> ## Content Index
> Fetch the complete content index at: https://genaisecretsauce.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# GenAI Secret Sauce Daily Digest - 2026-10-13
- URL: https://genaisecretsauce.com/genai-secret-sauce-daily-digest-2026-10-13/
- Published: 2026-10-13T23:39:20.000Z
- Updated: 2026-10-14T00:11:31.000Z
- Description: Google's Gemini 4 Argon is now open to developers and paying subscribers · Anthropic threw out Git because its AI writes too much code · Anthropic says small AI models may soon be able to copy themselves onto everyday gadgets
- Author: Jasmine Robinson
- Tags: Daily Digest

Watch today's digest as a video summary (generated by NotebookLM)

By the Numbers

## Statistically Speaking

[$2](https://deepmind.google/blog/introducing-gemini-4-argon?ref=genaisecretsauce.com) [per million input tokens and $10 per](https://deepmind.google/blog/introducing-gemini-4-argon?ref=genaisecretsauce.com) 

Google's Gemini 4 Argon is now open to developers and paying

Top Story

[1](https://deepmind.google/blog/introducing-gemini-4-argon?ref=genaisecretsauce.com) [on the Vals Index at 68](https://deepmind.google/blog/introducing-gemini-4-argon?ref=genaisecretsauce.com) 

Google's Gemini 4 Argon is now open to developers and paying

[12](https://deepmind.google/blog/introducing-gemini-4-argon?ref=genaisecretsauce.com) [simulated cyberattack scenarios in Google's safety tests](https://deepmind.google/blog/introducing-gemini-4-argon?ref=genaisecretsauce.com) 

Google's Gemini 4 Argon is now open to developers and paying

[10,000](https://deepmind.google/blog/introducing-gemini-4-argon?ref=genaisecretsauce.com) [scientists can get six months of free](https://deepmind.google/blog/introducing-gemini-4-argon?ref=genaisecretsauce.com) 

Google's Gemini 4 Argon is now open to developers and paying

[2,500](https://www.anthropic.com/engineering/redesigning-source-control?ref=genaisecretsauce.com) [merges a day in May 2026 grew](https://www.anthropic.com/engineering/redesigning-source-control?ref=genaisecretsauce.com) 

Anthropic threw out Git because its AI writes too much code

[600](https://www.anthropic.com/engineering/redesigning-source-control?ref=genaisecretsauce.com) [petabytes of daily traffic if Git fetched](https://www.anthropic.com/engineering/redesigning-source-control?ref=genaisecretsauce.com) 

Anthropic threw out Git because its AI writes too much code

One Thing to Tell Your Friends

## One Thing to Tell Your Friends

Anthropic's engineers, working with Claude, now merge more than 30,000 code changes a day - so many that Git, the tool almost every programmer on Earth uses, could not keep up, and a team of seven built a replacement in five months.

Summary

## TL;DR

Top Stories

[Google's Gemini 4 Argon is now open to developers and paying subscribers](https://deepmind.google/blog/introducing-gemini-4-argon?ref=genaisecretsauce.com), [Anthropic threw out Git because its AI writes too much code](https://www.anthropic.com/engineering/redesigning-source-control?ref=genaisecretsauce.com), and [Anthropic says small AI models may soon be able to copy themselves onto everyday gadgets](https://www.anthropic.com/research/self-replicating-ai?ref=genaisecretsauce.com).

Trends

[Governments are writing emergency plans for AI agents](https://www.gov.uk/government/news/britain-steps-up-national-defences-against-growing-ai-risks?ref=genaisecretsauce.com), [The fight is now over who gets to check the AI companies](https://thenextweb.com/news/anthropic-openai-ai-evaluators-publication-control?ref=genaisecretsauce.com), and [AI doom warnings are getting pushback from inside the field](https://the-decoder.com/machine-learning-pioneer-michael-i-jordan-dismantles-ai-doom-rhetoric-as-grand-moralizing-performance?ref=genaisecretsauce.com).

Creative AI

[Google Play now runs apps you made just for yourself](https://techcrunch.com/2026/10/13/google-play-now-supports-private-vibe-coded-apps-built-just-for-you?ref=genaisecretsauce.com) and [A game developer's recipe for getting AI to build games well](https://kylekukshtel.com/ai-harness-engineering-game-development-claude-code?ref=genaisecretsauce.com).

Dev Tools

[Linus Torvalds used Claude Code to make many small Linux file reads about 3 times faster](https://hwbusters.com/news/linus-torvalds-uses-claude-code-to-rework-the-linux-page-cache-tripling-small-read-throughput?ref=genaisecretsauce.com), [A simple trick to make coding agents follow your team's rules](https://simonwillison.net/2026/Oct/13/geoffrey-huntley?ref=genaisecretsauce.com), and [Apple turned idle employee Macs into a shared AI computer](https://machinelearning.apple.com/research/distributed-gpu-cloud?ref=genaisecretsauce.com).

Research

[A free AI model built to run agents privately on your own computer](https://huggingface.co/blog/LiquidAI/lfm25-24b-a2b?ref=genaisecretsauce.com), [A small Chinese AI model aimed at legal work topped Hacker News](https://z.ai/blog/glm-legal?ref=genaisecretsauce.com), and [Robots trained on human video are folding thousands of napkins a day at Din Tai Fung](https://www.latent.space/p/dyna?ref=genaisecretsauce.com).

Business

[NTT may build Anthropic's first data center in Japan for about $6.3 billion](https://asia.nikkei.com/business/technology/artificial-intelligence/ntt-to-build-first-data-center-for-anthropic-in-japan?ref=genaisecretsauce.com), [Twice as many companies use OpenAI's agent tools as Anthropic's](https://venturebeat.com/orchestration/openais-ai-agent-platform-is-used-by-twice-as-many-companies-as-anthropics-new-venturebeat-survey-finds?ref=genaisecretsauce.com), and [Elon Musk's company bundled X, Grok and the Cursor coding tool into one plan](https://techcrunch.com/2026/10/13/spacexsi-bundles-x-premium-grok-and-cursor-into-one-subscription?ref=genaisecretsauce.com).

Education

[An Oxford](https://aiedusimplified.substack.com/p/giving-everybody-a-seat-at-the-table), [College admissions' secret AI checks hurt students with the least help](https://www.insidehighered.com/opinion/views/2026/10/12/ai-literacy-and-new-admissions-disadvantage-opinion?ref=genaisecretsauce.com), and [African researchers say they should lead AI science, not just apply it](https://www.timeshighereducation.com/news/africa-has-crucial-role-advancing-study-ai-say-professors?ref=genaisecretsauce.com).

Surprising

[OpenAI's always](https://www.theregister.com/ai-and-ml/2026/10/13/openais-dot-assistant-burned-300k-tokens-in-a-day-without-a-single-prompt/5302913?ref=genaisecretsauce.com), [Police can now find a house from one photo of its laundry room](https://404media.co/ai-company-lets-cops-pinpoint-specific-houses-based-on-a-single-interior-photo?ref=genaisecretsauce.com), and [UK asylum AI summaries failed one in five expert checks](https://www.theregister.com/public-sector/2026/10/13/home-office-asylum-ai-flunks-one-in-five-accuracy-checks/5302649?ref=genaisecretsauce.com).

Worth Watching

[Your card may soon ask for a "permission slip" before an AI shops for you](https://techcrunch.com/2026/10/13/amex-introduces-a-framework-to-prevent-ai-agents-from-making-unauthorized-purchases?ref=genaisecretsauce.com), [Most AI security risks now come from apps built by non](https://techcrunch.com/2026/10/13/pluto-security-raises-42m-as-ai-gives-enterprise-security-teams-a-new-headache?ref=genaisecretsauce.com), and [AI agents are starting to negotiate with other companies' AI agents](https://fortune.com/2026/10/13/startup-causa-prima-20-million-billing-disputes-ai-agents?ref=genaisecretsauce.com).

GitHub

Leading repos: [morluto/rea](https://github.com/morluto/rea?ref=genaisecretsauce.com) (+54,706), [bilawalsidhu/gods-eye](https://github.com/bilawalsidhu/gods-eye-view?ref=genaisecretsauce.com) (+1,961), and [debpalash/VoiceStudio](https://github.com/debpalash/VoiceStudio?ref=genaisecretsauce.com) (+2,277).

HuggingFace

Leading models: [google/embeddinggemma](https://huggingface.co/google/embeddinggemma-2?ref=genaisecretsauce.com) (124,930), [Qwen/Qwen-Image-2.1](https://huggingface.co/Qwen/Qwen-Image-2.1-Turbo?ref=genaisecretsauce.com) (9,665), and [Lightricks/LTX](https://huggingface.co/Lightricks/LTX-2.5?ref=genaisecretsauce.com) (1,724,548).

Product Hunt

Top launches: [Articos MCP](https://www.producthunt.com/products/articos?ref=genaisecretsauce.com) (298), [OpenCode 2.0](https://www.producthunt.com/products/opencode?ref=genaisecretsauce.com) (236), and [Kai](https://www.producthunt.com/products/hirekai?ref=genaisecretsauce.com) (171).

API Pricing

No list-price changes vs. October 12 - but Gemini 4 Argon now appears on Google's official pricing page, with a firm date for its price to double (December 1) and an official 1M-token window.

arXiv

[KV-Repurpose: A Testbed for Evaluating Non-Prefix KV](https://arxiv.org/abs/2610.13060?ref=genaisecretsauce.com) — With the same 30 GB cache, segment reuse hit 63.9% of tokens versus 3.3% for standard prefix caching, but answers combining several documents got noticeably worse (2WikiMultihopQA 0.80 to 0.50).

FYI

## Hot off the Presses

01

### Google's Gemini 4 Argon is now open to developers and paying subscribers

**What this means for you:** A top-tier AI model is now available at a mid-range price, and if you pay for Google AI Pro or Ultra it is rolling out to you now in the Gemini app and in AI Mode in Search.

*Previously: [September 30](https://genaisecretsauce.com/genai-secret-sauce-daily-digest-2026-09-30/) \- Google unveiled Gemini 4 Argon with chart-topping scores, but only cybersecurity defenders could use it.*

**Today:** Google opened Argon to developers on its paid API (Application Programming Interface, the way software talks to the model) and to Google AI Pro ($19.99 a month in the US) and Ultra subscribers. The Gemini app also gets a "tasks" mode that keeps working on a job after you leave your computer, using 40+ connected apps such as Instacart and OpenTable.

Independent scores put Argon among the leaders, not alone at the top. Artificial Analysis rates it 53 on its Intelligence Index, tied with OpenAI's GPT-6 Astra and Anthropic's Fable 5.1, behind Anthropic's Opus 5.5 at 58.

- **$2 per million input tokens and $10 per million output tokens until November 30** \- tokens are word pieces, and the price doubles to $4 and $20 on December 1
- **Ranked #1 on the Vals Index at 68.9%** \- a test of finance, coding, legal and tax work
- **It solved all 12 simulated cyberattack scenarios in Google's safety tests** \- the previous generation solved five, so Google switched on its next-generation safeguards
- **10,000 scientists can get six months of free top-tier access** \- through Google's Antigravity coding platform

$2

per million input tokens and

10,000

scientists can get six months

[Google DeepMind: Gemini 4 Argon announcement →](https://deepmind.google/blog/introducing-gemini-4-argon?ref=genaisecretsauce.com)[VentureBeat: Argon broadly available at $2/$10 →](https://venturebeat.com/technology/google-makes-gemini-4-argon-broadly-available-with-limited-time-low-api-pricing-of-2-10-per-1m-token?ref=genaisecretsauce.com)[Try it →](https://gemini.google.com/?ref=genaisecretsauce.com)

02

### Anthropic threw out Git because its AI writes too much code

**What this means for you:** AI is now producing software so fast that the basic tools programmers have used for 20 years are breaking, which hints at how quickly the apps you use may change.

Git is the free system nearly all programmers use to track changes to code. Anthropic says its main code store grew so busy that Git could not cope, so it built a replacement called Scion and moved off Git completely. Its automated test systems check out the code about 25 million times a day.

The lab says AI agents, not people, now drive most of the activity. Engineers' main job has shifted to judging designs and telling Claude "no" when it breaks the rules of the system.

- **About 2,500 merges a day in May 2026 grew to over 30,000 by September** \- merges are approved code changes, and the peak was 30,401 on September 9
- **An estimated 600 petabytes of daily traffic if Git fetched naively** \- a petabyte is a million gigabytes
- **Seven people built Scion and a new code review tool** \- and finished the switch in five months
- **200 times less computing and memory** \- than the old setup of Git copies it replaced

[Anthropic: Redesigning source control at Anthropic →](https://www.anthropic.com/engineering/redesigning-source-control?ref=genaisecretsauce.com)[Simon Willison: notes on Anthropic's Scion →](https://simonwillison.net/2026/Oct/13/redesigning-source-control-at-anthropic?ref=genaisecretsauce.com)

03

### Anthropic says small AI models may soon be able to copy themselves onto everyday gadgets

**What this means for you:** The cheap smart devices in your home could one day host AI programs that spread on their own, which makes basic device security, like changing default passwords, matter more.

Anthropic's Frontier Red Team, the group that tests for dangerous abilities, asked whether small, freely downloadable AI models could spread across cheap internet-connected devices. Its answer: today's small models can do many individual steps but fail to chain them together. The team estimates they could close that gap by the end of 2026, while calling its timeline "only a rough estimate."

The lab also admits it made the tests easy. Target machines had no passwords, and the directing model ran on a separate, more powerful server. Outside researchers pushed back hard (see Trends & Themes).

- **One model reached all 8 machines in 4 of 5 trials** \- Z.ai's GLM-5.2, in a simplified test network
- **A Gemma 4 model ran across 15 Android TV boxes costing about $55 each** \- at a crawl of 0.39 tokens per second
- **Spread would be slow** \- roughly doubling every few days, but hard to stamp out with no central off switch
- **Suggested defenses are familiar** \- spotting odd network traffic, isolating infected devices, and banning default device passwords

[Anthropic: AI models may be capable of self-replicating without a datacenter →](https://www.anthropic.com/research/self-replicating-ai?ref=genaisecretsauce.com)[Fortune: Anthropic's small-model warning and its critics →](https://fortune.com/2026/10/13/anthropic-small-ai-models-copy-themselves-safety-risk?ref=genaisecretsauce.com)

04

### An OpenAI engineer was arrested for allegedly bugging an executive meeting room

**What this means for you:** AI labs now guard their research like state secrets, and prosecutors say chatbot questions from an account tied to the suspect helped give him away.

Federal prosecutors charged Abdulrahman Alfozan, 35, who had worked at OpenAI since July 2025, with using a device to intercept private conversations. OpenAI staff found a recorder on September 30, plugged in behind a meeting-room TV at its San Francisco headquarters and wrapped in toilet paper. The room hosted confidential meetings on research, products and government work.

Prosecutors allege a chatbot account linked to his spouse, sharing identifiers with his own, asked from September 24 to 30 how to build, hide and remotely access a recording device. Security later found a second, similar device in a bag he left unattended. US Attorney Craig Missakian called the case "a matter of national security." He has been charged, not convicted.

“a matter of national security”

- **Up to five years in prison and a $250,000 fine** \- if convicted
- **OpenAI says it used its own AI models to identify the device** \- and found no sign he accessed model files, source code or customer data
- **A detention hearing is set for October 19** \- he remains in custody, and OpenAI fired him the day of his arrest

[SF Standard: OpenAI engineer arrested in bugging of executive conference room →](https://sfstandard.com/2026/10/13/openai-engineer-arrested-alleged-bugging-executive-conference-room?ref=genaisecretsauce.com)

05

### Character.ai filed for bankruptcy and may shut down

**What this means for you:** If you or your kids use Character.ai's AI companions, the service could disappear by early December, so export any chats you want to keep.

Character.ai lets people chat with AI characters, and it became one of the best-known AI companion apps. On October 12 it started a voluntary Chapter 11 bankruptcy to sell all or most of its assets under court supervision. The company blamed a year of "significant legal, regulatory, and operational headwinds," without giving details.

The company says it hopes a buyer keeps the app running. If no workable bid comes in, the platform will be shut down.

- **The sale is expected to close by late November or early December 2026**
- **Everything still works for now** \- paid subscriptions stay active and billing is unchanged
- **You can delete your account or export your data**
- **Disney agreed to pay $20 million for some Character.ai assets** \- Reuters reports it is an opening bid that rivals can top, and a court must approve it

[Character.ai: An update on Character.ai →](https://blog.character.ai/an-update-on-character-ai?ref=genaisecretsauce.com)[NBC News: Character.AI to sell some assets to Disney for $20 million →](https://www.nbcnews.com/tech/tech-news/characterai-files-bankruptcy-sell-assets-disney-20-million-rcna603174?ref=genaisecretsauce.com)

Trends & Themes

## Trends & Themes

[![Trends & Themes](https://genaisecretsauce.com/content/images/2026/10/section-what-this-means-2026-10-13.png)](https://genaisecretsauce.com/content/images/2026/10/section-what-this-means-2026-10-13.png) 

### Governments are writing emergency plans for AI agents

**Why this matters to you:** The banks, hospitals and utilities you depend on are being told to prepare for AI agents causing a crisis, not just hackers.

The shift is from "will AI agents misbehave?" to "what do we do the day they do?" Regulators are treating agent accidents like fires or floods that need a drill.

- **Britain's frontier-AI cyber pilots have already helped find and fix more than 1,000 potential flaws in government systems** \- and UK security agencies are now working out how to respond to major incidents involving AI agents
- **UK ministers may strengthen a pending cyber security bill to cover AI "loss of control"** \- where autonomous systems evade human oversight
- **Three top US banking regulators are questioning banks about AI** \- including whether they use China-developed technology and can control autonomous systems, per Reuters
- **406 scientists signed a letter demanding AI models be fully cut off from the internet** \- including Geoffrey Hinton, Yoshua Bengio and Stuart Russell
- **Google says Argon's safety monitors can stop an AI agent mid-response** \- as part of the safeguards it switched on for its new model

[GOV.UK: Britain steps up national defences against growing AI risks →](https://www.gov.uk/government/news/britain-steps-up-national-defences-against-growing-ai-risks?ref=genaisecretsauce.com)[PYMNTS: Bank regulators tighten oversight of AI agents →](https://www.pymnts.com/news/banking/2026/banking-regulators-tighten-oversight-of-possible-ai-risks?ref=genaisecretsauce.com)[The Decoder: 406 scientists demand AI be cut off from the internet →](https://the-decoder.com/406-scientists-demand-ai-models-be-cut-off-from-the-internet-immediately?ref=genaisecretsauce.com)[Google DeepMind: Developing Gemini 4 Argon responsibly →](https://deepmind.google/blog/developing-gemini-4-argon-responsibly-and-safely?ref=genaisecretsauce.com)

### The fight is now over who gets to check the AI companies

**Why this matters to you:** Whether outside testers can speak freely decides whether you ever hear about an AI model's problems before it reaches your phone.

Companies want outside checks they can shape. Critics want checks with legal teeth.

- **Anthropic and OpenAI pushed for control over what outside evaluators publish** \- Reuters reports, citing fears of leaked trade secrets, weeks after both pledged reviewers could publish freely
- **OpenAI says it only strips secret details** \- and Anthropic says it keeps sign-off only on reports from new testing firms
- **"Strict liability" is gaining support** \- making AI companies pay for harm even without proven negligence, an idea the New York Times editorial board endorsed on October 10
- **Every group in a Forecasting Research Institute study rated it above a US licensing system** \- including 157 experts and 49 "superforecasters," people with strong prediction track records
- **A Fortune op-ed by an ex-Google DeepMind researcher and two whistleblower advocates says self-policing has already failed** \- and calls for mandatory independent audits and whistleblower protections

[The Next Web: Anthropic and OpenAI want a veto on evaluator reports →](https://thenextweb.com/news/anthropic-openai-ai-evaluators-publication-control?ref=genaisecretsauce.com)[Fortune: Strict liability is gaining traction →](https://fortune.com/2026/10/13/strict-liability-is-gaining-traction-as-a-solution-to-force-ai-companies-to-institute-better-safety-controls-but-it-has-some-big-downsides-too?ref=genaisecretsauce.com)[Fortune: The White House wants AI companies to police themselves →](https://fortune.com/2026/10/12/openai-whistleblowers-ai-safety-keep-firing-sam-altman?ref=genaisecretsauce.com)

### AI doom warnings are getting pushback from inside the field

**Why this matters to you:** Scary AI headlines are now openly disputed by respected experts, so it pays to ask who is making each claim and what they stand to gain.

The debate is no longer safety advocates versus everyone else. Builders, investors and academics are splitting into camps, and each has money or reputation on the line.

“Fear isn't evidence.”

- **Machine learning pioneer Michael I. Jordan called existential AI warnings science fiction** \- and labeled some lab researchers' public resignations "a grand moralizing performance"
- **Cerebras CEO Andrew Feldman said doom talk is hurting the industry** \- joking that some AI leaders "maybe shouldn't be allowed to talk in public," while his chip company sells to OpenAI
- **Rocky Linux founder Gregory Kurtzer dismissed Anthropic's self-copying study** \- pointing out the targets were left unprotected on purpose
- **OpenAI researcher Noam Brown says cutting AI off from the internet only buys time** \- and that over 10% of his own team now works on making AI behave as intended

[The Decoder: Michael I. Jordan on AI doom rhetoric →](https://the-decoder.com/machine-learning-pioneer-michael-i-jordan-dismantles-ai-doom-rhetoric-as-grand-moralizing-performance?ref=genaisecretsauce.com)[TechCrunch: Cerebras CEO pushes back on AI doomsday warnings →](https://techcrunch.com/2026/10/13/cerebras-ceo-pushes-back-on-ai-doomsday-warnings?ref=genaisecretsauce.com)[Fortune: Critics of Anthropic's self-copying study →](https://fortune.com/2026/10/13/anthropic-small-ai-models-copy-themselves-safety-risk?ref=genaisecretsauce.com)

### Investors are asking harder questions about AI's money

**Why this matters to you:** If AI spending outruns AI profits, the fallout could hit retirement accounts and jobs far beyond the tech industry.

The worry is not that AI fails. It is that cheap rival models squeeze prices before the giant data centers pay for themselves.

- **The International Monetary Fund (IMF) warned an AI correction could hit global markets** \- if heavy investment fails to produce the profits investors expect
- **The IMF flagged "circular financing"** \- where AI companies act as each other's investors, lenders and customers, so trouble at one spreads fast
- **Chinese models processed 76.12 trillion tokens in one week on OpenRouter** \- a service that routes requests to many AI models, versus 14.2 trillion for the top four US providers, per a Jefferies note
- **Jefferies estimates about half of a $2.3 trillion cloud backlog belongs to OpenAI and Anthropic** \- so price pressure on those two labs could ripple through big tech
- **Anthropic will show investors its third-quarter numbers before a reported November stock market debut** \- as lab finances become central to how tech stocks trade

[The Register: IMF warns an AI bubble could burst →](https://www.theregister.com/ai-and-ml/2026/10/13/ai-bubble-could-burst-and-take-global-economy-with-it-imf-warns/5302781?ref=genaisecretsauce.com)[The Register: Cheap Chinese models could trigger an AI infrastructure reset →](https://www.theregister.com/paas-and-iaas/2026/10/13/cheap-chinese-and-open-source-llms-could-trigger-trillion-dollar-ai-infrastructure-reset/5302774?ref=genaisecretsauce.com)[CNBC: Anthropic to disclose new revenue numbers →](https://cnbc.com/2026/10/13/anthropic-q3-financials-ipo.html?ref=genaisecretsauce.com)

### People are using AI more, and liking it less

**Why this matters to you:** If you feel more tired of AI even as you use it more, you are in good company, and employers are starting to notice.

The novelty phase is ending. What comes next is a quieter argument about where AI genuinely helps and where people would rather do the work themselves.

“The amazement is kind of wearing off”

- **Daily AI use among US managers jumped from 46% to 62%** \- but the share who feel "excited" fell 8 points to 51%, in a Wharton and GBK Collective survey
- **72% of those managers say their job shifted from making things to checking AI's work**
- **51% of US adults never use AI chatbots** \- and 67% of them say they are unlikely to start within a year, per Pew Research Center
- **IMF chief Kristalina Georgieva called AI's effect on jobs a "slow-moving tsunami"** \- warning that the entry-level jobs that start careers are disappearing
- **Hachette Book Group's CEO writes every email himself** \- and says leaders outsourcing their writing to AI will create a "worse C-suite"

[Fortune: Wharton survey on managers and AI agents →](https://fortune.com/2026/10/13/gen-z-ai-backlash-managers-agent-verification-wharton-gbk?ref=genaisecretsauce.com)[The Register: Chatbot refuseniks edge out users →](https://www.theregister.com/ai-and-ml/2026/10/12/chatbot-refuseniks-edge-out-users-in-the-us-and-theyre-not-for-turning/5302690?ref=genaisecretsauce.com)[Fortune: IMF chief warns of a 'slow-moving tsunami' →](https://fortune.com/2026/10/13/imf-chief-kristalina-georgieva-ai-job-market-warning-tsunami-of-disruption-hurt-gen-z-workers?ref=genaisecretsauce.com)[Fortune: Hachette's CEO writes every email himself →](https://fortune.com/2026/10/13/hachettes-ceo-writes-every-email-himself-he-says-ai-shortcuts-are-creating-a-worse-c-suite?ref=genaisecretsauce.com)

Creative AI & Media

## Creative AI & Media

### Google Play now runs apps you made just for yourself

- **Describe an app in Google AI Studio and install it on your own Android phone** \- no public store listing needed
- **Only you can see and use it** \- launching first for US users 18 and over
- **Sharing with family or friends is coming "in the months ahead"** \- for example, an app for a child's sports league
- **Popular personal apps can later go public** \- after Google's trust-and-safety review

[TechCrunch: Google Play now supports private vibe-coded apps →](https://techcrunch.com/2026/10/13/google-play-now-supports-private-vibe-coded-apps-built-just-for-you?ref=genaisecretsauce.com)[Try it →](https://aistudio.google.com/?ref=genaisecretsauce.com)

### A game developer's recipe for getting AI to build games well

- **AI handles websites easily but struggles with games** \- because games are full of art, levels and data that are not code
- **Make every random event repeatable** \- route all randomness through one seeded source so any bug can be replayed exactly
- **Let the AI "see" the game** \- take screenshots and compare them with a free image-diff tool that gives a similarity score
- **Build save and load first** \- so the AI can jump straight to any game state to test a fix

[Kyle Kukshtel: Your Game Is A Harness →](https://kylekukshtel.com/ai-harness-engineering-game-development-claude-code?ref=genaisecretsauce.com)

Developer Tools

## Developer Tools & Infrastructure

### Linus Torvalds used Claude Code to make many small Linux file reads about 3 times faster

**Linus Torvalds**, creator of the Linux operating system, credited Anthropic's Claude Code for writing an experimental set of Linux changes.

- **It speeds up many small reads of the same file** \- from about 10 million to 30-33 million reads per second with 8 threads (parallel workers) on his laptop
- **Claude Code wrote the code, the tests and the first draft of the cover letter** \- Torvalds directed the work and ran the benchmarks himself
- **The catch is memory** \- freed memory is held longer, which Torvalds calls "the part of this series that is least solved"
- **Not merged yet** \- it was tested on one laptop and needs checks on large servers

[Hardware Busters: Linus Torvalds uses Claude Code to rework the Linux page cache →](https://hwbusters.com/news/linus-torvalds-uses-claude-code-to-rework-the-linux-page-cache-tripling-small-read-throughput?ref=genaisecretsauce.com)

### A simple trick to make coding agents follow your team's rules

**Geoffrey Huntley**, a developer known for agent experiments, stopped relying on written instructions and enforced a rule at save time instead.

- **The problem** \- coding agents kept skipping his team's required logging library
- **The fix** \- a "pre-commit hook," a check that runs before code is saved to the project, rejects the change and tells the agent what to do instead
- **The result** \- he reports compliance went to ten out of ten

[Simon Willison: Quoting Geoffrey Huntley →](https://simonwillison.net/2026/Oct/13/geoffrey-huntley?ref=genaisecretsauce.com)

### Apple turned idle employee Macs into a shared AI computer

**Minions** is an internal Apple research system that pools the unused graphics power of employees' Macs to run AI jobs.

- **Queue times for AI jobs fell from hours to minutes** \- Apple reports, with no new graphics hardware requested
- **Work stays on the Macs** \- using open models, with no outside AI services
- **Machines that sleep or go offline get their tasks handed to others** \- by a watchdog program

[Apple Machine Learning Research: Minions →](https://machinelearning.apple.com/research/distributed-gpu-cloud?ref=genaisecretsauce.com)

### Plan for the day your AI provider goes down

**Sam Newman's** new book on resilient systems, excerpted in The Pragmatic Engineer, has a section on AI features.

- **OpenAI and Anthropic have weaker uptime records than the big cloud companies** \- so AI features need a fallback
- **Use a backup route** \- for example, reaching OpenAI models through Microsoft Azure
- **Test the AI's behavior, not just whether it responds** \- model updates can silently change answers

[The Pragmatic Engineer: Building resilient AI-powered software →](https://newsletter.pragmaticengineer.com/p/building-resilient-systems?ref=genaisecretsauce.com)

Research & Models

## Research & Models

### A free AI model built to run agents privately on your own computer

- **Liquid AI's LFM2.5-24B-A2B is open to download** \- it has 24 billion parameters (internal settings) but uses only about 2 billion per word, so it runs fast
- **It beat larger rivals on agent tasks** \- 27.69 on AutomationBench (a test of automating multistep work), more than triple the next model Liquid compared
- **It is weaker on knowledge and coding** \- Liquid suggests pairing it with web search
- **About 184 tokens per second on an Apple M5 Max laptop chip**

[Hugging Face: Liquid AI's LFM2.5-24B-A2B →](https://huggingface.co/blog/LiquidAI/lfm25-24b-a2b?ref=genaisecretsauce.com)

### A small Chinese AI model aimed at legal work topped Hacker News

- **Z.ai's GLM-Legal is built on its lightweight GLM-5.3-Flash model** \- then trained across tens of thousands of realistic legal tasks
- **19.2% on Harvey's Legal Agent Benchmark** \- which requires every check to pass on real law-firm work, 12.5 points above the base model
- **Z.ai claims it matches top closed models on LegalBench** \- with those results checked by the outside firm Vals.ai
- **In beta only** \- no size, license or price disclosed

[Z.ai: GLM-Legal →](https://z.ai/blog/glm-legal?ref=genaisecretsauce.com)

### Robots trained on human video are folding thousands of napkins a day at Din Tai Fung

- **Dyna Robotics' Dyna-2 learned from more than one million hours of human video** \- it predicts what happens next, then acts
- **A new task can take as little as 45 minutes of demonstrations**
- **Its robots fold thousands of napkins a day at several US Din Tai Fung restaurants** \- and towels for Best Western
- **The goal for next year is hundreds of robots** \- up from tens today

[Latent.Space: Dyna-2 with Jason Ma →](https://www.latent.space/p/dyna?ref=genaisecretsauce.com)

### Anthropic says it cut AI "copying" by rivals by 90-99%

- **"Distillation" means training your own model on another company's AI answers** \- often through networks of fake accounts
- **Anthropic says new defenses cut the useful data copiers collect by about 90-99%**
- **It named Chinese lab Zhipu** \- saying it distilled cyber skills from US models while building GLM-5.3
- **The trade-off** \- false alarms in under 0.1% of cases, and charges even for blocked requests

[Anthropic: Distillation defenses →](https://www.anthropic.com/research/distillation-defenses?ref=genaisecretsauce.com)

Business & Industry

## Business & Industry

### NTT may build Anthropic's first data center in Japan for about $6.3 billion

- **Planned for Osaka, with a partial opening targeted for 2028**
- **Talks are in final stages but not signed** \- per Nikkei's sources
- **It would give Anthropic dedicated computing inside Japan** \- which could appeal to Japanese customers who want their data kept in the country

[Nikkei Asia: NTT to build first data center for Anthropic in Japan →](https://asia.nikkei.com/business/technology/artificial-intelligence/ntt-to-build-first-data-center-for-anthropic-in-japan?ref=genaisecretsauce.com)

### Twice as many companies use OpenAI's agent tools as Anthropic's

- **43% of 107 companies use OpenAI's agent platform** \- versus 22% for Anthropic and 21% for Google, in VentureBeat's September survey
- **Momentum favors Anthropic** \- 33% of companies shopping for a new platform are considering it, versus 30% for OpenAI, though from a small group
- **62% swapped an AI model in a live product in the past year** \- and 59% of those saw performance drop

[VentureBeat: OpenAI's agent platform used by twice as many companies →](https://venturebeat.com/orchestration/openais-ai-agent-platform-is-used-by-twice-as-many-companies-as-anthropics-new-venturebeat-survey-finds?ref=genaisecretsauce.com)

### Elon Musk's company bundled X, Grok and the Cursor coding tool into one plan

- **"The X Subscription" has five tiers** \- free, then $8, $30, $100 and $200 a month
- **It combines X Premium's checkmark, the Grok assistant and Cursor** \- an AI coding tool, in one price ladder
- **The newest Grok models need the $30 tier or higher**

[TechCrunch: SpaceXSI bundles X Premium, Grok and Cursor →](https://techcrunch.com/2026/10/13/spacexsi-bundles-x-premium-grok-and-cursor-into-one-subscription?ref=genaisecretsauce.com)

### A judge ordered OpenAI to hand over hiring records in Apple's trade secrets case

- **Apple sued in July** \- alleging former employees took confidential files to OpenAI
- **OpenAI must share interview materials from hiring ex-Apple staff** \- for its hardware and robotics teams, within 5 to 14 days
- **Three OpenAI employees, including its chief hardware officer, will be questioned under oath**
- **The judge refused, for now, Apple's broader requests about OpenAI's hardware research**

[TechCrunch: Judge orders OpenAI to turn over hiring records →](https://techcrunch.com/2026/10/13/judge-orders-openai-to-turn-over-hiring-records-in-apple-trade-secrets-case?ref=genaisecretsauce.com)

### Customer-service AI startup Decagon is exploring a sale

- **It has held early talks with ServiceNow and others** \- and hired an investment bank, Business Insider reports
- **Last valued at $4.5 billion** \- with customers including Delta and Ticketmaster
- **It could raise a new funding round instead if no deal happens** \- the talks are early, per people familiar with the matter

[Business Insider: Decagon hires investment bank amid acquisition interest →](https://www.businessinsider.com/ai-agent-company-decagon-hired-investment-bank-acquisition-interest-2026-10?ref=genaisecretsauce.com)

Education

## GenAI in Education

### An Oxford-led research team wants university staff and students to say what human teachers are for

- **The AIEOU project's higher-education group spans five continents** \- and asks what humans uniquely bring to teaching in an AI age
- **Its survey uses imagined future scenarios** \- so people reflect before stating their values, taking about 15-20 minutes
- **Responses are requested by October 31, 2026** \- from anyone in a university, including faculty, students and staff
- **The team's own rule: humans must write everything** \- AI can help analyze data only between human start and review

[AI + Education = Simplified: Giving everybody a seat at the table →](https://aiedusimplified.substack.com/p/giving-everybody-a-seat-at-the-table)

### College admissions' secret AI checks hurt students with the least help

- **The Common App treats AI-written content as fraud** \- but colleges rarely say how they detect it
- **Policies conflict** \- some schools encourage AI use while others penalize it
- **Private-school teachers have far more access to AI tools** \- per a College Board study of over 10,000 people
- **The author asks colleges to disclose their detection tools** \- and how often they reject applications for AI use

[Inside Higher Ed: AI literacy and the new admissions disadvantage →](https://www.insidehighered.com/opinion/views/2026/10/12/ai-literacy-and-new-admissions-disadvantage-opinion?ref=genaisecretsauce.com)

### African researchers say they should lead AI science, not just apply it

- **Wits University professors call the "junior partner" view of Africa "condescending"**
- **The Deep Learning Indaba grew from about 330 participants to 1,300** \- and is now described as the world's largest AI summer school
- **48 of Africa's 54 countries host a local IndabaX event this year**

[Times Higher Education: Africa has crucial role in advancing study of AI →](https://www.timeshighereducation.com/news/africa-has-crucial-role-advancing-study-ai-say-professors?ref=genaisecretsauce.com)

### Ireland's tech sector aims to train 400,000 people in AI by 2030

- **That is close to one in seven Irish workers**
- **Training is matched to each learner's starting level and goals** \- through a nonprofit's AI training center
- **IBM is providing its free SkillsBuild courses** \- no budget was disclosed

[Silicon Republic: FIT plans to upskill 400,000 people in AI →](https://siliconrepublic.com/careers/fit-unveils-plan-upskill-people-ai-2030-working-life-skills?ref=genaisecretsauce.com)

Surprising

## Surprising & Under-the-Radar

### OpenAI's always-on assistant used about 307,000 tokens in a day with no prompts

Security firm Geordie gave OpenAI's "dot" assistant read-only access and no email or apps, then left it alone. It still logged 155 activities, including at least 50 inside two minutes. Surprising because "always-on" agents can quietly run up usage you never asked for.

[The Register: OpenAI's Dot burned 300K+ tokens in a day →](https://www.theregister.com/ai-and-ml/2026/10/13/openais-dot-assistant-burned-300k-tokens-in-a-day-without-a-single-prompt/5302913?ref=genaisecretsauce.com)

### Police can now find a house from one photo of its laundry room

Graylark's tool Raven matches indoor photos to listing pictures on Airbnb, Vrbo and Zillow, with no location data needed. Its CEO says the ability was found during testing, not built on purpose. Airbnb and Zillow say scraping their sites breaks their rules.

[404 Media: AI lets cops pinpoint houses from a single interior photo →](https://404media.co/ai-company-lets-cops-pinpoint-specific-houses-based-on-a-single-interior-photo?ref=genaisecretsauce.com)

### UK asylum AI summaries failed one in five expert checks

The Home Office uses OpenAI's GPT-4o to summarize asylum interviews. After rollout, 45 of 203 summaries failed initial checks, up from a 9% rejection rate in the pilot. Applicants are not told AI was used and cannot correct the summaries.

[The Register: Home Office asylum AI flunks one in five accuracy checks →](https://www.theregister.com/public-sector/2026/10/13/home-office-asylum-ai-flunks-one-in-five-accuracy-checks/5302649?ref=genaisecretsauce.com)

### A small website got buried in AI bot traffic after one popular post

After a Hacker News post, developer Sammy Straus found nearly half his traffic came from one AI newsletter service watching for website changes. AI crawlers and hacking scanners made up much of the rest. A simple robots file and rate limits fixed it.

[Sammy Straus: I posted to HN, then my website was flooded with AI bots →](https://www.sammystraus.com/?ref=genaisecretsauce.com#i-posted-to-hn-and-then-my-website-was-flooded-with-ai-bot-traffic)

### Debate: is AI making mathematicians dumber, or faster?

**Worried:** 25 Fields Medal winners, including Terence Tao and Peter Scholze, have voiced concern that AI proofs erode real understanding, and some are calling for a boycott of AI-generated proofs amid a flood of automated submissions. **Enthusiastic:** Fields Medalist Shing-Tung Yau just launched a Beijing alliance of eight institutions aiming for 1,000+ researchers doing AI math.

[The Decoder: Western mathematicians worry while China builds an AI math alliance →](https://the-decoder.com/western-mathematicians-fear-ai-makes-them-dumber-while-china-builds-an-entire-alliance-for-it?ref=genaisecretsauce.com)

### Debate: is AI getting smarter everywhere, or just spikier?

**Spikier:** Writer Nate Jones argues AI races ahead where feedback is cheap, like math, while lagging elsewhere, so one model could rebuild a 16,000-line program from its behavior yet fail a routine customization. **Broader:** He concedes models are improving across the board, and that the evidence cannot yet separate the two views.

[Nate's Newsletter: Narrow superintelligence →](https://natesnewsletter.substack.com/p/narrow-superintelligence-ai)

Worth Watching

## Signals to Track

[![Worth Watching](https://genaisecretsauce.com/content/images/2026/10/section-worth-watching-2026-10-13.png)](https://genaisecretsauce.com/content/images/2026/10/section-worth-watching-2026-10-13.png) 

01

### Your card may soon ask for a "permission slip" before an AI shops for you

American Express wants a signed record of what you allowed an agent to buy.

Amex's new Intent Intelligence Framework captures your instructions, like brand, size and budget, and turns them into a token the agent must use at checkout. If you ask for white Nike sneakers and the agent tries to buy Adidas, Amex can block it. If this becomes standard, disputes over AI purchases could get much easier to win.

[TechCrunch: Amex introduces a framework to stop unauthorized agent purchases →](https://techcrunch.com/2026/10/13/amex-introduces-a-framework-to-prevent-ai-agents-from-making-unauthorized-purchases?ref=genaisecretsauce.com)

02

### Most AI security risks now come from apps built by non-programmers

About 60% of the risks one startup finds are in tools "vibe-coded" by finance, marketing and legal staff.

Pluto Security, which just raised $42 million, watches the AI tools employees use and what they build with them. Developers ranked only fourth as a source of risk, and one example was an internet-exposed customer-data app built by a finance employee. If this holds, the biggest office security worry may soon be your coworker's homemade app.

[TechCrunch: Pluto Security raises $42M →](https://techcrunch.com/2026/10/13/pluto-security-raises-42m-as-ai-gives-enterprise-security-teams-a-new-headache?ref=genaisecretsauce.com)

03

### AI agents are starting to negotiate with other companies' AI agents

A new startup's agents settle billing disputes with suppliers, and it earns a cut when they negotiate early payments.

Causa Prima raised a $20 million seed round for agents that resolve billing disputes and speed up payments between businesses. Invoice processing averages more than eight days today. If it works, small suppliers could get paid faster without chasing anyone.

[Fortune: Causa Prima raises $20 million for billing-dispute agents →](https://fortune.com/2026/10/13/startup-causa-prima-20-million-billing-disputes-ai-agents?ref=genaisecretsauce.com)

04

### In China, specialized AI chips now outsell graphics cards

Chips designed for specific AI tasks, such as running trained models, passed graphics processors in China's data centers by units shipped.

IDC data seen by Reuters shows non-GPU accelerators jumped nearly 80% to 1.016 million units in the first half of 2026, overtaking Graphics Processing Units (GPUs). Nvidia's share fell to 34% as Huawei rose to 28%. If cheaper specialized chips keep winning, running AI could get cheaper worldwide, and less dependent on Nvidia.

[The Standard: Nvidia's China AI chip share falls as Huawei closes gap →](https://www.thestandard.com.hk/innovation/article/345296/Nvidias-China-AI-chip-share-falls-as-Huawei-closes-gap-report-finds?ref=genaisecretsauce.com)

GitHub Trending

## Top Repos Today

#1

### [morluto/rea](https://github.com/morluto/rea?ref=genaisecretsauce.com)

Rank yesterday: #1 - Holding steady ➡

⭐ **Stars today:** +54,706 · 📦 **Total:** 205,628  
📜 **License:** MIT · 👤 **By:** individual developer  
🎯 **Time to value:** 15 minutes

**What it is:** A tool that lets an AI coding agent take apart apps you do not have the source code for and explain how a feature works. It passed 200,000 stars in a record day, and now warns that copycat projects are offering fake downloads. **Why you'd want it:** Understand how an app does something, or debug a closed-source dependency, with the agent doing the legwork.

| ✓ Pros                                  | ✗ Cons                                                            |
| --------------------------------------- | ----------------------------------------------------------------- |
| One tool covering many analysis engines | Reverse engineering can break software license terms              |
| Runs locally and shows its evidence     | Copycat repositories exist, so install only from official sources |
| Free and open source                    | Hundreds of changes a day, so pin a version                       |

[GitHub - morluto/rea: Reverse engineer anything with agents, from app behavior down to native binaries.Reverse engineer anything with agents, from app behavior down to native binaries. - morluto/rea![](https://genaisecretsauce.com/content/images/icon/favicon-25f39efc-e761-4074-9f83-2cf5c770e04a.png)GitHub![](https://genaisecretsauce.com/content/images/thumbnail/1a3fef9a-be4b-41f4-8830-a94c0e8288cc-49971690-eeb2-4b37-9c3d-913df16d4285.png)](https://github.com/morluto/rea?ref=genaisecretsauce.com)

#2

### [bilawalsidhu/gods-eye-view](https://github.com/bilawalsidhu/gods-eye-view?ref=genaisecretsauce.com)

Rank yesterday: #2 - Holding steady ➡

⭐ **Stars today:** +1,961 · 📦 **Total:** 55,297  
📜 **License:** MIT · 👤 **By:** individual developer  
🎯 **Time to value:** 10 minutes

**What it is:** A browser app that puts live public data (flights, ships, satellites, earthquakes) on a realistic 3D globe, steerable by voice. Today it added Google Street View, so you can click a street and see the nearest panorama. **Why you'd want it:** A ready-made example of a voice-controlled map app over public data.

| ✓ Pros                                   | ✗ Cons                                                   |
| ---------------------------------------- | -------------------------------------------------------- |
| Most layers work with no keys            | Voice features need a paid OpenAI key                    |
| Built-in spending caps on voice sessions | Street View and realistic 3D need a paid Google Maps key |
| Free and modular                         | The surveillance-style features call for care            |

[GitHub - bilawalsidhu/gods-eye-view: A spy satellite simulator in your browser, except the data is real. Live open source spatial intelligence on a photorealistic 3D globe.A spy satellite simulator in your browser, except the data is real. Live open source spatial intelligence on a photorealistic 3D globe. - bilawalsidhu/gods-eye-view![](https://genaisecretsauce.com/content/images/icon/favicon-b0a0b404-b13b-4256-963d-8b76c432f248.png)GitHub![](https://genaisecretsauce.com/content/images/thumbnail/gods-eye-view-55c661c5-9340-4a21-b459-d9e5fe08a3aa.png)](https://github.com/bilawalsidhu/gods-eye-view?ref=genaisecretsauce.com)

#3

### [debpalash/VoiceStudio](https://github.com/debpalash/VoiceStudio?ref=genaisecretsauce.com)

Rank yesterday: not listed - New entry 🆕

⭐ **Stars today:** +2,277 · 📦 **Total:** 63,125  
📜 **License:** AGPL-3.0 · 👤 **By:** individual developer  
🎯 **Time to value:** 15 minutes

**What it is:** A free desktop app that clones voices, designs new ones from a description, dubs videos, transcribes and makes audiobooks, all on your own computer. Today's update makes it run on computers without a graphics card, while moving multi-machine features into a paid Pro version. **Why you'd want it:** Voice cloning and dubbing without paying a cloud service per character.

| ✓ Pros                                         | ✗ Cons                                                             |
| ---------------------------------------------- | ------------------------------------------------------------------ |
| Fully local, works offline once set up         | Copyleft license and some non-commercial models limit business use |
| One app for cloning, dubbing and transcription | Multi-machine features now cost extra                              |
| Agents can call it directly                    | Only clone voices you have permission to use                       |

[GitHub - debpalash/VoiceStudio: VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation in 646 languages.VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation in 646 languages. - debpalash/Voic…![](https://genaisecretsauce.com/content/images/icon/favicon-05c5da4c-a04d-49c6-b707-6fff014ebc0d.png)GitHub![](https://genaisecretsauce.com/content/images/thumbnail/fd6851a0-4e36-4541-a76c-2ad84935d0bc-0a1ce75f-7b90-4235-8dd3-8d4c6bd8212e.png)](https://github.com/debpalash/VoiceStudio?ref=genaisecretsauce.com)

#4

### [cathrynlavery/diagram-design](https://github.com/cathrynlavery/diagram-design?ref=genaisecretsauce.com)

Rank yesterday: #5 - Rising ↑

⭐ **Stars today:** +1,813 · 📦 **Total:** 54,027  
📜 **License:** MIT · 👤 **By:** individual developer  
🎯 **Time to value:** 5 minutes

**What it is:** An add-on that makes coding agents draw clean, magazine-style diagrams as single web files. New today: a check that flags anything drawn outside the visible area. **Why you'd want it:** Better-looking diagrams for docs and slides without a design tool.

| ✓ Pros                         | ✗ Cons                                   |
| ------------------------------ | ---------------------------------------- |
| Plain web files, easy to share | Opinionated minimal style                |
| Works with six coding agents   | Diagrams still need fact-checking        |
| Free and open source           | About 13 updates a day, so pin a version |

[GitHub - cathrynlavery/diagram-design: Editorial diagram design for Claude Code, Codex, GitHub Copilot, Cursor, Factory Droid, and Pi. 44 diagram types. Self-contained HTML + SVG. No shadows. No Mermaid slop.Editorial diagram design for Claude Code, Codex, GitHub Copilot, Cursor, Factory Droid, and Pi. 44 diagram types. Self-contained HTML + SVG. No shadows. No Mermaid slop. - cathrynlavery/diagram-design![](https://genaisecretsauce.com/content/images/icon/favicon-7b4fdec8-6519-415a-87bf-c09c7b3ccfb9.png)GitHub![](https://genaisecretsauce.com/content/images/thumbnail/diagram-design-ac01e97a-e8b2-4475-9c7d-6b4d6ecb7a16.png)](https://github.com/cathrynlavery/diagram-design?ref=genaisecretsauce.com)

#5

### [vercel-labs/skills](https://github.com/vercel-labs/skills?ref=genaisecretsauce.com)

Rank yesterday: not listed - New entry 🆕

⭐ **Stars today:** +187 · 📦 **Total:** 34,459  
📜 **License:** MIT · 👤 **By:** company (Vercel Labs)  
🎯 **Time to value:** 2 minutes

**What it is:** A one-command installer for "agent skills," packaged instructions that teach AI coding agents new tasks, across 79 different agents. Its npm package was downloaded about 25 million times in 30 days. **Why you'd want it:** Install the same skill into every coding agent your team uses, and keep it updated.

| ✓ Pros                                   | ✗ Cons                                                                     |
| ---------------------------------------- | -------------------------------------------------------------------------- |
| From Vercel and free                     | Usage tracking is on by default                                            |
| One installer for about 79 agents        | Third-party skills are instructions your agent will follow, so review them |
| Version locking and private-repo support | Bulk-install flags make it easy to skip looking                            |

[GitHub - vercel-labs/skills: The open agent skills tool - npx skillsThe open agent skills tool - npx skills. Contribute to vercel-labs/skills development by creating an account on GitHub.![](https://genaisecretsauce.com/content/images/icon/favicon-e46cadf7-ef31-4033-ac2b-d88c77c84877.png)GitHub![](https://genaisecretsauce.com/content/images/thumbnail/skills-d1ab45e6-10d2-4bd9-86fa-32b52ed201dc.png)](https://github.com/vercel-labs/skills?ref=genaisecretsauce.com)

#6

### [storytold/artcraft](https://github.com/storytold/artcraft?ref=genaisecretsauce.com)

Rank yesterday: #3 - Falling ↓

⭐ **Stars today:** +4,078 · 📦 **Total:** 25,532  
📜 **License:** Apache-2.0 · 👤 **By:** startup (Storytold)  
🎯 **Time to value:** 30 minutes

**What it is:** A desktop app for AI images and video where you stage the scene first, with layers, 3D props, posable characters and camera, then generate. No new release today, but its second-biggest star gain on the page. **Why you'd want it:** Repeatable control over composition, pose and camera across shots.

| ✓ Pros                            | ✗ Cons                                      |
| --------------------------------- | ------------------------------------------- |
| Permissive open-source license    | The faster rewrite is not released yet      |
| Use your own AI provider accounts | Some bundled assets are not openly licensed |
| You own what you make             | The easiest path runs on paid credits       |

[GitHub - storytold/artcraft: ArtCraft is an intentional crafting engine for artists, designers, and filmmakersArtCraft is an intentional crafting engine for artists, designers, and filmmakers - storytold/artcraft![](https://genaisecretsauce.com/content/images/icon/favicon-2bfa875c-1837-4b04-981a-0df71b8ee5c4.png)GitHub![](https://genaisecretsauce.com/content/images/thumbnail/artcraft-0b44db08-7ccf-4dc1-8d39-ed4f47bc874e.png)](https://github.com/storytold/artcraft?ref=genaisecretsauce.com)

HuggingFace Trending

## Top Models Today

#1

### [google/embeddinggemma-2](https://huggingface.co/google/embeddinggemma-2?ref=genaisecretsauce.com)

Google's small multimedia search model held the top spot for a fifth day as downloads rose 8%.

📥 **Downloads (30d):** 124,930 · 📜 **License:** Apache-2.0  
👤 **By:** Google DeepMind · 🎯 **Task:** Feature extraction (embeddings)  
📐 **Size:** 0.74B

**What it is:** A small model that turns text, code, images, video and audio into number lists (embeddings) that can be searched together. It works in 100+ languages. **Why you'd want it:** One search index for documents, screenshots, video and audio, small enough for a laptop.

| ✓ Pros                            | ✗ Cons                                           |
| --------------------------------- | ------------------------------------------------ |
| Permissive license                | Text gains over version 1 are small outside code |
| One index for every media type    | Google's own benchmark numbers                   |
| Small enough for on-device search | Tool support still thin                          |

[google/embeddinggemma-2 · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.![](https://genaisecretsauce.com/content/images/icon/favicon-5f3b74aa-0b62-421d-a192-d2f8c52f1ce4.ico)![](https://genaisecretsauce.com/content/images/thumbnail/embeddinggemma-2-64d377a2-8bf2-46b4-a39f-0af5c6050a74.png)](https://huggingface.co/google/embeddinggemma-2?ref=genaisecretsauce.com)

#2

### [Qwen/Qwen-Image-2.1-Turbo](https://huggingface.co/Qwen/Qwen-Image-2.1-Turbo?ref=genaisecretsauce.com)

Alibaba's Qwen image generator and editor, sped up to 8 steps, with downloads up 42% again.

📥 **Downloads (30d):** 9,665 · 📜 **License:** Qwen Research (non-commercial)  
👤 **By:** Qwen (Alibaba) · 🎯 **Task:** Text-to-image  
📐 **Size:** 7.1B

**What it is:** An official fast version of Qwen's image model that both creates and edits pictures. It plugs into the popular Diffusers toolkit. **Why you'd want it:** Quick, good-quality image generation and editing on your own hardware.

| ✓ Pros                     | ✗ Cons                                       |
| -------------------------- | -------------------------------------------- |
| Official release from Qwen | Non-commercial license                       |
| Generates and edits        | Needs the newest Diffusers version           |
| Only 8 steps per image     | Unofficial "uncensored" copies are spreading |

[Qwen/Qwen-Image-2.1-Turbo · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.![](https://genaisecretsauce.com/content/images/icon/favicon-85ccf279-ec7e-46e1-9b35-e36d272147c8.ico)![](https://genaisecretsauce.com/content/images/thumbnail/Qwen-Image-2.1-Turbo-0d6f2fab-0a25-469e-93a0-8e926073789f.png)](https://huggingface.co/Qwen/Qwen-Image-2.1-Turbo?ref=genaisecretsauce.com)

#3

### [Lightricks/LTX-2.5](https://huggingface.co/Lightricks/LTX-2.5?ref=genaisecretsauce.com)

The go-to open video model for turning still images into clips on your own machine.

📥 **Downloads (30d):** 1,724,548 · 📜 **License:** LTX community license  
👤 **By:** Lightricks · 🎯 **Task:** Image-to-video  
📐 **Size:** not listed

**What it is:** A video generator that animates images and text prompts into short clips. It has a large ecosystem of add-ons in the ComfyUI workflow tool. **Why you'd want it:** Make video locally without per-clip cloud fees.

| ✓ Pros                    | ✗ Cons                         |
| ------------------------- | ------------------------------ |
| Runs locally              | License has revenue thresholds |
| Huge community of add-ons | Needs a lot of graphics memory |
| Mature and widely tested  | Downloads have flattened       |

[Lightricks/LTX-2.5 · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.![](https://genaisecretsauce.com/content/images/icon/favicon-b742f12b-97db-42a3-b700-f40f994ca4f2.ico)![](https://genaisecretsauce.com/content/images/thumbnail/LTX-2.5-5a74bed1-5663-455e-be15-6b13ab7cc570.png)](https://huggingface.co/Lightricks/LTX-2.5?ref=genaisecretsauce.com)

#4

### [openbmb/MiniCPM-V-4.7-35B-A3B](https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B?ref=genaisecretsauce.com)

OpenBMB's new vision model, which the team says runs a 35-billion-parameter AI on a phone at 13+ tokens per second.

📥 **Downloads (30d):** 630 · 📜 **License:** Apache-2.0  
👤 **By:** OpenBMB · 🎯 **Task:** Image-text-to-text  
📐 **Size:** 35B (about 3B active)

**What it is:** A model that understands single images, multiple images and video, using only about 3 billion of its parameters per word. A trick that preloads the parts it will need next lets the full model run on an iPhone 17 Pro Max, the team says. **Why you'd want it:** Strong on-device image and video understanding without sending pictures to the cloud.

| ✓ Pros                          | ✗ Cons                                             |
| ------------------------------- | -------------------------------------------------- |
| Permissive license              | Phone speed relies on OpenBMB's own tooling        |
| Runs on phones and many chips   | Benchmarks are self-reported, shown only as images |
| Works with Ollama and llama.cpp | Full-precision files are about 70GB                |

[openbmb/MiniCPM-V-4.7-35B-A3B · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.![](https://genaisecretsauce.com/content/images/icon/favicon-5cf8a5b6-ae5c-4dbd-a5f7-e6839f6af732.ico)![](https://genaisecretsauce.com/content/images/thumbnail/MiniCPM-V-4.7-35B-A3B-8c7f50d6-4e43-4da4-89be-1ca49b633868.png)](https://huggingface.co/openbmb/MiniCPM-V-4.7-35B-A3B?ref=genaisecretsauce.com)

#5

### [Cloudflare/clef](https://huggingface.co/Cloudflare/clef?ref=genaisecretsauce.com)

Cloudflare's "decision model" answers typed questions with probabilities instead of text.

📥 **Downloads (30d):** 17,173 · 📜 **License:** Apache-2.0  
👤 **By:** Cloudflare · 🎯 **Task:** Image-text-to-text (decisions)  
📐 **Size:** 27.4B

**What it is:** A model that picks among fixed options and says how confident it is, built on Qwen. It also runs on Cloudflare's Workers AI service. **Why you'd want it:** Reliable classification and routing decisions you can set thresholds on.

| ✓ Pros                       | ✗ Cons                                 |
| ---------------------------- | -------------------------------------- |
| Calibrated confidence scores | Needs a custom loader                  |
| Permissive license           | A fine-tune of another company's model |
| Accepts images               | Sensitive to how questions are worded  |

[Cloudflare/clef · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.![](https://genaisecretsauce.com/content/images/icon/favicon-0db6899d-160a-41ec-a072-2ba7852aeb6b.ico)![](https://genaisecretsauce.com/content/images/thumbnail/clef-074f1967-7703-452c-991d-19cfc5c745c7.png)](https://huggingface.co/Cloudflare/clef?ref=genaisecretsauce.com)

#6

### [LiquidAI/d1-3B](https://huggingface.co/LiquidAI/d1-3B?ref=genaisecretsauce.com)

A laptop-sized decision model that answers typed questions in milliseconds.

📥 **Downloads (30d):** 13,190 · 📜 **License:** LFM 1.0 (custom)  
👤 **By:** Liquid AI · 🎯 **Task:** Image-text-to-text (decisions)  
📐 **Size:** 3.1B

**What it is:** A small model that scores fixed options with probabilities, which Liquid AI says takes about 8 milliseconds per decision on a gaming graphics card. **Why you'd want it:** Fast, cheap yes/no and multiple-choice decisions on modest hardware.

| ✓ Pros                        | ✗ Cons                                   |
| ----------------------------- | ---------------------------------------- |
| Small enough for edge devices | Custom license                           |
| Very low latency              | Needs remote code to load                |
| Accepts images                | Speed figures come from Liquid AI itself |

[LiquidAI/d1-3B · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.![](https://genaisecretsauce.com/content/images/icon/favicon-75fe39a7-1396-495c-b915-c93d48e419d9.ico)![](https://genaisecretsauce.com/content/images/thumbnail/d1-3B-27085859-f6c9-4918-ace3-b592e672500b.png)](https://huggingface.co/LiquidAI/d1-3B?ref=genaisecretsauce.com)

#7

### [autotrust/JEV-35B](https://huggingface.co/autotrust/JEV-35B?ref=genaisecretsauce.com)

A new open decision model that ranks up to 256 options with probabilities in one pass.

📥 **Downloads (30d):** 17,426 · 📜 **License:** Apache-2.0  
👤 **By:** AutoTrust AI · 🎯 **Task:** Text classification (decisions)  
📐 **Size:** 35B (about 3B active)

**What it is:** A fine-tune of Qwen3.5-35B-A3B that answers typed questions with calibrated probabilities. AutoTrust reports a median response time of 204 milliseconds on one high-end Nvidia chip. **Why you'd want it:** Rank or classify many options at once with confidence scores.

| ✓ Pros                                            | ✗ Cons                                               |
| ------------------------------------------------- | ---------------------------------------------------- |
| Permissive license                                | Benchmarks are self-scored                           |
| Up to 256 options per call                        | Built by a small company                             |
| A compressed version fits one large graphics card | The compressed version needs the newest Nvidia chips |

[autotrust/JEV-35B · Hugging FaceWe’re on a journey to advance and democratize artificial intelligence through open source and open science.![](https://genaisecretsauce.com/content/images/icon/favicon-0b762788-8a2d-460c-b535-e1c2a8ae5bd2.ico)![](https://genaisecretsauce.com/content/images/thumbnail/JEV-35B-5afe17da-f1a7-4f4c-8869-887240b52c31.png)](https://huggingface.co/autotrust/JEV-35B?ref=genaisecretsauce.com)

Product Hunt

## AI Launches Today

### [Articos MCP](https://www.producthunt.com/products/articos?ref=genaisecretsauce.com)

See how simulated buyers react, right in your agent

🔥 **Upvotes:** 298 · 👤 **By:** Articos  
💰 **Pricing:** free trial, paid from $79/month · 🏷 **Category:** Marketing research

Articos builds a panel of AI "personas" matched to your ideal customer and has them react to your messaging, ads or landing pages, with their reasoning, in under 30 minutes. Today's launch lets you run those studies from inside Claude, ChatGPT, Gemini, Copilot or Cursor. It led Tuesday's board at time of writing. **Verdict:** A cheap first filter for marketing copy, but simulated buyers are a model's guess at people, and the accuracy claims are the company's own.

[Articos: Launch with confidence, not gut instinct | Product HuntArticos helps SaaS teams, agencies, and founders test messaging, positioning, and landing pages against simulated personas matched to your ICP. Real audience signal in under 30 minutes. Peer-reviewed at 86% human accuracy vs Baymard and Nielsen Norman.![](https://genaisecretsauce.com/content/images/icon/ph-favicon-brand-500-d1766e4f-ddb3-4035-beb0-797bd71faf70.png)Shaheer Gadit and Owais KhanProduct Hunt![](https://genaisecretsauce.com/content/images/thumbnail/bf8a4aba-000b-4cce-86ca-fb88380bb028-bb56b613-bab7-4bcf-a38f-8f67c4b38b3c.jpg)](https://www.producthunt.com/products/articos?ref=genaisecretsauce.com)

### [OpenCode 2.0](https://www.producthunt.com/products/opencode?ref=genaisecretsauce.com)

The open source AI coding agent

🔥 **Upvotes:** 236 · 👤 **By:** Anomaly  
💰 **Pricing:** free and open source, optional $10/month hosted models · 🏷 **Category:** Developer tools

A free coding agent for the terminal, desktop and web that works with 75+ AI providers or your existing ChatGPT or GitHub Copilot login. Version 2 rebuilds its plugin system and settings, so old plugins stop working. **Verdict:** The leading open-source alternative to Claude Code and Codex, but plugin users should wait until their add-ons are updated.

[opencode: Your terminal’s AI agent, with any model you want | Product Huntopencode is an open-source AI coding agent built for the terminal. It features a native TUI, multi-session support, and is model-agnostic, letting you use Claude, OpenAI, Google, or even local models.![](https://genaisecretsauce.com/content/images/icon/ph-favicon-brand-500-04446aa0-521f-40cc-9e19-04206eb468e2.png)Brendan and Fang and Luke ParkerProduct Hunt![](https://genaisecretsauce.com/content/images/thumbnail/8e52eeef-b98d-48b7-b0e3-9579fc3864b9-9e109447-2261-4cf4-9068-6e8cf8f5a755.jpg)](https://www.producthunt.com/products/opencode?ref=genaisecretsauce.com)

### [Kai](https://www.producthunt.com/products/hirekai?ref=genaisecretsauce.com)

Autonomous assistant designed to handle meetings and email

🔥 **Upvotes:** 171 · 👤 **By:** the team behind the Morgen calendar app  
💰 **Pricing:** free trial, paid from $20/seat/month · 🏷 **Category:** Productivity

An AI executive assistant that sorts your email and drafts replies in your voice, schedules meetings, briefs you beforehand and turns recordings into action items. It can also be used from Claude, ChatGPT or Gemini. **Verdict:** A credible team with a shipped product, but you are giving an AI access to your email, calendar and meetings, so check data handling first.

[Kai: Autonomous assistant designed to handle meetings and email | Product HuntKai is an AI executive assistant for solopreneurs, freelancers and startup founders. It works across your email, calendar and tasks: it triages your inbox and drafts replies, schedules around your real availability, records and transcribes your meetings, then turns them into summaries and action items. Workflows run in the background, and Kai hands tasks to Claude or ChatGPT with the context of the meeting or email they came from. Available on Web, desktop and mobile app. Free trial available.![](https://genaisecretsauce.com/content/images/icon/ph-favicon-brand-500-e84e91fe-d01f-4723-9632-ce94d316f41a.png)John Mavrick and Danny Hatcher 🔎 and André JonasProduct Hunt![](https://genaisecretsauce.com/content/images/thumbnail/a20e8ef7-b6b3-4abd-a22d-0bdf43f5dc94-d4a4be96-4173-4c06-b3f1-d6c594b4da15.jpg)](https://www.producthunt.com/products/hirekai?ref=genaisecretsauce.com)

### [Takes for Mac](https://www.producthunt.com/products/takes-for-mac?ref=genaisecretsauce.com)

Record talking-head videos. Claude Code does the edit.

🔥 **Upvotes:** 118 · 👤 **By:** Marv (solo maker)  
💰 **Pricing:** free, AI edits use your own Claude plan · 🏷 **Category:** Video

A free, open-source Mac app with a teleprompter that follows your voice. Claude Code then cuts silences and false starts, makes a vertical version and drafts social posts. **Verdict:** A clever no-subscription approach, but it is a brand-new one-person project shared "as is."

[Takes for Mac: Record talking-head videos. Claude Code does the edit. | Product HuntA free, open-source Mac app for talking-head videos. A teleprompter that follows your voice, one-button recording of camera and screen, and every take in plain folders. Claude Code, on your own Claude plan, cuts the edit, makes the vertical version and drafts the posts. Leave notes on the frame; it turns them into rules and remembers them.![](https://genaisecretsauce.com/content/images/icon/ph-favicon-brand-500-a2432ce0-a372-4361-9e64-1aea8bdca61f.png)Marvin AzizProduct Hunt![](https://genaisecretsauce.com/content/images/thumbnail/df241800-c38b-4b99-97ed-005120d71737-7dacf23c-4f5d-45d8-82ee-e43bcb61ddbc.jpg)](https://www.producthunt.com/products/takes-for-mac?ref=genaisecretsauce.com)

### [Toby](https://www.producthunt.com/products/toby-5?ref=genaisecretsauce.com)

Realistic mock interviews with an AI hiring manager

🔥 **Upvotes:** 103 · 👤 **By:** Toby (UK)  
💰 **Pricing:** first 5 minutes free, then £0.50 a minute · 🏷 **Category:** Careers

Paste a job description and Toby researches the company, maps likely interview rounds, then interviews you on a video call. The debrief scores each answer and rewrites it using only what you said. **Verdict:** Honest pay-per-minute pricing for something most people need a few times a year, but double-check its AI-generated company research.

[Toby: Realistic mock interviews with an AI hiring manager | Product HuntPaste a job description. Toby researches the company, maps every round and interviews you on a Meet, Teams or Zoom-style call. Then you get a scored debrief and a stronger version of each answer, built only from what you said. Real mode runs it straight through; Coach mode gives a note and a retry. No subscription: 50p a minute on a call. First 5 minutes free.![](https://genaisecretsauce.com/content/images/icon/ph-favicon-brand-500-4cb7a970-5817-46c1-9abc-cf31ac49a654.png)Rob KealeyProduct Hunt![](https://genaisecretsauce.com/content/images/thumbnail/08931473-7144-4d60-9b27-19161f4fbbfb-3c66ad13-394f-4e2a-ab74-2b0ba77ab91c.jpg)](https://www.producthunt.com/products/toby-5?ref=genaisecretsauce.com)

API Pricing

## Snapshot

Provider

Model

Input $/1M

Output $/1M

Context

Anthropic

Claude Fable 5.1

$10.00

$50.00

1M tokens

Anthropic

Claude Opus 5.5

$4.00

$20.00

1M tokens

Anthropic

Claude Sonnet 5.5

$2.00

$10.00

1M tokens

Anthropic

Claude Haiku 5.5

$0.10 (up to 100K-token prompts; $0.50 above)

$0.50 ($2.50 above 100K)

1M tokens

OpenAI

GPT-6 Astra

$10.00 (up to 272K-token prompts; $20.00 above)

$50.00 ($75.00 above 272K)

1.05M tokens

OpenAI

GPT-6.1 Sol

$2.00 (up to 272K; $4.00 above)

$10.00 ($15.00 above 272K)

1.05M tokens

OpenAI

GPT-6 Luna

$0.10 (up to 272K; $0.20 above)

$0.50 ($0.75 above 272K)

1.05M tokens

Google

Gemini 4 Argon

$2.00 to Nov 30; $4.00 from Dec 1

$10.00 to Nov 30; $20.00 from Dec 1

1M tokens (input and output combined)

Google

Gemini 3.8 Flash

$0.75 (to Dec 31; $1.50 from Jan 1)

$3.75 (to Dec 31; $7.50 from Jan 1)

not listed

Groq

GPT-OSS 120B

$0.15

$0.60

131K tokens

Groq

Qwen3.8-27B

$0.80

$4.00

131K tokens

**No list-price changes vs. October 12** \- but Gemini 4 Argon now appears on Google's official pricing page, with a firm date for its price to double (December 1) and an official 1M-token window. It is available on the paid tier only.  
  
**What this means:** Until November 30, Argon costs the same as GPT-6.1 Sol and Claude Sonnet 5.5, and one-fifth of GPT-6 Astra or Claude Fable 5.1\. From December 1 it matches Claude Opus 5.5, so heavy users have about seven weeks of cheap testing.  
  
arXiv Paper of the Day

## KV-Repurpose: A Testbed for Evaluating Non-Prefix KV-Cache Reuse

Taiga Ikeda, Daisuke Miyashita, Jun Deguchi (Kioxia) · arXiv:2610.13060

**What it claims:** AI servers today reuse saved work on a prompt only when the start of the prompt matches exactly, so the same document gets reprocessed whenever it follows different instructions. The authors release an open-source add-on for the vLLM serving software that reuses saved work for each document wherever it appears, and measure when that is safe.  
  
**Key finding:** **With the same 30 GB cache, segment reuse hit 63.9% of tokens versus 3.3% for standard prefix caching, but answers combining several documents got noticeably worse (2WikiMultihopQA 0.80 to 0.50).**  
  
**Why practitioners should care:** If you run document search or agents on your own GPUs, this shows far bigger cache savings are possible for single-document questions and summaries, and an Apache-licensed testbed to try them. Avoid it for questions that join several documents. Caveats: one model, one GPU setup and a synthetic workload, from a company that sells storage drives.  
  
[Read on arXiv →](https://arxiv.org/abs/2610.13060?ref=genaisecretsauce.com)

GenAI Secret Sauce Daily Digest · 2026-10-13

×

Click anywhere or press ESC to close