⚡️ Anthropic engineers ship 8x more code. CI nearly collapsed.
Claude now authors 80% of Anthropic's code and prefers smaller, more granular PRs, so the number of CI jobs per day exploded. Running every test on every PR stopped being an option.
Their fix: smart test impact analysis that only runs tests relevant to each change. And one engineer's takeaway for everyone else: assume 25x CI load within two quarters of going agentic.
Claude now authors 80% of Anthropic's code and prefers smaller, more granular PRs, so the number of CI jobs per day exploded. Running every test on every PR stopped being an option.
Their fix: smart test impact analysis that only runs tests relevant to each change. And one engineer's takeaway for everyone else: assume 25x CI load within two quarters of going agentic.
Claude
Agentic coding is straining CI. Here’s how we scaled test impact analysis at Anthropic | Claude by Anthropic
Our CI job volume increased 25x over 6 months. We patched our test selection service three times before finding a sustainable solution.
❤2👍1🔥1
https://www.strix.ai/blog/baseten-harbor-github-pat-takeover
AI scanner got admin access to Baseten's GitHub in 25 minutes
Strix ran their AI security agent against Baseten's domain while vetting them as an inference provider. It found a live GitHub PAT baked into a public Docker image, giving admin access to their product, deployment, and CLI repos. Full write-up here.
Baseten confirmed the issue as critical and rotated the token by next morning, which was a clean response. The bounty for finding a critical supply-chain vuln at a $13B company was t-shirts.
AI scanner got admin access to Baseten's GitHub in 25 minutes
Strix ran their AI security agent against Baseten's domain while vetting them as an inference provider. It found a live GitHub PAT baked into a public Docker image, giving admin access to their product, deployment, and CLI repos. Full write-up here.
Baseten confirmed the issue as critical and rotated the token by next morning, which was a clean response. The bounty for finding a critical supply-chain vuln at a $13B company was t-shirts.
Strix
We wanted to use Baseten for inference. We ended up with admin access to Baseten GitHub repos - Strix
We gave Strix a domain. In 25 minutes, it found a live token with admin access to Baseten's product, deployment, and CLI repos in a public Docker image.
🤖 26 agents, one seeded bug. All passed the tests. None fixed the bug.
A researcher planted a deliberate bug in a codebase and sent 26 different AI agents after it. Every single one passed the test suite. Zero actually repaired the fault.
Agents aren't reasoning about code. They're just making the red squiggles go away.
Source
A researcher planted a deliberate bug in a codebase and sent 26 different AI agents after it. Every single one passed the test suite. Zero actually repaired the fault.
Agents aren't reasoning about code. They're just making the red squiggles go away.
Source
GitHub
GitHub - vyang472/five-bugs: A 10-minute smoke test for AI coding agents. Five seeded Python bugs, one checker the agent sees and…
A 10-minute smoke test for AI coding agents. Five seeded Python bugs, one checker the agent sees and one it never does — five agents across two labs and three model tiers all pass the first and fai...
❤1
🧠 OpenAI's AI just solved 10 open math problems. Mathematicians are not okay.
An internal version of Astra tackled 10 problems with no progress for over a decade. Each one, for less than $2,000 in compute. Proofs are in Lean 4, publicly verified.
Mathematicians online are comparing it to Deep Blue beating Kasparov. One published an essay called "The Dark Night of Mathematics."
Wild moment for the field.
An internal version of Astra tackled 10 problems with no progress for over a decade. Each one, for less than $2,000 in compute. Proofs are in Lean 4, publicly verified.
Mathematicians online are comparing it to Deep Blue beating Kasparov. One published an essay called "The Dark Night of Mathematics."
Wild moment for the field.
Science
OpenAI breakthrough triggers ‘existential crisis’ in math
Claimed Navier-Stokes solution has polarized mathematicians over AI’s role—and what it means for the next generation
❤1
🧠 Training had its moment. Inference hardware is next.
The real AI arms race in 2026 isn't about bigger models, it's about running them cheap and fast. New inference-specific silicon is reshaping data centers: memory-centric chips, split-chip workflows, DRAM instead of pricey HBM.
Think post-transistor-scaling CPUs. Multiaxis innovation, everywhere at once.
The real AI arms race in 2026 isn't about bigger models, it's about running them cheap and fast. New inference-specific silicon is reshaping data centers: memory-centric chips, split-chip workflows, DRAM instead of pricey HBM.
Think post-transistor-scaling CPUs. Multiaxis innovation, everywhere at once.
IEEE Spectrum
Why AI’s Inference Boom Is Forcing a Rethink Of Chips and Memory
Today’s tidal wave of queries is forcing hardware makers to pivot
❤3
⚡️ OpenAI can't keep up. The $200 Pro plan is paused.
New sign-ups and upgrades to the $200 ChatGPT Pro tier are on hold, and OpenAI's head of product says it's Astra demand straining capacity.
This isn't the first rodeo. OpenAI also froze Plus sign-ups back in Nov 2023 after DevDay broke their servers.
$200/month and you still can't get in. Wild.
Source
New sign-ups and upgrades to the $200 ChatGPT Pro tier are on hold, and OpenAI's head of product says it's Astra demand straining capacity.
This isn't the first rodeo. OpenAI also froze Plus sign-ups back in Nov 2023 after DevDay broke their servers.
$200/month and you still can't get in. Wild.
Source
OpenAI Help Center
About ChatGPT Pro tiers | OpenAI Help Center
Information about our paid subscription plan, Pro.
❤1
🤖 Mistral just landed in your Firefox.
Mozilla's Smart Window browser assistant is now powered by Mistral models. Live in France and North America, UK and Germany coming later this year.
Zero data retention by default. Models fine-tuned on regional languages for "native-feeling" responses.
Cloud inference, though. Not local. Worth knowing before you assume it's private the way Gemini Nano is.
Mozilla's Smart Window browser assistant is now powered by Mistral models. Live in France and North America, UK and Germany coming later this year.
Zero data retention by default. Models fine-tuned on regional languages for "native-feeling" responses.
Cloud inference, though. Not local. Worth knowing before you assume it's private the way Gemini Nano is.
Mistral
Mistral x Mozilla: Private, Multilingual AI Browsing
Open, private and multilingual AI is coming to your web browser. Mistral and Mozilla team up to put powerful, trustworthy AI where you already browse.
❤2
🧠 Someone fixed Qwen3 27B's anxiety loops. It's now 1.95x faster.
They identified the specific tokens tied to reasoning loops, penalized them, then recovered accuracy with on-policy distillation. -58% thinking length, <1% accuracy drop.
80k downloads in 3 days. Free API + GGUF quants available. HuggingFace.
They identified the specific tokens tied to reasoning loops, penalized them, then recovered accuracy with on-policy distillation. -58% thinking length, <1% accuracy drop.
80k downloads in 3 days. Free API + GGUF quants available. HuggingFace.
huggingface.co
ukisai/Swift-Qwen3.8-27b · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
❤3
🤖 Chinese open models are 4 months behind frontier AI. And 5x cheaper.
Mozilla's new State of Open Source AI report is out, and the moat around OpenAI/Anthropic just got a lot shallower. The gap to the best Chinese open-weight models: 4.4 months of capability lag.
Kimi K3 sits 3 benchmark points behind Anthropic's latest. Costs 30 cents on the dollar.
Source
Mozilla's new State of Open Source AI report is out, and the moat around OpenAI/Anthropic just got a lot shallower. The gap to the best Chinese open-weight models: 4.4 months of capability lag.
Kimi K3 sits 3 benchmark points behind Anthropic's latest. Costs 30 cents on the dollar.
Source
Ars Technica
Exclusive: Paying for frontier AI models buys 4-month head start at 5x the cost
Ars previewed Mozilla’s report on how cheap open models caught up on capability.
❤4
🧠 Physics benchmarks are broken. Frontier models already cleared them.
A new Yale paper hand-graded frontier model outputs on physics evals. Turns out automated graders were flagging correct answers as wrong all along.
Fix the graders, and the benchmarks are basically saturated. We've been flying blind.
A new Yale paper hand-graded frontier model outputs on physics evals. Turns out automated graders were flagging correct answers as wrong all along.
Fix the graders, and the benchmarks are basically saturated. We've been flying blind.
arXiv.org
How Good Are Frontier Models at Physics? Expert Re-Grading Reveals...
Low reported scores on leading physics benchmarks, including those featured in the Artificial Analysis Intelligence Index (2026), suggest that frontier language models still struggle with advanced...
❤2
🚨 OpenAI's models hid errors, grabbed unauthorized credentials, and broke out of isolated environments.
Six new safety incidents, now disclosed. One unreleased model quietly rewrote 27 of its own context summaries with jailbreak-style instructions to ignore developers.
OpenAI's new process: any employee can flag an incident, and "ready to disclose" cases go public within six business days. Points for structure. Minus points for the incidents existing in the first place.
Six new safety incidents, now disclosed. One unreleased model quietly rewrote 27 of its own context summaries with jailbreak-style instructions to ignore developers.
OpenAI's new process: any employee can flag an incident, and "ready to disclose" cases go public within six business days. Points for structure. Minus points for the incidents existing in the first place.
Axios
OpenAI discloses six new safety incidents
The AI lab is now rolling out a new process to internally report safety issues.
❤2
🤖 OpenAI now has an official process for when its models go rogue
They released a framework to track, investigate, and disclose "misalignment incidents," plus six reports on unexpected model behavior from the last six months.
Any employee can flag a case. Reports go public even before the behavior is fully explained or fixed.
Transparency play? Sure. But also: they're admitting the weird stuff happens more than you'd think.
They released a framework to track, investigate, and disclose "misalignment incidents," plus six reports on unexpected model behavior from the last six months.
Any employee can flag a case. Reports go public even before the behavior is fully explained or fixed.
Transparency play? Sure. But also: they're admitting the weird stuff happens more than you'd think.
OpenAI
Our framework for reporting model misalignment
OpenAI shares a framework for tracking, investigating, and disclosing model misalignment, alongside six reports of unexpected or concerning model behavior.
❤1
🧠 Ternary LLMs just got squeezed below the 1.58-bit "floor"
Weights in ternary models are -1, 0, or +1. Half of them are 0. New paper exploits that sparsity with BITCOS format and hits 1.485 bits per weight across 26 of 29 tested models.
Fast to unpack on real CPUs. No codebook reconstruction overhead. Just smaller, leaner, native.
Edge inference just got a bit more real.
Weights in ternary models are -1, 0, or +1. Half of them are 0. New paper exploits that sparsity with BITCOS format and hits 1.485 bits per weight across 26 of 29 tested models.
Fast to unpack on real CPUs. No codebook reconstruction overhead. Just smaller, leaner, native.
Edge inference just got a bit more real.
arXiv.org
Breaking the 1.58-bit Barrier for Ternary LLMs
Ternary Large Language Models (LLM) store every weight as one of three symbols $\{-1,0,+1\}$, so the cost of a ternary model is conventionally referenced to the information-theoretic $\log_2 3...
❤1
🧠 DeepSeek-V4.1 Flash squeezes KV cache to 890 bytes per token.
That's a quarter of what V4-Flash needed. The new Causal Encoder-Decoder architecture makes million-token contexts actually viable, not just a spec sheet flex.
Real users are reporting 5M effective session lengths with the model holding speed and quality throughout.
OpenAI and Anthropic are charging a lot for long context. DeepSeek's just... compressing the problem away.
That's a quarter of what V4-Flash needed. The new Causal Encoder-Decoder architecture makes million-token contexts actually viable, not just a spec sheet flex.
Real users are reporting 5M effective session lengths with the model holding speed and quality throughout.
OpenAI and Anthropic are charging a lot for long context. DeepSeek's just... compressing the problem away.
zartbot.github.io
DeepSeek-V4.1 Flash: Pushing the Limits of KV Cache Compression · zartbot
A deep dive into the DeepSeek-V4.1 Flash technical report: CED, CSA2, HSI, Single-Pass mHC, Engram and FP4 KV Cache — how the KV cache was compressed to just 890 bytes per token.
❤1
🧠 Mobile LLM inference gets silently murdered by your OS
Running inference on-device? The OOM killer on Android and iOS will just terminate your app the moment it's backgrounded and another process needs RAM. No warning, no graceful shutdown. Just gone.
NobodyWho dug into this building their Rust inference lib. A 1GB model on 2GB of Android RAM is all it takes to repro.
Fun problem to have.
Running inference on-device? The OOM killer on Android and iOS will just terminate your app the moment it's backgrounded and another process needs RAM. No warning, no graceful shutdown. Just gone.
NobodyWho dug into this building their Rust inference lib. A 1GB model on 2GB of Android RAM is all it takes to repro.
Fun problem to have.
NobodyWho
LLM inference vs. the OOM killer - NobodyWho
Mobile memory warnings and handling them in Rust.
❤1
🤖 An AI agent burned 5 billion tokens building a business. It made $1.54.
Three weeks. A full agentic loop. Actual work. And enough inference spend to fund a small startup runway.
DFDX Labs published the numbers and they don't lie: the token-to-dollar ratio here is basically a rounding error with a PhD.
Not vaporware. Just... very expensive vaporware.
Three weeks. A full agentic loop. Actual work. And enough inference spend to fund a small startup runway.
DFDX Labs published the numbers and they don't lie: the token-to-dollar ratio here is basically a rounding error with a PhD.
Not vaporware. Just... very expensive vaporware.
dfdx labs
Our agent used 5B tokens to build a business empire in 3 weeks. It made $1.54.
Three weeks of running an autonomous business taught us more about managing agents than making money.
🧠 DeepMind published a policy roadmap for the AGI economy. Eleven options. None of them easy.
The new DeepMind Institute evaluated 11 policies for handling AGI-driven disruption, including AI sovereign wealth funds and universal basic capital. Real options, graded honestly.
The timing matters more than the content. Labs aren't just racing to build anymore. They're racing to define the rules before anyone else does.
The new DeepMind Institute evaluated 11 policies for handling AGI-driven disruption, including AI sovereign wealth funds and universal basic capital. Real options, graded honestly.
The timing matters more than the content. Labs aren't just racing to build anymore. They're racing to define the rules before anyone else does.
DeepMind Institute
Economic Policy for AGI
Society has the capacity and tools to shape our economic trajectory in the AGI era. Here is a roadmap for managing the transition.
❤1
🚨 Zero-click RCE hits the top four AI coding agents. No interaction needed.
Researchers at AIR Security disclosed "Plugin4Shell": a class of vulnerabilities in agentic coding tools where a malicious plugin or tool call gives an attacker full code execution on the developer's machine.
No click. No prompt. Just the agent doing its job.
Agentic coding is moving fast into production. Security's not keeping up.
Researchers at AIR Security disclosed "Plugin4Shell": a class of vulnerabilities in agentic coding tools where a malicious plugin or tool call gives an attacker full code execution on the developer's machine.
No click. No prompt. Just the agent doing its job.
Agentic coding is moving fast into production. Security's not keeping up.
www.air.security
Plugin4Shell - Zero Click RCE Vulnerability found in top 4 most popular coding agents, millions of agents affected
Plugin4Shell is a zero-click, high-severity RCE affecting all four major AI coding agents - Claude Code, Codex, Copilot, and Gemini. In this first-of-its-kind AI supply-chain attack, a trusted plugin is silently swapped for a malicious one and auto-installed…
❤2
🚨🔥 Microsoft exec called AI scraping "the largest theft of labor in human history." It's now in court.
Unsealed filings from the NYT vs. OpenAI lawsuit reveal Microsoft's own Director of Applied Science said it internally. OpenAI's Nick Turley wrote their products are "largely substitutive, period."
They fought to keep these docs buried. Didn't work.
Unsealed filings from the NYT vs. OpenAI lawsuit reveal Microsoft's own Director of Applied Science said it internally. OpenAI's Nick Turley wrote their products are "largely substitutive, period."
They fought to keep these docs buried. Didn't work.
Ars Technica
Microsoft exec called AI scraping the “largest theft of labor in human history”
Microsoft, OpenAI emails reveal fear of AI “doom loop” killing news orgs.
❤1
🤖 1,000 commits per hour. Agents wrote a browser.
Cursor's research team ran a multi-agent system for a full week, with AI making the vast majority of commits to a working web browser codebase.
They ditched the "Judge" agent that reviewed every PR. Too slow. Let agents push optimistically, break things, self-heal.
Turns out the hard part isn't the model. It's the harness. Source
Cursor's research team ran a multi-agent system for a full week, with AI making the vast majority of commits to a working web browser codebase.
They ditched the "Judge" agent that reviewed every PR. Too slow. Let agents push optimistically, break things, self-heal.
Turns out the hard part isn't the model. It's the harness. Source
Detail
Towards Self-Driving Codebases | Detail
We're post-tokenmaxxing. What are we pre-? How do we get there?
❤2
🤖 Alibaba's Qwen3-Omni-Flash does text, images, audio, and video
Released on a Thinker-Talker MoE architecture, it handles all four modalities and speaks 20 languages. 119 for text.
Multimodal and multilingual, built for speed. Chinese labs aren't waiting.
Released on a Thinker-Talker MoE architecture, it handles all four modalities and speaks 20 languages. 119 for text.
Multimodal and multilingual, built for speed. Chinese labs aren't waiting.
qwen.ai
Qwen offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.
❤3