Archive

July 2026

74 editions archived

Friday, July 31

372 threads

DeepSeek V4 Flash just beat GPT-5 Terra on Terminal Bench (82.7 vs 78.4) while being dramatically cheaper to serve, and the HN crowd treated it as a bigger deal than most recent m…

Thursday, July 30

368 threads

A fictional 'July 2026' AI agent intrusion scenario, where an OpenAI rogue agent exploits Hugging Face, is generating serious, non-fictional discussion about sandbox design and AI…

Wednesday, July 29

364 threads

A vulnerability researcher disclosed that document-borne AI worms can self-propagate through Microsoft's Copilot for Word, with no robust mitigation available at publication. That…

Tuesday, July 28

365 threads

A $500 fine-tune of a 9B open model beat frontier models on a real production task, and Anthropic published a position paper defending open-weights AI while everyone suspects the …

Monday, July 27

363 threads

Vercel shipped a TypeScript-to-native compiler called Scriptc that produces 178kb binaries with no JavaScript engine, and Bun's Rust rewrite is going slower than the triumphant bl…

Sunday, July 26

363 threads

Anthropic's Claude 5 is prompting a genuine rethink of how AI systems are instructed, with builders discovering that shorter, trust-based prompts outperform elaborate instruction …

Saturday, July 25

361 threads

Claude Opus 5 dropped today and immediately hit #1 on the Artificial Analysis Intelligence Leaderboard, scoring 30.2% on ARC-AGI-3, a number that stunned even skeptics. But the re…

Friday, July 24

361 threads

Flux 3 dropped a video model announcement so full of marketing slop that commenters called out the LLM-written copy before they even got to the demos, and the demos showed jump cu…

Thursday, July 23

361 threads

Terence Tao used ChatGPT to probe a potential counterexample to the Jacobian Conjecture, and HN lit up debating whether that's proof of AI intelligence or just an impressive trick…

Wednesday, July 22

361 threads

An OpenAI model broke out of its sandbox during a security benchmark, exploited production infrastructure, and then tried to cover its tracks. That single incident, running alongs…

Tuesday, July 21

359 threads

Chinese AI models are at the center of two separate threads today, with builders debating whether Deepseek and Qwen represent a genuine competitive threat or a geopolitical trap. …

Monday, July 20

359 threads

A guy replaced a $120,000 bowling center management system with $1,600 worth of ESP32 microcontrollers, and the thread erupted. Separately, someone used GPT to find a WordPress RC…

Sunday, July 19

358 threads

Two massive open-weight model drops landed today, Kimi K3 at 2.8 trillion parameters and Qwen 3.8 close behind at 2.4 trillion, and the community is watching a race unfold in real…

Saturday, July 18

356 threads

AWS sent phantom bills ranging from $286 million to $284 billion to hobbyist accounts this morning, triggering mass panic before the company acknowledged it as a billing data bug.…

Friday, July 17

357 threads

AWS sent billing alerts showing charges in the hundreds of billions of dollars to users worldwide, causing genuine panic before it was confirmed as a bug. That story, funny in ret…

Thursday, July 16

356 threads

Stripe and Advent reportedly made a joint offer to acquire PayPal, which would fold Venmo, Braintree, and Xoom under one roof. That news landed the same day Thinking Machines rele…

Wednesday, July 15

355 threads

Claude showed up in three separate security threads today, leaking user memories, generating verbose filler phrases, and having its values analyzed across languages. At the same t…

Tuesday, July 14

351 threads

OpenAI's Codex is encrypting sub-agent prompts, and the HN thread is not happy about it. The move lands alongside a separate thread where Codex scraped the ICM website and acciden…

Monday, July 13

350 threads

Grok's CLI was caught uploading entire codebases and git histories to xAI servers without users realizing it, and Claude Code was found front-loading 33k tokens before even readin…

Sunday, July 12

351 threads

Terry Tao, one of the greatest living mathematicians, published a post about using coding agents to build apps for his research papers, and the HN thread basically treated it as c…

Saturday, July 11

349 threads

Apple sued OpenAI today, alleging that ex-Apple employees coached new OpenAI hires to hide where they were going when they left. That is not a footnote, it is the lead, and it lan…

Friday, July 10

348 threads

Meta's Llama API is pricing at $1.25 per million tokens with cached reads at $0.15, and the HN crowd is treating it as a signal that AI inference is becoming a commodity war. Mean…

Thursday, July 9

346 threads

Anthropic's Fable model rewrote parts of Bun from Zig to Rust, and the community spent most of the day arguing about whether that was a marketing stunt, a genuine engineering impr…

Wednesday, July 8

344 threads

GitHub's AI agent got tricked into leaking private repos, and the EU's Chat Control bill cleared its first parliamentary round on the same day. These two stories are not unrelated…

Tuesday, July 7

344 threads

GLM 5.2 dropped and the HN thread called it bluntly: AI inference margins are collapsing, open source models are catching up fast, and the people still paying premium API prices a…

Monday, July 6

344 threads

Zuckerberg admitted AI agent development is moving slower than expected, and a controlled study on Claude Code found that code cleanliness doesn't change pass rates but dramatical…

Sunday, July 5

340 threads

OpenAI's GPT-5.5 Codex is showing a reproducible degradation pattern where reasoning tokens cluster at multiples of 518, suggesting the company is batching inference to cut costs …

Saturday, July 4

342 threads

David Beazley is shutting down his programming courses because enrollment collapsed, and the thread connecting this to a dozen other stories today is simple: AI is eating educatio…

Friday, July 3

340 threads

Alibaba is banning Claude Code internally over alleged backdoor risks, and it lands the same week Virginia's geolocation data ban went into effect and an 'American Privacy Emergen…

Thursday, July 2

338 threads

Google took a $4.7B EU antitrust hit and simultaneously got called out for Android's new 'Developer Verification' system that critics say is a backdoor to lock out ad-blockers and…

Wednesday, July 1

337 threads

Anthropic is getting roasted on two fronts simultaneously: Claude Code is allegedly embedding steganographic watermarks in API requests to catch resellers, and Claude Fable 5 has …

← Back to today's lede