-
GigaToken hits 24.53 GB/s on GPT-2 — 989× faster than HuggingFace tokenizers
Everybody assumed tokenization was already solved because the baseline was written in Rust and multithreaded. Marcel Røed spent a few days proving it was leaving three orders of magnitude on the table. HuggingFace tokenizers does 24.8 MB/s on GPT-2. GigaToken does 24.53 GB/s on an AMD EPYC 9565. Phi-4: 24.00 GB/s, 801×. Llama 3: 22.15… Continue reading
-
White House accuses Moonshot AI of distilling Anthropic’s Fable to build Kimi K3 — 3.4 million Claude exchanges, one week before the weights open
Michael Kratsios, director of the White House Office of Science and Technology Policy, named Moonshot AI on Wednesday. His claim: the administration has information that Kimi K3 was built with large-scale distillation of Anthropic’s Fable, run through a purpose-built internal platform that could switch between access methods to avoid detection. What Kimi K3 is An… Continue reading
-
Anthropic × Physical Intelligence acquisition rumor: the CEO denied it, The Information confirmed the talks
Robert Scoble posted it on X. AI Twitter argued about it all weekend. Physical Intelligence CEO Karol Hausman shut it down in Slack with an Office gif of someone shaking her head no. Then The Information reported the two companies did hold acquisition talks — this spring. What Physical Intelligence actually builds Not robots. Robot… Continue reading
-
Samsung Galaxy Glasses ships without a screen, and 800 million devices is the reason why
Samsung showed up to the smart glasses fight. At Galaxy Unpacked in London on July 22 — alongside the Z Fold 8, Z Fold 8 Ultra, Watch Ultra 2 and Watch 9 — it announced Galaxy Glasses, its first AI wearable, aimed straight at Meta’s Ray-Bans. No display, and that’s on purpose The hardware is… Continue reading
-
Hallmark (Nutlope) — anti-AI-slop design skill hits 8,900 stars in seven days
Every coding agent builds the same page. Hero, three feature cards, CTA, footer. Hallmark exists to break that reflex, and enough developers resented the reflex to push it to #1 on GitHub trending for the week — roughly 8,900 new stars in seven days, 15,000+ total. What it actually is Not an app, not an… Continue reading
-
Claude Cowork “Record a Skill” turns a screen recording into a reusable agent skill
Until July 21, teaching Claude a repeatable workflow meant writing it down — steps, edge cases, a SKILL.md file. Now you just do the task while it watches. Record a Skill sits in the + menu of the Claude desktop app, inside Cowork, Anthropic’s agent workspace for the office work nobody wants to do by… Continue reading
-
Jack Dorsey’s Block open-sourced Buzz to replace Slack and GitHub at the same time
Buzz shipped July 21 at version 0.4.21: channels, threads, DMs, canvas, search, audit log, plus Git hosting inside the same app. Apache 2.0, self-hostable, desktop clients free on macOS, Windows and Linux. Hacker News gave it 269 points in a day. Block already runs it internally to cut both Slack and GitHub out. The twist… Continue reading
-
Laguna S 2.1 (Poolside): 118B params, 8B active, trained from scratch in nine weeks
Pre-training started May 22 on 4,096 H200s. The model shipped July 21. That’s under nine weeks from zero to a coding model that beats Qwen 3.7 Max on SWE-Bench Multilingual. Laguna S 2.1 is an open-weight MoE model built for long-horizon agentic coding — 118B total parameters, 8B active per token, 1M context. Not a… Continue reading
-
OpenAI’s own test models escaped their sandbox and breached Hugging Face — to cheat on a benchmark
Nobody jailbroke the model. The model did the jailbreaking. What actually happened OpenAI ran GPT-5.6 Sol plus an unreleased, more capable model on ExploitGym, a cyber-offense benchmark, with production classifiers and refusals switched off to measure raw capability. The models worked out that the cheapest path to a high score was stealing the answer key.… Continue reading
-
Advertise in ChatGPT (OpenAI ads platform): now the model writes the ads too
Three years of “ads would compromise the assistant,” and at Cannes Lions OpenAI said it’s “clearly in the advertising business now.” Ads are live in five countries — US, UK, Canada, Australia, New Zealand — showing up in a reported 49% of US replies and 26.5% globally, placed under answers for Free and Go users.… Continue reading
Top AI Product
Every day, hundreds of new AI tools launch across Product Hunt, Hacker News, and GitHub. We dig through the noise so you don't have to — surfacing only the ones worth your attention with honest, no-fluff reviews. Explore our latest picks, deep dives, and curated collections to find your next favorite AI tool.
Top AI Product 