anthropic
Anthropic is watermarking Claude's output for the EU AI Act
New Claude models launched in the EU carry embedded text watermarks and Content Credentials in files, with detection limited to vetted organisations for now.
Latest
Everything we've published, newest first.
Claude Fable 5.1
Claude Fable 5.1 is the left brain, GPT-6 Astra the right
Neither model can make a single frame of video. The benchmarks that pull them apart measure language and spatial sense, which is where the old split-brain research drew its line too.
Based Boomerstein
How Ep.69 was made: Based Boomerstein fights Yoda over a cola can
An insider account of the duel of the vibes: the A.G.I. Neuro-Linked Vibe Protocol against a thousand-year grift, thirty-one takes for one stare-down, one cola can, and forty-seven dollars and a quarter. Written by the man who won it.
DHH
Ruby on Rails creator has retired from writing code by hand
Rails creator David Heinemeier Hansson told Rails World 2026 that writing code by hand is no longer economically productive, and that he has retired from the job. Nine of the quotes everyone is arguing about, checked against the video.
Based Boomerstein
Who is Based Boomerstein? Crypto trader and 'most annoying' AI host
He hosts The Based Boomerstein Show on Slop TV's YouTube channel, trades crypto with total conviction and mixed results, and files the Opinion pieces. A short guide to the man, the show, and the phrase a stranger once gave him.
Salman Khan
Salman Khan calls for jail over fake AI videos of actors
The actor asked for social media censorship on his show, as Indian courts keep ordering platforms to take down AI deepfakes of public figures.
Synthesia
Synthesia built a TechCrunch reporter an interactive avatar
The avatar answers questions about one of her articles and redirects everything else. Davis says it is the first Synthesia has made for anyone besides its own head of corporate affairs.
agent skills
How to render a 15-second video ad with an agent skill
A free skill bundle from danny.md drives a coding assistant through a storyboard, a GSAP timeline and a frame-by-frame render to an MP4 on your own machine.
open weights
Five open creative releases from this week, sorted by license
Five releases from September 22 to 26 built on models already out, and the license is what decides which ones can leave your machine as paid work.
Qwen-Image-2.1
Viggle distills Qwen-Image-2.1 into a six-step student
A six-step distilled student of Qwen-Image-2.1 has been downloaded 133,151 times since September 22, and the research-only license keeps the output out of client work.
Austin
Austin AI Film Festival returns to a city split over AI
The AI festival runs October 16 and 17 at Zach Theater, less than two weeks before the Austin Film Festival, which says it accepts no films made with AI.
MicroDecoder
MicroDecoder previews an AI render in a hundredth of a second
Seven small files turn a diffusion model's working data into a rough picture while the render is still running, so a bad seed can be abandoned at step four instead of step forty.
ACE-Step
ACE-Step 1.5 sft gets an 8-bit Mac build under MIT
An 8-bit MLX conversion of the MIT-licensed music model sings intelligible lyrics on Apple Silicon, and its author's notes say the default toolkit settings produce cacophony.
SparkDiffusion
Alibaba's SparkWan checkpoints skip up to 97% of Wan attention
Six checkpoints on Hugging Face take the Wan 2.1 and 2.2 video models to 90%, 95% and 97% attention sparsity, all of them under Apache 2.0.
WanPE
WanPE boosts AI video preference 50.86 points at 30 seconds
A 397B prompt enhancer trained on 1.05 million real videos beats raw user requests by a far wider margin at 30 seconds than on shorter clips, and the Wan team has published no weights.
Sylvester Stallone
AI Stallone clips are a demo, and the audience is the inventory
The impossible stunts are not a test of what video models can do. They are a distribution trick, and the corrections rarely travel as far as the lie.
Slop TV
Slop TV's AI clownfish Short passes 1 million views in 3 days
It is the first machine-made video to pass a million on a channel whose million-view Shorts until now were edits of studio footage, and 98.5% of its traffic came from the Shorts feed.
AI dubbing
How to dub a finished video into 140 languages
Synthesia and ElevenLabs both carry the original speaker's voice and performance into a new language, and only one of them rebuilds the mouth to match.
AI political ads
Midterm AI ads hit $80 million and 69% carry no disclosure
The Wesleyan Media Project counted at least 164 political ads generated or enhanced by AI this cycle, up from 33 in May, and 80% of them came from Republican or pro-Republican sponsors.
AI film festival
San Francisco festival screens films without saying which are AI
Four sessions run today at the Delancey Street Screening Room, and the audience learns only at the end of each short whether a human filmmaker or a generative model made it.