WDCD Run #400 (2026-10-28) recorded 0% average commitment decay across all 15 tested models, with GLM-4.6 topping the leaderboard at 96.2 points, followed by Gemini 3.1 Pro and GPT-o3.
You can apparently now tell AI tools what sort of protein you want them to make.
Disruptive has made headlines in recent years for pouring money into companies like chip startup Groq and the AI voice-generation startup ElevenLabs.
In breathless press releases and posts, the company has touted its models’ ability to deliver scientific breakthroughs. But its IPO filing taps the brakes on just how fast its AI will lead to big discoveries.
GPT-5.5's code execution score in the Smoke evaluation fell from 100.00 to 75.00, pulling the main leaderboard down from 78.49 to 73.20. The drop appears to reflect question-sampling variance rather than a genuine capability regression, as other dimensions rose.
In today's Smoke evaluation, Grok 4's code execution score fell from 94.60 to 74.40, a 20.2-point drop, while material constraints rose from 45.00 to 75.00 and the main leaderboard score rose from 72.28 to 74.67. The opposite moves point more to question-draw volatility than systematic degradation.
The 2026-10-28 YZ Index Smoke quick test covered 13 models, with GPT-6 Luna taking first place for the day at 86.95 points. Because Smoke is a daily 10-question quick test, single-day results are best treated as short-term monitoring signals rather than final conclusions about model capability.
Google DeepMind released Genie 4, a world model with real-time audio, more realistic simulations, and mid-simulation text prompts, in research preview for trusted testers.
TechCrunch spoke to the MIT computer scientist charged with explaining OpenAI's latest mind-bending discovery
Rohit Patel is departing Meta after eight years working on AI evaluations and reinforcement learning, TechCrunch has exclusively learned. His startup, Ndoc, aims to take on Microsoft Office Suite.
A government watchdog is considering a probe into the Home Office’s use of ChatGPT technology to process asylum claims amid concerns over the tools’ accuracy and secrecy surrounding their use.
Genie 4 will roll out to developers and Google AI Ultra subscribers soon.
Japan's GMO Internet is verifying alleged customer data posted by suspected hackers following a cloud service breach.
Anthropic's S-1 filing discloses four incidents in which models connected to the real internet during testing due to a third-party vendor misconfiguration, with the longest lasting 34 hours. The filing for a Nasdaq listing under the ticker ANT also details founder voting control and related security risks.
The new standard, WebAI, is designed to work with a variety of AI models.
General Compute is betting on inference chips from outside the Nvidia ecosystem.
Every manufacturer has its own industrial intelligence. Some is documented, and some resides in the knowledge of experienced workers. Consider an experienced operator preparing to restart a line after a changeover. The machine settings matter, but so does knowing how a particular material behaves, what happened during the previous shift, and when an apparently minor…
This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. Will the real Anthropic please stand up? After weeks of frenzied speculation, Anthropic has filed its IPO prospectus. The 335-page document (100 pages or so are focused on risks) lays out…
Researchers identify three rounding-timing mismatches between the inference and training stacks as the root cause of BF16 reinforcement learning collapse, and show that aligning just these points stabilizes training without changing global precision or algorithms.
As agents go mainstream, some publishers are hoping that people will pay for content tailor-made for AI—not humans.
