Fable 5’s NanoGPT Speedrun used 811 experiments to close one gap
Prime Intellect ran 153 autonomous optimization attempts. Fable 5 led the NanoGPT Speedrun, but the result measures a model-harness-effort system.
Author: Abdessalam Alaoui
I'm Abdessalam Alaoui, founder of Musthave.AI. I write practical guides to AI models, tools, generative media, and ways real people make money with this stuff. No press-release fluff. 20+ years shipping digital products; you'll also find my work around HowTik, Yumasty, and IncomeHelpful.
Thomson Reuters’ Thomson model cost $40M. The outside test is still missing
Thomson Reuters spent $40 million building a legal-focused model. Its first product use is clear, but the public benchmark evidence remains internal.
NVIDIA says AVO hit 100 on ARC-AGI-3. The private set is still the test
NVIDIA says AVO completed all 183 public ARC-AGI-3 levels with 100 RHAE. The architecture is useful evidence; private-set generalization is still untested.
LinkedIn got one million AI-slop reports. One click still cannot judge a post
LinkedIn says its AI-slop feedback passed one million clicks and classified content lost 40% of views. The missing denominator and appeal path still matter.
NVIDIA SkillEvaluator reports a 41-point lift. Repeat the test
NVIDIA's paired agent tests report large gains from verified skills. The catalog average is promising, but most skills received only one attempt per task.
Uber automated-decision fine makes human review operational
Uber faces a €824.99 million Dutch privacy fine over automated driver deactivations. The builder lesson is what meaningful human review must do.
MLflow SSRF is exploited. Version 3.15.0 is the line
CISA says CVE-2026-64849 is being exploited. An MLflow webhook could follow a safe-looking redirect into cloud metadata or internal services.
SpaceX Cursor acquisition closed at $60B. Test your exit before the roadmap changes
Cursor is now a wholly owned SpaceX subsidiary after a $60B stock deal. The filing proves the handoff; builders should now test contracts, exports,…
Anthropic risk report: Claude writes most code. Oversight still broke
Anthropic says Claude writes a large majority of its production code. Its new risk report also documents review, training, permission, and monitoring failures.
Adobe Express AI Assistant picks the model. Your workflow still needs checkpoints
Adobe Express AI Assistant automatically routes design tasks across Firefly and partner models. Here is the versioning workflow that keeps the beta useful.
EVOKE world model moves video memory outside the context window
Alaya-EVOKE stores persistent world state outside its bounded video context. The paper reports near-playback generation, but no public weights yet.
YouTube AI detector flagged original animation, Kurzgesagt says
Kurzgesagt says YouTube privately traced a severe reach decline to a false AI-content classification. Creators need evidence and a human appeal path.