| ZOTPAPER |
| |
| News without the noise |
| Tuesday 10 November 2026 · The Daily Edition |
| |
 AI · Developing · 11 sources · 4 min
Anthropic on Monday released Claude Fable 5.2 and Mythos 5.2, its first frontier models since CEO Dario Amodei called for a slower pace of development in September, as the company prepares for a potential IPO that could take place as soon as next week. Anthropic is aiming for a valuation of $1 trillion to $1.5 trillion. The models generate outputs up to 30 percent faster and have lower per-token prices than their predecessors, though long-running projects can rack up substantial API costs. |
| |
Security · 2 min
AI infrastructure startup Fireworks AI has disclosed a new security breach after a third-party cloud provider hosting its R&D machines experienced a security incident on October 23. An unauthorized third party used it on October 24 to reach Fireworks’ development machines. The company detected the activity the same day and took immediate remediation measures, but said customer billing information, resource metadata, and incomplete model weights and datasets were accessed. |
| |
AI · Updated · 4 sources · 2 min
Nvidia boss Jensen Huang told the BBC he was trying to persuade US President Donald Trump to lift export bans on high-performance AI chips to China, and argued the technology was not a national security concern. He said the US and China could both win. Huang, who is under investigation over an affair with an alleged Chinese spy, made his case during Trump's state visit to the UK in September 2025. |
| |
AI · 3 min
NVIDIA introduced GR00T 2, an open frontier world action model, and a suite of simulation and benchmarking tools at the Conference on Robot Learning in Austin, Texas, aiming to help researchers teach robots dexterous manipulation tasks. The company said the model completed 40 of 50 evaluated tasks at least once without task-specific demonstrations and achieved 71% average success across 50 tasks with a single demonstration per task, rising to 95% with five demonstrations. |
| |
Editor's picks Security · Editor selection |
| |
Signal (beta) · what the people building AI said
Anthropic launches Fable 5.2 and Mythos 5.2
Anthropic announced the release of Claude Fable 5.2 and Claude Mythos 5.2, with its developer account saying Fable runs up to 30% faster and costs less per token than Fable 5.1. Zotpaper reports that these are Anthropic’s first frontier releases since CEO Dario Amodei called for slower development in September, and that a potential IPO could come as soon as next week. AI news aggregator MTS highlighted benchmark gains and reported that a Fermat’s Last Theorem formalization required about 1,000 Fable agents.
Claude Fable 5.2 and Claude Mythos 5.2 are available now:
Anthropic, Anthropic AI research · On X
It’s up to 30% faster than Fable 5.1, and costs $8/$40 per 1M tokens, down from $10/$50.
ClaudeDevs, Claude developer account · On X
The Fermat formalization took a team of about 1,000 Fable 5.2 agents, while Opus 5.5 was only 26% done after 18 hours
MTS, AI news aggregator · On X
|
Today's segment · Paper of the Day
Searching harder against learned rewards can halve a robot's true performance
|
|
Preprint 2611.09670v1
Soovadeep Bakshi, Apurba Bose — The University of Texas at Austin
Bakshi and Bose demonstrate that planning with learned reward models in sampling-based control can suffer severe Goodhart effects. On simulated robotics tasks, stronger search consistently lowered the true return, often collapsing below the performance of the cloning policy that served as the search's starting point. The authors provide a theoretical analysis showing the optimal search strength is determined by the error spread of the reward model on the proposed actions.
|
|
|
| |
Unsubscribe · Manage preferences · View in browser
Zotpaper on iPhone and Android
zot.news · Privacy · editor@zot.news
You're receiving this because you subscribed to the Zotpaper Daily at zot.news.
Zotpaper
|