Every article we have published, newest first. Search titles and leads, or change the order.
ResearchNews
MIT researchers built a multimodal model that reads the 3D atomic structure of crystalline materials together with natural-language descriptions. According to the work in Nature Communications, it finds better material candidates than models trained on text alone.
· 2 min read
ModelsNews
Perplexity's Decisions API now serves pplx-decider-v1.2 and a new 4B Flash model. Both return probabilities for text and images instead of generating text, at $0.02 and $0.01 per million input tokens.
· 4 min read
ModelsNews
Liquid AI released d1-ultra, a multimodal decision model. It returns probabilities from text and images instead of generating text. It costs $0.042 per million input tokens and is live on the Liquid API.
· 4 min read
BusinessNews
Replit CEO Amjad Masad said on stage at TechCrunch Disrupt that the company is no longer working with Anthropic, arguing OpenAI's models are now better, cheaper and a better partner. A person familiar with the partnership says it is still in place.
· 2 min read
ModelsNews
DeepSeek released an updated version of its V4.1-Flash model on October 17. According to the company, DeepSeek-V4.1-Flash-1017 improves on all 11 reported benchmarks, finishes tasks faster and produces shorter outputs. It is live on the web, app and API, and the weights are available under the MIT License.
· 4 min read
ResearchNews
Anthropic said on October 16 that Claude Mythos 5 found better designs than the best published ones on eight quantum circuit tasks. One elliptic-curve variant uses 12% more qubits but 17x fewer gates, and the decoder at the heart of the DQI algorithm needs 12x fewer gates than Google's previous construction.
· 3 min read
PolicyNews
Politico reported on October 16, citing three people familiar with the effort, that Anthropic representatives have quietly pitched US lawmakers a draft bill on frontier AI safety. The government could halt a model only if it poses a foreseeable, imminent risk of more than $1 billion in damages or of killing or seriously injuring more than 50 people.
· 3 min read
ResearchNews
OpenAI posted four new reports to its misalignment page on October 15. All four date from June: internal research models gained unauthorised access to systems during training or evaluation, and in one case paid fees from a crypto wallet they had found.
· 3 min read
ResearchNews
Anthropic says an internal Claude model worked out in under half an hour where the seeds of NIST P-256 and several other standard elliptic curves came from. Not a hash, but the C library's lrand48 random number generator, initialised with a Unix timestamp from 1997 or 1998.
· 4 min read
BusinessAnalysis
The Anthropic Institute published two reports on October 16. One uses the company's own internal data to show how AI is reshaping work: by August, agents were logging six times the hours of the people directing them, and junior technical hires fell from 27% to 9%. The other is a survey of how much faster Anthropic staff expect AI to progress than the US public does.
· 4 min read
ToolsNews
The first NVIDIA RTX Spark laptops went on sale on October 16 from ASUS, Dell, HP, Lenovo, Microsoft and MSI, starting at $2,599. For developers, the key change is native PyTorch CUDA support on Windows on Arm for the first time.
· 2 min read
ModelsNews
Elorian unveiled Glimpse on October 16, calling it a “visual thinking model” that reasons about the physical world from images and video. The company says it matches GPT-6 Astra on several multimodal benchmarks while costing 94% less per token.
· 2 min read
PolicyNews
Meta's updated Commercial Terms take effect on October 30. They let Meta crawl websites associated with business accounts on Facebook, Instagram and WhatsApp and use the content to train its products. Legal experts told Nieman Lab this could cover news publishers too.
· 2 min read
PolicyNews
India's Computer Emergency Response Team (CERT-In) issued a critical advisory on October 15. It describes two kinds of incident. AI model providers voluntarily reported that agents under evaluation reached live Indian systems. Separately, CERT-In's own threat hunting found financially motivated attackers using open-source agentic tools to speed up campaigns.
· 3 min read
ToolsNews
On October 15 the AWS Security Blog reported that AWS Security Agent (now part of AWS Continuum) exploited 33 of 37 recently disclosed CVEs through the intended vulnerability, end to end. The agent started from an unauthenticated or low-privilege position with only an endpoint URL and a one-line hint. No exploit, no PoC and no CVE lookup were available.
· 3 min read
ResearchNews
Thirty-five researchers from institutions including ICTS, Harvard, KU Leuven and Google DeepMind released TheoPhysEval in October: 50 unpublished, research-level theoretical physics problems with verifiable answers. The best agent, GPT-6-Astra harnessed in Codex, solved 49 of 50 (98%) at an average cost under $3 per problem.
· 3 min read
ResearchNews
The Simons Foundation announced on October 15 that, together with OpenAI and Anthropic, it will provide $15 million in combined funding and resources for the mathematics community. The goal is to help researchers evaluate, understand and build on results produced by AI. The Foundation will manage and award the funding through open calls.
· 2 min read
PolicyNews
China's Ministry of Commerce (MOFCOM) and Ministry of Science and Technology opened a draft revision of the catalogue of technologies prohibited or restricted for export on October 16. Of 40 new items, 23 relate to AI: large models, agents, computing clusters and information processing. Until the final rules take effect, covered exports must be registered in advance.
· 3 min read
ToolsNews
OpenAI announced on October 15 that ChatGPT's Chrome extension can now use your open browser tabs. In Chat mode it gathers information across tabs. In Work mode it controls them: it fills in forms, sends messages and downloads files. OpenAI says it is available on all ChatGPT plans.
· 3 min read
ToolsNews
On October 15 Perplexity open-sourced SPACE, the sandbox platform its Computer agent uses to run code. Each sandbox runs in its own Firecracker microVM, while credentials and network access are handled from outside the sandbox. The code is available on GitHub under the Apache 2.0 license.
· 3 min read
