… Hallucinations have been an ongoing problem for AI models, but OpenAI says its new GPT-5.5 Instant model has “significant improvements in factuality across the board.” The company claims that, based on “internal evaluations,” GPT-5.5 Instant produced “52.5% fewer hallucinated claims” than its Insta… …
Kolter laid out OpenAI’s different safety groups: the safety systems team, which works on guardrails and evaluations; the preparedness team, which deals with OpenAI’s preparedness framework; the alignment team, which helps train models on ways that “align with human values”; the model policy team, … …
… As part of a “research partnership,” DeepMind will “work with an offline version of EVE Online running on a local server to test and evaluate models in a controlled setting,” according to a blog post. …
… In Fable’s system card — a public document AI developers release to explain how a system works — Anthropic said it would handle queries it believed were distillation attempts by altering and degrading the model’s answers directly. …
… As part of its earnings, Sonly also said that PS5 sales dropped by 46 percent year over year following significant price hikes.
… Bloomberg reports the group accessed Mythos by making “an educated guess about the model’s online location,” using information about Anthropic’s other models exposed in the breach of Mercor — a company that makes AI training data — along with access one member had through contract work evaluating A… …
… Meanwhile, support for Google’s SynthID watermarks are coming to more AI content generated by OpenAI, Kakao, and ElevenLabs, though Google doesn’t specify what models or content this will apply to. …
… Separately, XBOW released data suggesting “frontier models have taken a major step forward in vulnerability discovery.” Meanwhile, Microsoft said its multi-model agentic setup, MDASH , was used to discover 16 CVEs in this week’s Patch Tuesday updates and is the leader on the CyberGym security evalu… …
… And it’s eaten up completely by indirect operation costs, like building out more compute and the “ungodly” expense of constantly training the next big model. “As soon as you then add all of the infrastructure that needs to be built for the next generation of model, and you look at how these models … …
… Microsoft used its own AI model to find security vulnerabilities. Microsoft’s MDASH multi-model agent is the leader on the CyberGym security evaluation framework, and the company has used it to discover vulnerabilities in its own products. …