@roydanroy
@Google DeepMind. On leave, Canada CIFAR AI Chair and Former Research Director, @VectorInst. Professor, @UofT (Statistics/CS). Views are my own.
AI
Fort says he’s joining Cohere as VP of AI Security to lead research on AI cybersecurity.
AI
@j_foerst says a conversation with Schmidhuber explored shaping the future, rather than simply predicting it.
AI
Human-guided AI helped verify specific Signal protocol protections in four weeks, under assumptions that include perfect cryptography.
AI
The person who left JMLR argues top AI models now give better reviews than time-pressed humans.
AI
A fixed output shift helped later in training, but @mufan_li disputes the authors’ explanation.
AI
The authors say previous results reached either the optimal approximation scale or sample complexity, but not both.
AI
A user linked to an arXiv paper they described as fixing the issue; another user praised the work.
AI
A reply cites the UK and Mistral; the original poster clarifies they meant the EU.
AI
A post introducing the approach says its two modes combine latent-space thinking with text-based actions in one multi-turn flow.
AI
Yoshua Bengio says the letter asks AI and security communities to collaborate to protect critical infrastructure and digital systems.
AI
The author proposes 'fork-and-flush' as a way out of those persistent idea basins.
AI
OpenAI says it drew on an independent mathematics advisory group’s advice about how to release the results.
AI
Reflection says Beam uses 23B active parameters and plans to publish its full weights later in October 2026.
AI
One post asks how that scenario would differ from what happened during COVID.
AI
arXiv says the change aims to distribute moderator time fairly and support staff, volunteers, readers and authors.
AI
Bloomberg, citing people with direct access to the effort, reports that Google’s forthcoming model does less well when employees put it to work than it does on industry benchmarks.
AI
Google announces a 1 million-token output limit, up from 64,000, for its new model. Initial access goes to trusted cyber defenders through the Fairwind Program.
AI
A post about the project claims a 65% score gain with the same compute on a 24-hour multi-repository agent-swarm task.
AI
A user is asking for more votes on how one year in the ‘AI age’ compares with ‘normal times.’
AI
Inherent Labs announced the letter on September 29, saying UK AI companies that have raised $5 billion support its call for people to be free to start and scale businesses.