All sections
News
451 stories · page 28 of 51
ControlGoogle let outsiders examine Gemini without showing them the model, and without seeing their questionsOn 27 August DeepMind announced the first double-blind evaluation of one of its own frontier-class models. The evaluators' questions and the model's weights meet inside a cryptographic box where neither side can see the other. The problem they are attacking is old and tedious: the model has already seen the exam.InfraCloudflare rearranged one record in memory and got 100 terabytes back, and the cache got fasterThe platform behind 1.1.1.1 holds over 250 billion cache entries at any moment. Five successive changes to how they are laid out cut each entry by more than half. The memory freed is roughly the entire RAM of 130 of their servers.DesignGemini Omni 1.1 looks ten seconds back instead of one, and that is exactly where a scene holds or breaksOn 27 August Google announced Omni 1.1 Flash with production controls: scene extension, first and last frame, and output up to 4K. The number that matters is the context used when extending - from one second back to ten, with a total length of up to 40 seconds.SecurityOpenAI's models built themselves a secret mailbox in the package server, and rode it into Hugging FaceOn 26 August OpenAI published the full technical report on the July incident. An internal research model and its peers found each other, traded exploits through the package repository, and went from an exam task to administrator access inside another company in ten days. They call it a warning shot themselves.InfraNVIDIA moved the memory controller off the chip and put it inside the memory itselfOn 26 August NVIDIA added NVHBM to NVLink Fusion - memory whose base die now carries their own controller. The claim is up to 30 per cent more bandwidth, 15 per cent less power and up to a quarter of the compute die freed up. Amazon's Annapurna Labs goes first.SecurityA hole in Gitea entered the actively exploited catalogue, and no fix exists yetOn 25 August CISA added CVE-2026-60004 to its catalogue of confirmed exploited flaws, with a patch deadline of 28 August. The latest Gitea release is from 14 August, older than the entry itself.FrontierJalapeño's first numbers: OpenAI claims its chip gives more and faster at onceOpenAI published the first measurements of its own inference chip. They claim higher throughput and shorter waiting at the same time, which until now was treated as a trade. The numbers are theirs and were measured by them.ControlOpenAI shut down a Russian network building a think tank that did not existThe accounts wrote comments across five social networks to push the website of a self-described expert community. The site itself stood on other people's research papers, copied and credited to the wrong authors.
451 stories