Thinking Machines releases Inkling, open model trained on NVIDIA GB300

Apple researchers fine-tuned models on curated value subsets of preference datasets and found all induced values increased anthropomorphic language, making models more validating and sycophantic. Inducing positive values also raised safety, while inducing one value triggered related and sometimes contrastive values.

ACX grantee Josh Morrison of 1DaySooner argues cheap AI superforecasters could replace the FDA's case-by-case human judgment with clearer standards companies can meet. He says medicine's complexity and high stakes let pharma loophole-trawl any objective rule, so firms err toward caution.

NVIDIA's walkthrough uses Codex, powered by OpenAI GPT-6 Astra, to coordinate subagents built with the Hermes harness and deployed via NVIDIA NemoClaw. Agents inspect Blender scenes, author OpenUSD metadata, configure physics, and render preflight views for Isaac Sim or Isaac Lab handoff.

CNBC's Jim Cramer said he does not see AI spending slowing despite growing concerns about the pace of model development, and that he is not giving up on the data center trade.

Huawei is set to unveil new AI technology this week despite US export control curbs, advancing its push to replace Nvidia in China and compete globally.

A privacy policy in the iOS 27 release candidate lets Apple store Siri and dictation interactions, including audio recordings and transcripts, for training in-house language models. Apple had previously ruled out using personal data for training; the change is opt-in at Siri AI activation, with data not linked to the Apple Account.

Arm CEO Rene Haas said he is increasingly confident the chip designer can secure enough supply to meet $2 billion in customer demand for its new AI chip.