AI
"A few months ago running something this capable locally meant serious hardware and serious tradeoffs on quality..."
Anthropic turns off the compute taps for third-party tools and rushes to get more TPUs online.
NVIDIA dominates results but others find their own wins.
What did we learn from it: Capybaras, undercover mode, and unreleased features...
Similar to "academic and professional research" processes, says Microsoft of defaulting to two-engine research mode.
First debt raise for the company that shares NVIDIA's philosophy: the real money lies in giving away the models for free.
No hyperscaler stone has been left unturned as the AI startup needs immense amounts of cloud compute.
Enterprise users are excluded from the GitHub training data hoover, but will have security agents turned on, with free basic use, for Microsoft 365 E5.
TurboQuant reduces the working memory needed for vector quantisation without sacrificing accuracy, offering a material reduction in inference costs.