Keep pulling the thread on Stephanie Palazzolo.
OpenAI has reportedly discovered a new optimization technique that can cut its inference requirements in half for existing models.
AWS announced it will invest $1 billion to create a new unit staffed with forward-deployed engineers to help customers set up and use AI tools.
The Information reports that Anthropic plans to bring a version of its Claude Tag agent to Microsoft Teams.
The U.S. government intervened in a lawsuit to shut down the gas turbines at SpaceX's Colossus data center, claiming the facility is a matter of national security.
The U.S. Department of Commerce has lifted export controls on Anthropic's Fable-5 model, allowing the company to begin redeploying it.
Beginning July 1st, Anthropic's Fable-5 model was made available again to all global users across all paid subscriptions.
For the relaunch of Fable-5, Anthropic has trained a new classifier designed to block the specific behavior described in the Amazon report, with a claimed success rate of 99%.
Using a new optimization technique, OpenAI was able to serve its entire user base of non-signed-in ChatGPT users on just 100 GPUs.
Stephanie Palazzolo of The Information stated that OpenAI's new cost-cutting technique is a closely guarded secret, not even shared with all OpenAI employees, to prevent it from being copied by other labs.
Deepseek has open-sourced a speculative decoder system called DeSpark that can speed up inference by 85% during testing on small models.
AI aggregator Andrew Curran predicted that a team spun out of OpenAI will soon announce a significant breakthrough in AI architecture related to memory efficiency.
Harry Stebbings of 20 Minute VC reported that five founders, from 10-person startups to a $200 billion public company, have cut their inference spend by 75% or more with little effort.