
#154 - Google Gemini, Waymo Collision, Smaug-72B, EU AI Act final text, image watermarks
Last Week in AI
Open Sourcing Training Checkpoints and Evaluation Frameworks
The Allen Institute is releasing 500 training checkpoints for each model every 1000 steps during the training process, enhancing openness. They are also making their Coloma evaluation framework code base available. This move is seen as beneficial for safety, allowing the open source community to scrutinize how the model was evaluated. The initiative may also put pressure on other organizations, like meta, to increase their level of openness.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.