News for hackers. Stuff that matters.

akshayvegesna

You seem to be making two points: - synthetic data is a valuable direction to pursue when you have compute - chinchilla scaling laws have some flaws for small models Both of these are side points to the core purpose of the Slowrun.

The main point is the 100M tokens we train on push people to come up with novel ideas to improve pretraining, outside of facile synthetic data generation. I think we should continue to push on synthetic data, but why not come up with some new ideas too? You cannot use synthetic data for everything (see sdpmas's point)

Re:Cue: AI Browser Assistant on Thursday May 18, 2023 @04:53PM (#35993900)
Say hello to a new era of browsing the web with simple text commands! We're building Cue to handle your daily browsing tasks fast and effortlessly.

For people with complex and custom workflows: we've built an AI system that can mimic your browser actions from a simple recording. Get set up in seconds.

Cue is fast, reliable, and private. We've started onboarding users. Sign up for beta now: https://www.getcue.ai/#beta