AI - AI Dynamics

→ View original post on X — @goodside

Tuning models on filtered self-generations for smarter AI

By

@goodside

–

15 December 2022 13h53

By tuning the model on its own generations, filtered to those deemed perfect by human evaluators, greater volumes of data could be used, yielding a more intelligent and obedient model.

→ View original post on X — @goodside

Human demonstrations limited early instruction tuning models

By

@goodside

–

15 December 2022 13h53

Early attempts at instruction tuning relied entirely on demonstrations from humans. This made the model easier to prompt, but the approach was limited by the inherent difficulty of manufacturing new humans.

Company built for task now replaced by Twitter usage

By

–

15 December 2022 13h47

Yeah we actually made a whole company that does just that… But now I just use Twitter instead.

Model Evaluations Counting Details Fine Print Clarification

By

–

15 December 2022 13h19

Good point – the fine print states we're actually counting model evals

Model Evaluation Trade-offs: Epochs vs Computational Budget

By

–

15 December 2022 13h17

Sure. But you only get to do 60,000*20 model evals in total either way! (i.e if the 2nd order method needs twice as many evals, you only get 10 epochs)

→ View original post on X — @ai21labs

AI21 Labs partners with Verb on AI-enhanced writing applications

By

@ai21labs

–

15 December 2022 13h16

Our friends at Verb have launched their AI-enhanced writing apps for authors. It's been a pleasure working with their wonderful team! 🥳 https://t.co/ezxpwgIZhH
— AI21 Labs (@AI21Labs) 15 décembre 2022

Our friends at Verb have launched their AI-enhanced writing apps for authors. It's been a pleasure working with their wonderful team!

Random Initialization Achieving 94% Accuracy in Classification

By

–

15 December 2022 13h11

If you can find a random initialization that by coincidence happens to get >94% of 60,000 10-class classification decisions correct then yes, that counts.