Eric Florenzano / is on Nostr

Building assistance! Used to make AR/VR. Before that made mobile apps. Before that made web things. People first, then products, systems, data. he/him

PixelFed: https://pixelfed.social/ericflo
Twitter: https://twitter.com/ericflo
Instagram: https://www.instagram.com/ericflo/
Website: https://www.eflorenzano.com/

Last Notes

2024-08-15 17:11:02

- reply

They're all doing it. Rejection sampling is pretty much the first thing you do before moving forward with RLHF-style contrastive learning like PPO/DPO/KTO/RLOO etc. You do it when you want to get better at some task(s), so you want more training data to elicit that behavior and then demonstrate it, and use rejection to move the distribution in the direction you want

2024-05-12 00:30:06

- reply

This will not stop at registration and you well know that. Did you read the bill? It requires you to be able to pull the weights from all computers. How can you and people be so credulous

2024-05-12 00:25:16

- reply

I think I'm done here, we disagree I guess. What a bummer

2024-05-12 00:21:30

- reply

It will chill development of open models and slow development of all models.

2024-05-12 00:09:26

- reply

Because I disagree with your premise. How people use AI is the point of contention: to propogate misinformation, replace jobs, produce nonconsentual imagery, etc. Regulate that stuff, not math.

2024-05-11 23:56:23

- reply

It regulates development instead of use. I don't want companies to have to get governmental approval to open source new models, which this mandates. It's so bad.

2024-05-11 23:39:43

So, how can we band together to stop SB-1047? Or is it a fait accompli?