@thom_wolf - AI Dynamics

Training Data Transparency in Large Language Models Debate

By

–

11 May 2023 14h27

Interesting. You think it’s better to hide the training data then? Currently in NLP the only chatbots widely deployed (and replacing search) are closed source so the situation is maybe a bit different. I’ll try to take some time write a blog post on this topic one day. It’s a

→ View original post on X — @thom_wolf

11 May 2023

Closed source models debate in AI development

By

@thom_wolf

–

11 May 2023 14h16

Closed source models for instance?

→ View original post on X — @thom_wolf

11 May 2023

Credit vs Monetary Reward in Open Source Communities

By

@thom_wolf

–

11 May 2023 13h24

I make a distinction between crediting & monetary reward. I know many people happy to share without monetary reward (the OSS community is all about that) but I know very few people who don't care at all about being credited (even anonymously) for something they've spent time on

→ View original post on X — @thom_wolf

11 May 2023

Need for More Research Beyond Bing Chat

By

@thom_wolf

–

11 May 2023 12h57

Huge need for more research here imo. Bing chat is not nearly enough.

→ View original post on X — @thom_wolf

11 May 2023

AI Content Reposting Threatens Creator Attribution and Artistic Value

By

@thom_wolf

–

11 May 2023 12h57

There is no excitation in seeing your blog post ideas reposted by a chat AI as his own personal thoughts No excitation in sharing art if the main way to consume it becomes through a white page anonymous art drawing chatbot

→ View original post on X — @thom_wolf

11 May 2023

Knowledge Creator Attribution Crisis in Generative AI Models

By

@thom_wolf

–

11 May 2023 12h57

The disconnection between knowledge creators & knowledge consumers in Gen AI models is a pressing problem that's not yet discussed enough Nobody will be excited to share knowledge on the internet if the default way to consume it becomes through anonymizing/unreferencing chatbots

→ View original post on X — @thom_wolf

11 May 2023

Search Engines vs Anonymous AI Training: Creator Economics Shift

By

@thom_wolf

–

11 May 2023 12h57

Search engines had the huge advantage that they would point to your personal page where you could get credit/revenue for your ideas, build your own personal brand Writing anonymous tokens for a held-out training dataset is the dream of no stackoverflow or wikipedia contributor

→ View original post on X — @thom_wolf

11 May 2023

StarCoder Fine-tuned as Chat Assistant Using Community Datasets

By

@thom_wolf

–

10 May 2023 13h30

So by popular demand, we've starting exploring the finetuning of BigCode's recently released code-completion model StarCoder (
https://
x.com/BigCodeProject
/status/1654174941976068119
…) to be a…. Chat Assistant We finetuned it on two high-quality datasets created by the community:
– OpenAssistant’s

→ View original post on X — @thom_wolf

10 May 2023