Suchir Balaji

6 posts

Suchir Balaji

@suchirbalaji

previously: AI research at OpenAI

San Franciscosuchir.netJoined June 2024

1 Following

578 Followers

Suchir Balaji’s posts

Suchir Balaji

@suchirbalaji

Oct 24

I recently participated in a NYT story about fair use and generative AI, and why I'm skeptical "fair use" would be a plausible defense for a lot of generative AI products. I also wrote a blog post (suchir.net/fair_use.html) about the nitty-gritty details of fair use and why I

Replying to

Here is the article:

Former OpenAI Researcher Says the Company Broke Copyright Law

Replying to

Also, since I see some incorrect speculation: The NYT didn't reach out to me for this article; I reached out to them because I thought I had an interesting perspective, as someone who's been working on these systems since before the current generative AI bubble. None of this is

Replying to

(and thanks

for advice while writing this!)

Replying to

It's nuanced. Generally speaking, it actually is fair use to train models on copyrighted data for research purposes. The problems happen when the models are commercially deployed in a way that competes with their data sources. When I worked on training datasets for GPT-4 in

Replying to

and

(research is explicitly highlighted as an example of "fair use" in section 107: law.cornell.edu/uscode/text/17. The importance of the commerciality of the use is also seen in the first factor)

1.1K