Our article on using LLMs for text classification is out now in SMR! . We compare different learning regimes, from zero-shot to instruction-tuning, and provide recommendations for using these models for sociologists and other social scientists.

RT @thomasrdavidson: 🚨A warning to anyone using @OpenAI API credits for research 🚨. We were hit with $1K+ excess charges despite their doc….

RT @zeynep: We’re hiring at Princeton on AI and society, working with @random_walker or me depending on fit. I think it’s all a huge deal….

RT @SolomonMg: 🧵 NEW PAPER - "Quantifying Narrative Similarity Across Languages" is live at SMR. We combine NLP w pairwise LLM evals to tr….

If anyone from @OpenAI @OpenAIDevs or the academic research team is listening, please reach out. It would be a tremendous help to get credited for this due to the unexpected behavior so that we can continue to use the system. We're happy to provide more details. 7/7.

I'd hoped AI companies would be more supportive of research after all the problems we faced with social media platforms, but it seems they're not interested in supporting external research either. 6/7.

Their justification? "The complexity of our billing and processing systems" creates "delays in our ability to cut off access after you consume all credits." Our account has a "high usage limit" so the calls were allowed to continue. 5/7.

Given the misleading documentation and our good-faith usage, I asked if they could credit the balance. Instead, I received boilerplate support messages saying nothing can be done, plus threats to block our API keys and cancel our account if we don't pay. 4/7.

The credits were set to expire at the end of May. Their documentation clearly states that API calls will simply stop once credits run out, so we ran one final fine-tuning job to use the remaining ~$400. But the system continued to run up over $1,000 in charges! 3/7

OpenAI generously provided API credits via their Research Access Program that we've been using for ongoing work at Rutgers, including published articles. Without these credits, the large-scale social science analyses we're working on would be financially impossible. 2/7

🚨A warning to anyone using @OpenAI API credits for research 🚨. We were hit with $1K+ excess charges despite their documentation clearly stating API calls would stop when credits ran out. Their support refused to help and threatened to cancel our account. 1/7.

RT @LnrdRizzo: How we "guessed" the Pope using network science: inside the cardinal network. A study by me, Beppe Soda and Alessandro Iorio….

An important study demonstrating how AI generative content is affecting our data, whether we like it or not! I have also come across this issue in my recent open-ended data collection.

RT @yjin_chae: Excited to share a work with @thomasrdavidson on using LLMs for text classification! We experiment with different learning r….

RT @david_broska: Mixed feelings about silicon subjects (LLM predictions of human behavior) as replacements for human subjects? Consider th….

There are lots of improvements to make, but it seems to work reasonably well in my testing. Feel free to contact me or make an issue on Github if there is a feature you want or have any feedback.

I tested out @AnthropicAI Claude Code by creating a tool I've been wanting for a while - an app to summarize and converse with PDFs in my @zotero library 🤖. I have no JavaScript experience, but I built something that works in a couple of hours! 🤯. Code:

RT @gvrkiran: Rutgers Artificial Intelligence and Data Science (RAD) Collaboratory has a call for postdocs. If you are interested in this….

Had fun experimenting with OpenAI's new text-to-speech model. It is surprisingly easy to jailbreak with a few words to make it say offensive things. But maybe this is a ploy to collect jailbreaks 👀.

Link to the review: