DeBoris

Information-Guided Identification of Training Data Imprint in (Proprietary) Large Language Models High-quality training data has proven crucial for developing performant large language models (LLMs). However, commercial LLM providers disclose few, if any, details about the data used for training. ...

Want to know what training data has been memorized by models like GPT-4?

We propose information-guided probes, a method to uncover memorization evidence in *completely black-box* models,

without requiring access to
🙅‍♀️ Model weights
🙅‍♀️ Training data
🙅‍♀️ Token probabilities 🧵 (1/5)

a man with long hair is giving a thumbs up and saying he 's gonna fix the economy . ALT: a man with long hair is giving a thumbs up and saying he 's gonna fix the economy .

This is definitely lowering the cost of eggs

My friends are organizing a new workshop bringing together NLP and CSS research with psychology: First Workshop on Integrating NLP and Psychology to Study Social Interactions (NLPSI) at ICWSM 2025 nlpsi-workshop.github.io Consider submitting a paper (long/short) or an extended abstract :)

Imagine if @aoc @sanders.senate.gov , every dem, SIMULTANEOUSLY held Town Halls where they allowed grant and contract recipients to explain to the country what it is they do and why it's important

Invite all media. Including RW podcasters. @spaces Flood the zone

Call it a Day Of Transparency

“I, too” by Langston Hughes

Alumni too! Call/email your schools and tell them that they have a legal and ethical responsibility to protect student data. I’m still paying OSU and they should be safeguarding my data from third parties.

Is this based on fear from politicians cutting funding or are some of these “scholars” being emboldened to show their true colors?

I firmly believe that there needs to be more conversation about ethical AI development and less about “SkyNet” like scenarios.

A statistical approach to model evaluations A research paper from Anthropic on how to apply statistics to improve language model evaluations

These seem like some very common (statistical) sense recommendations for eval. Still the title of the paper is somewhat confusing as if these ideas are new. I would’ve suggested “Bringing a statistical approach…”

www.anthropic.com/research/sta...

I can already feel my timeline getting smarter and happier

Hello all DeBoris here. I work on ML models in the healthcare space. Fulltime blerd, husband and father.

06 🤙🏿

🤙🏿

AI is biased against speakers of African American English, study finds Large language models attributed negative attributes, less prestigious jobs and more convictions to speakers

news.uchicago.edu/story/ai-bia...

What can we do to make improve NLU and mitigate bias in LLMs towards certain dialects?

#AI #NLP #EthicalAI

This app feels like 2009. I hope it stays that way