Matthew Finlayson
@mattf.nl
NLP PhD @ USC Previously at AI2, Harvard mattf1n.github.io
We discovered that language models leave a natural "signature" on their API outputs that's extremely hard to fake. Here's how it works 🔍 📄 arxiv.org/abs/2510.14086 1/
Every Language Model Has a Forgery-Resistant Signature
The ubiquity of closed-weight language models with public-facing APIs has generated interest in forensic methods, both for extracting hidden model details (e.g., parameters) and for identifying...
arxiv.org
I didn't believe when I first saw, but: We trained a prompt stealing model that gets >3x SoTA accuracy. The secret is representing LLM outputs *correctly* 🚲 Demo/blog: mattf1n.github.io/pils 📄: arxiv.org/abs/2506.17090 🤖: huggingface.co/dill-lab/pi... 🧑💻: github.com/dill-lab/PILS
I wish the ML community would stop trying to turn every technique into a brand name. Just give the thing a descriptive name and call it what it is. Forced backronyms like this are counter productive.
David Marx@digthatdata.bsky.social · last yr.too distracted by this to read the actual content
It appears that the only fonts with optical sizes that work with pdflatex are the computer/latin modern fonts. I would kill for a free pdflatex-compatible Times clone with optical sizes so my small text can look good in ArXiv/conference submissions.
If you are writing a paper for #colm2025 and LaTeX keeps increasing your line height to accommodate things like superscripts, consider using $\smash{2^d}$, but beware of character overlaps.
This project was made feasible by the excellent open-source LLM training library @fairseq2.bsky.social; I highly recommend giving it a look! It made both SFT and DPO a piece of cake 🍰
Matthew Finlayson@mattf.nl · last yr.🧵 Adapting your LLM for new tasks is dangerous! A bad training set degrades models by encouraging hallucinations and other misbehavior. Our paper remedies this for RAG training by replacing gold responses with self-generated demonstrations. Check it out here: https://arxiv.org/abs/2502.10
🧵 Adapting your LLM for new tasks is dangerous! A bad training set degrades models by encouraging hallucinations and other misbehavior. Our paper remedies this for RAG training by replacing gold responses with self-generated demonstrations. Check it out here: https://arxiv.org/abs/2502.10
In Vancouver for NeurIPS but don't have Taylor Swift tickets? You can still spend the day going through our tutorial reading list: cmu-l3.github.io/neurips2024-... Tuesday December 10, 1:30-4:00pm @ West Exhibition Hall C, NeurIPS
Curious about all this inference-time scaling hype? Attend our NeurIPS tutorial: Beyond Decoding: Meta-Generation Algorithms for LLMs (Tue. 1:30)! We have a top-notch panelist lineup. Our website: cmu-l3.github.io/neurips2024-...
What's that? A fully open LM competitive with Gemma and Qwen*? Happy to have helped a bit with this release (Tulu 3 recipe used here)! OLMo-2 13B actually beats Tulu 3 8B on these evals, making it a SOTA fully open LM!!! (*on the benchmarks we looked at, see tweet for more)
Ai2@ai2.bsky.social · 2y agoMeet OLMo 2, the best fully open language model to date, including a family of 7B and 13B models trained up to 5T tokens. OLMo 2 outperforms other fully open models and competes with open-weight models like Llama 3.1 8B — As always, we released our data, code, recipes and more 🎁
#socalnlp is the biggest it's ever been in 2024! We have 3 poster sessions up from 2! How many years until it's a two-day event?? 🤯
This is niche but the LLM360 logo always reminds me of the 2014 iOS game Oquonie
Hottest new research challenge: find the lost LLM head!
Hamish Ivison@hamishivi.bsky.social · 2y agoWe actually have all the weights.... except the LM head for this blursed checkpoint. Don't ask.
Everyone follow Sean! He's been working nonstop to perfect our upcoming NeurIPS tutorial
@wellecks.bsky.social · 2y agoI was honored to give a talk at Simons Institute on inference-time algorithms and meta-generation! simons.berkeley.edu/talks/sean-w... It was a sneak-preview subset of our NeurIPS tutorial: cmu-l3.github.io/neurips2024-...
As "X is all you need" and "Transformers are Y" paper titles have died, I propose that we similarly retire: - X of thought - Chain of Y - "[topic] a comprehensive survey" where [topic] is exclusively post-2022 papers - Claims of reasoning/world model/planning based on private defns of one
Today I learned you can add a citation link to your GitHub repo. citation-file-format.github.io
citation-file-format.github.io
I’m proud of this tikz drawing I made today for our upcoming NeurIPS tutorial on decoding (our paper: arxiv.org/abs/2406.16838)
I'll be presenting "Distributional Properties of Subword Regularization" with @zouharvi.bsky.social and Naoaki Okazaki at #EMNLP. arxiv.org/abs/2408.11443 The idea is that stochastic variants of BPE/MaxMatch produce very biased tokenization distributions, which is probably bad for modeling. #NLP
USC NLP folks are on Bluesky! Follow my amazing colleagues here go.bsky.app/KUwSZ6W
A starter pack for #NLP #NLProc researchers! 🎉 go.bsky.app/SngwGeS
How many times did the 14 LA Metro board members + CEO use their Metro TAP cards in the 4.5 years from 1/1/20 to 6/30/24? 2,259 times. 33.5 taps per person per year. 1112 workdays in period. If CEO tapped 2x/workday, 14 board members in total tapped 35 times in 4.5 years.
Just landed in Philly for COLM where I’ll be presenting my work on extracting secrets from LLM APIs at the Wednesday afternoon poster sesh. Please reach out if you wanna hang and talk about sneaky LLM API hacks, accountability, and the geometry of LLM representations! arxiv.org/abs/2403.09539
Logits of API-Protected LLMs Leak Proprietary Information
The commercialization of large language models (LLMs) has led to the common practice of high-level API-only access to proprietary models. In this work, we show that even with a conservative assumption...
arxiv.org
Ooh we got the answer from Steven Bird! Sorry to link to Twitter, but the story is worth reading. x.com/StevenBird/s... "I got the idea at SIGMOD'99... I had been living in West Africa and experienced inequities of access... I wanted ours to be a society that made its content free to all."
Maria Antoniak@mariaa.bsky.social · 2y agoWhat is the story behind the ACL Anthology? How did NLP escape issues around closed access, high publication costs, and governance by external societies/publishers? I guess there must be some interesting history and heroes to thank for this situation!
“Media companies aren't on their workers' side in just the same way that Amazon wasn't on publishers' side. The promise of a new, broader copyright [for AI] is like the promise of DRM: it's a prison pretending to be a fortress” pluralistic.net/2024/03/13/h...
Please help me, probabilists of the internet: what is the probability that an unfair m-sided dice rolls its most-likely value more times than any of its other values after n rolls?
