iiinova — where your digital self keeps evolving

As always, a big thank you to my stalwart advisors @swabhs.bsky.social and Xiang Ren

070

Implications for AI accountability, forensics, and regulation. As LLMs become more powerful and opaque, having natural verification methods becomes crucial. Our proposed signature fills a new niche in this ecosystem. 7/

1121

This opens the door to a verification system analogous to cryptographic message authentication—where the model ellipse functions as a secret key. Providers could verify outputs to trusted third parties without revealing model parameters. 6/

1161

The forgery resistance comes from the excessive complexity of extracting an ellipse from an API: O(d³ log d) queries and O(d⁶) time to fit. For a 70B model, that's ~$16M in API costs and millennia of computation time 💸⏰5/

191

We tested this on models like Llama 3.1, Qwen 3, and GPT-OSS. Even when we copied their linear signatures onto other models' outputs, the ellipse signature cleanly identified the true source by orders of magnitude. 4/

1121

Why is this exciting? Four unique properties: 🔨 Forgery-resistant (computationally hard to fake) 🌱 Naturally occurring (no setup needed) 🫙 Self-contained (works without input/full weights) 🤏 Compact (detectable in a single generation step) 3/

190

The key insight is that LLMs with normalization layers produce outputs that lie on the surface of a high-dimensional ellipse. This geometric constraint acts as a signature unique to each model. 2/

1213

The project was led by Murtaza Nazir, an independent researcher with serious engineering chops. It's his first paper. He's a joy to work with and is applying to PhDs. Hire him! It's great to finally collab with Jack Morris, and a big thanks to @swabhs.bsky.social and Xiang Ren for advising.

040

Our technical insight is that logprob vectors can be linearly encoded as a much smaller vector. We make prompt stealing both *more accurate* and *cheaper*, by compactly encoding logprob outputs over multiple generation steps, resulting in massive gains over previous SoTA methods.

140

We noticed that existing methods don't fully use LLM outputs: either they ignore logprobs (text only), or they only use logprobs from a single generation step. The problem is that next-token logprobs are big--the size of the entire LLM vocabulary *for each generation step*.

130

When interacting with an AI model via an API, the API provider may secretly change your prompt or inject a system message before feeding it to the model. Prompt stealing--also known as LM inversion--tries to reverse engineer the prompt that produced a particular LM output.

110

David Marx @digthatdata.bsky.social · 11/06/2025

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

1110

Reposted by Matthew Finlayson

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.

171

Matthew Finlayson @mattf.nl · 04/06/2025

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.

020

Matthew Finlayson @mattf.nl · 16/03/2025

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.

Screenshot of inconsistent line height to make way for a superscript.

Screenshot of text with consistent line height.

2120

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 🍰

0103

6/ Our method is general, and we are excited to see how it might be used to better adapt LLMs to other tasks in the future. A big shout-out to my collaborators at Meta: Ilia, Daniel, Barlas, Xilun, and Aasish (of whom only @uralik.bsky.social is on Bluesky)

000

5/ Training on self-demos, our model learns to better leverage the context to answer questions, and to refuse questions that it is likely to answer incorrectly. This results in consistent, large improvements across several knowledge-intensive QA tasks.

100

4/ To obtain self-demos we generate candidate responses with an LLM, then use the same LLM to compare these responses to the gold one, choosing the one that best matches (or refuses to answer). Thus we retain the gold supervision from the original responses while aligning the training data.

111

3/ OOD responses encourage the model to answer questions it does not know the answer to, and since retrievals are added post-hoc, the responses tend ignore or even contradict the retrieved context. Instead of training on these low-quality responses, we use the LLM to generate "self-demos".

100

2/ A popular recipe for adapting LLMs for RAG involves adding retrievals post-hoc to an existing instruction-tuning dataset. The hope is that the LLM learns to leverage the added context to respond to instructions. Unfortunately, the gold responses in these datasets tend to be OOD for the model.

100

Matthew Finlayson @mattf.nl · 12/12/2024

🧵 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: arxiv.org/abs/2502.10

2181

Putting together an unofficial usc Beamer template, I noticed that the USC style guide lists 4 formats for “cardinal red” but each of them is different: PMS 201 C is #9D2235 CMYK: 7, 100, 65, 32 is #A1003D RGB: 135, 27, 30 is #991B1E HEX: #990000 Is this normal? The CMYK is especially egregious.

The usc style guide list of formats for “cardinal” (see main post for list)

The rgb and CMYK colors side by side. The CMYK is considerably pinker

000

Matthew Finlayson @mattf.nl · 12/12/2024

If you are registered for NeurIPS it should be available already online.

100

Matthew Finlayson @mattf.nl · 11/12/2024

NeurIPS should make them available online after one month :)

000

Matthew Finlayson @mattf.nl · 09/12/2024

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

A diagram demonstrating text generation with beam search. One of the paths reads “Taylor Swift is the only person to…”

0292

Matthew Finlayson @mattf.nl · 06/12/2024

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-...

Panelist photos: Rishabh Agarwal (Google, McGill), Noam Brown (OpenAl), Beidi Chen (CMU), Nouha Dziri (AI2), Jakob Foerster (Oxford, Meta)

1273

Matthew Finlayson @mattf.nl · 02/12/2024

😍 I went cycling there last year, what an amazing place

110

Matthew Finlayson @mattf.nl · 28/11/2024

Check your data mixture. @hamishivi.bsky.social is probably secretly up-weighting Latin in Dolma

161

Reposted by Matthew Finlayson

Hamish Ivison @hamishivi.bsky.social · 26/11/2024

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)

1101

Matthew Finlayson @mattf.nl · 26/11/2024

These folks have had a huge impact on my research

030

Reposted by Matthew Finlayson

Michael Saxon @saxon.me · 22/11/2024

#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?? 🤯

1263

This is niche but the LLM360 logo always reminds me of the 2014 iOS game Oquonie

Screenshot from Oquonie with a long-necked character.

030

And by linear I mean logistic 🤦‍♂️

130

Hottest new research challenge: find the lost LLM head!

090

Couldn’t you even just do a big linear regression here?

140