PhD student at the University of Maryland
Joined November 2023
- Tweets205
- Following165
- Followers673
- Likes633
Pinned Tweet
Humans don’t maintain exact, line-by-line recall of huge contexts like full codebases or long legal documents. We keep a high-level mental model, then look things up when precision matters. We enable LLMs to do this, with high speed.
Sean McLeish retweeted
Driving video shot through a windshield is full of cabin reflections, and 3D Gaussian Splatting bakes them straight into the scene. RIDE-Splat removes them, with no clean images, no reflection masks and no pretrained networks. 🧵
Sean McLeish retweeted
Rumor has it frontier labs are using looped LMs to scale quality without scaling parameters. But KV-cache memory still grows with every loop. Is that necessary?
No! Our method BeNice! unlocks near-lossless sharing of the final KV-cache across loops: more depth, same KV memory.
This lossless KV-sharing isn’t guaranteed out of the box with looped models. 👀👇for our method BeNice! to teach your model that sharing is caring!
Sean McLeish retweeted
This next paper is a bit quirky, but here goes:
Introspection is hard to study in both humans and LLMs. But, we do have internal senses that are verifiable, like proprioception (the sense of where your arm is right now).
What would proprioception look like in language models?
As a digital analogue to the sense of position, a model could verbalize which expert activated in a certain layer for any previous token.
It turns out that LLMs can learn a range of such proprioception skills where they understand and report on their internals. We can calibrate and score these reports, and evaluate whether training for proprioception changes model behaviors.
Sean McLeish retweeted
🚨 New Paper! 🚨
Multiple-choice benchmarks use accuracy (boring), but education has way cooler scoring schemes! 🤩
👎 Lose points if wrong
🔁 Try until you're right
🃏 Wager points
At #EMNLP2026, we show these schemes shift LLM ranks, align with LLM Arena, + unmask LLM quirks
Sean McLeish retweeted
Want to know who's the most cited person in ICLR 2027, and which submissions cite your paper?
I downloaded all 43,013 public ICLR 2027 submissions through the OpenReview API and asked my good friends Codex and Claude to build a site that ranks the most-cited papers and authors.
#1 paper is not Attention Is All You Need 👀
Link: modal-labs-leon-dev--iclr202…
Sean McLeish retweeted
🚨 LLMs are trained on past data and deployed in the future, but we evaluate them like time doesn't exist. By treating LLM evaluation as a forecasting problem, we can better predict how models will actually perform in the future. 📈 1/n
Sean McLeish retweeted
Announcing Exploration-Distillation, a RL framework that unlocks aggressive exploration and large performance boosts over standard algorithms like DAPO. In principle, an LLM can discover new things during RL, but existing algorithms struggle. Here’s how we do it. 🧵1/n
Sean McLeish retweeted
As autoresearch agents run longer, do they keep exploring new approaches, or mostly refine the ones they found early?
Turns out, runs settle into different "idea basins" and stay there.
We introduce fork-and-flush, a simple way out 🧵👇
Sean McLeish retweeted
I'll be in San Francisco all week for @COLM_conf and to talk about investing with @RockboundCap. Let me know if you want to chat about the strategies used in high net worth investing and how they can work for you.
Sean McLeish retweeted
"Sphere Encoder 2"
This paper fixes why one-step autoencoder generation looks blurry.
It trains the latent rotation all the way from an encoded image to the 90° equator region where random generation actually starts.
Then instead of forcing those highly ambiguous latents to reconstruct the exact source image pixel-by-pixel, it uses semantic, latent-consistency, and feature-space score-matching losses so the decoder can generate a sharp plausible image rather than average many possibilities into blur.
On ImageNet, it reaches FDr6 4.92 with just 4 steps and ~23x less sampling compute than PixNerd, though its gFID is worse.
alphaxiv.org/abs/2610.02208
Sean McLeish retweeted
🔁 𝐖𝐡𝐚𝐭 𝐢𝐟 𝐥𝐨𝐨𝐩𝐞𝐝 𝐋𝐌𝐬 𝐜𝐨𝐮𝐥𝐝 𝐮𝐬𝐞 𝟑× 𝐟𝐞𝐰𝐞𝐫 𝐩𝐚𝐫𝐚𝐦𝐞𝐭𝐞𝐫𝐬 𝐚𝐧𝐝 𝐚 𝟑× 𝐬𝐦𝐚𝐥𝐥𝐞𝐫 𝐊𝐕 𝐜𝐚𝐜𝐡𝐞 𝐚𝐧𝐝 𝐬𝐭𝐢𝐥𝐥 𝐛𝐞 𝐜𝐨𝐦𝐩𝐚𝐫𝐚𝐛𝐥𝐞 𝐭𝐨 𝐚 𝐬𝐭𝐚𝐧𝐝𝐚𝐫𝐝 𝐓𝐫𝐚𝐧𝐬𝐟𝐨𝐫𝐦𝐞𝐫?
Introducing 𝐑𝐞𝐭𝐡𝐢𝐧𝐤𝐢𝐧𝐠 𝐚𝐭 𝐅𝐢𝐱𝐞𝐝 𝐏𝐨𝐢𝐧𝐭𝐬. We argue that looped LMs don't have to pay for every loop. They can scale by adding FLOPs at constant memory.
The key is the fixed point of the loop. We introduce:
– 𝐅𝐢𝐱𝐞𝐝-𝐏𝐨𝐢𝐧𝐭 𝐒𝐡𝐨𝐫𝐭𝐜𝐮𝐭𝐬: what fixed points buy a looped model in training (pre-training, post-training) and inference (prefill, decoding)
– A 𝐥𝐞𝐚𝐫𝐧𝐞𝐝 𝐝𝐞𝐩𝐭𝐡 𝐩𝐫𝐢𝐨𝐫 and 𝐨𝐫𝐭𝐡𝐨𝐠𝐨𝐧𝐚𝐥 𝐢𝐧𝐣𝐞𝐜𝐭𝐢𝐨𝐧: two ways to shape those fixed points
🧵 Long thread ahead (0/n).
Sean McLeish retweeted
Frontier LLMs like Claude don’t come with uncertainty estimates, and their verbalized confidence is bad. We trained Pinocchio, a lightweight model that assigns confidence to outputs of popular API models, making it fast and easy to get well-calibrated uncertainty estimates. 📄🧵
Sean McLeish retweeted
Can reasoning models learn to be more token-efficient without ever being trained for efficiency?
Length-penalty RL rewards shorter reasoning. Early stopping cuts reasoning at inference.
Our approach does neither: the only thing we train is confidence prediction. 🧵
Sean McLeish retweeted
🚨 New Paper! 🚨
Your agent is writing code faster than you can understand it 😬
In our controlled online study, coding agents helped users finish tasks quickly, but reduced how well they understood the code they just “wrote”
We also find three ways to avoid this tradeoff 👇🧵
Sean McLeish retweeted
1/9 🧵 What if agents should learn to predict what they will never generate?
ActObs supervises actions AND environment observations during SFT, teaching action consequences with no extra data, parameters, sequence tokens, or forward passes.
📄 arxiv.org/abs/2609.20715
Sean McLeish retweeted
Though this challenge was impossible I learned a lot trying to attempt it anyway. For example, I learned that backprop is capable of factoring small semiprimes. This is contrary to the intent of the problem!
Over 15,000 submissions to One Layer Deeper. Nobody solved the problem as intended, but there are some interesting ideas for learning reusable operations and composing them into deeper computations
coreauto.com/blog/what-the-m…
Sean McLeish retweeted
Deep learning is still not solved.
It has been great to see the creativity of the community and we all have more work to do to understand the universe.
Thank you for being in this journey together ❤️
Over 15,000 submissions to One Layer Deeper. Nobody solved the problem as intended, but there are some interesting ideas for learning reusable operations and composing them into deeper computations
coreauto.com/blog/what-the-m…
Sean McLeish retweeted
Over 15,000 submissions to One Layer Deeper. Nobody solved the problem as intended, but there are some interesting ideas for learning reusable operations and composing them into deeper computations
coreauto.com/blog/what-the-m…
Sean McLeish retweeted
1/🚨 New paper #emnlp2026
Do multilingual #LLMs work in English? It depends on how you ask. In "Lingua Franca or Probing Artifact?", we probe the same hidden states with three latent language identification methods and get different answers. 🧵👇
#interpretability #NLProc
Sean McLeish retweeted
Excited to share two papers at #EMNLP2026! 🥳
1️⃣ Check the Scoreboard: We extend *scoring schemes* from education research and validate their benefits for LLM evals (Main) 🧑🏫📈
2️⃣ VibeJam: Our system for human-AI studies with coding agents (Demo) 🧑💻🌐
See you in Budapest! 🤠
ALT emnlp :)