Matt Gardner

@nlpmattg

MTS at @MicrosoftAI. Formerly at @ScaledCognition, Semantic Machines, @allen_ai (@ai2_allennlp, #nlphighlights).

Irvine, CA Joined May 2017

119 Following 8.9K Followers 3K Posts

Latest posts 5

Pinned post

Matt Gardner @nlpmattg Dec 1, 2025

New blog post - Prompt Trees: Training-time Prefix Caching. By the research team at @scaledcognition.

TL;DR: Training speedups of up to 70x on tree-structured data. Not 70%. _70x_.

scaledcognition.com/blog/prompt-tr…
(preprint version coming soon)

www.scaledcognition.com Prompt Trees: Training-time Prefix Caching | Scaled Cognition We present a method for encoding tree-structured data (like you might get from conversation rollouts while doing RL training for conversational agents) using a standard transformer. This method gives...

Matt Gardner @nlpmattg Dec 2, 2025

I'm headed to NeurIPS now, excited to see people!

Also, ICYMI, we just published a way to represent RL rollouts as trees and encode them with a single pass of a standard transformer, giving 70x speedups to post-training: x.com/nlpmattg/statu…

Matt Gardner @nlpmattg Dec 1, 2025

New blog post - Prompt Trees: Training-time Prefix Caching. By the research team at @scaledcognition.

TL;DR: Training speedups of up to 70x on tree-structured data. Not 70%. _70x_.

scaledcognition.com/blog/prompt-tr…
(preprint version coming soon)
View quoted post on X

Matt Gardner @nlpmattg Oct 22, 2025

We read this paper in our reading group today. It's a cool paper; it spurred some interesting discussion. Our high-level reaction: it sure looks like the results can be explained by sampling k tokens at a time, taking the most likely sequence, then continuing.

Aayush Karan Verified account @aakaran31 Oct 17, 2025

We found a new way to get language models to reason. 🤯

No RL, no training, no verifiers, no prompting. ❌

With better sampling, base models can achieve single-shot reasoning on par with (or better than!) GRPO while avoiding its characteristic loss in generation diversity.

View quoted post on X

Matt Gardner @nlpmattg Oct 16, 2025

Some pretty interesting research has gone into training the models behind this. It's been fun to work on this the last ~2.5 years, and I'm excited for what's coming. Also, we're hiring ;).

Scaled Cognition @ScaledCognition Oct 16, 2025

@ScaledCognition + @Genesys = a new era of action-driven AI for CX.

Together, we’re helping enterprises deploy deterministic systems that deliver reliable, policy-aligned outcomes — built for action, not just words.

Learn more → gsys.cx/3J7hl9U
View quoted post on X

Matt Gardner @nlpmattg Mar 30, 2023

This is so sad. Traveling to Slovenia as a chaperone for NACLO with Drago and @LoriLevinPgh was one of my favorite experiences ever. RIP Drago, you will be missed.

Harlan Krumholz @hmkyale Mar 30, 2023

The #AI community, the #computerscience community, the @YaleSEAS community, and humanity have suddenly lost a remarkable person, @dragomir_radev - kind and brilliant, devoted to his family and friends... gone too soon. A sad day @Yale @YINSedge @YaleCompsci #NLP2023

View quoted post on X