Picture

Kevin Lin 林冠言 @nlpkevinl

research @Letta_AI @berkeleynlp @ucbrise @ai2_allennlp kevinlin.io Joined September 2017
  • Tweets
    55
  • Followers
    481
  • Following
    308
  • Likes
    1K
Kevin Lin 林冠言 retweeted
charlespacker Profile Picture

Charles Packer @charlespacker

2 weeks ago

Prior to GPT-5, Sonnet & Opus were the undisputed kings of AI coding. It turns out the GPT-5 is significantly better than Sonnet in one key way: the ability to recover from mistakes. Today we're excited to release our latest research at @Letta_AI on Recovery-Bench, a new…

Kevin Lin 林冠言 retweeted
alexgshaw Profile Picture

Alex Shaw @alexgshaw

a month ago

Congrats to @Letta_AI for building the best performing open source agent on Terminal-Bench!

Letta_AI Profile Picture

Letta @Letta_AI

a month ago

Congrats to @Letta_AI for building the best performing open source agent on Terminal-Bench!

Kevin Lin 林冠言 retweeted
alexgshaw Profile Picture

Alex Shaw @alexgshaw

2 months ago

Evaluating agents on benchmarks is a pain. Each benchmark comes with its own harness, scoring scripts, and environments and integrating can take days. We're introducing the Terminal-Bench dataset registry to solve this problem. Think of it as the npm of agent benchmarks. Now…

Kevin Lin 林冠言 retweeted
realJessyLin Profile Picture

Jessy Lin @realJessyLin

2 months ago

User simulators bridge RL with real-world interaction // jessylin.com/2025/07/10/use… How do we get the RL paradigm to work on tasks beyond math & code? Instead of designing datasets, RL requires designing environments. Given that most non-trivial real-world tasks involve…

Kevin Lin 林冠言 retweeted
sarahwooders Profile Picture

Sarah Wooders @sarahwooders

3 months ago

This has been in the works for a very long time, so very excited to share it with more people! The first and only model agnostic Agent API that is packaged with the beautiful ADE

Letta_AI Profile Picture

Letta @Letta_AI

3 months ago

This has been in the works for a very long time, so very excited to share it with more people! The first and only model agnostic Agent API that is packaged with the beautiful ADE

Kevin Lin 林冠言 retweeted
charlespacker Profile Picture

Charles Packer @charlespacker

3 months ago

introducing the Open AI agent cloud 🚀 unlike OpenAI Assistants / Responses API, Mistral Agents API, etc, you have complete control and can freely move state + memory from local <-> cloud (using agent file!)

Letta_AI Profile Picture

Letta @Letta_AI

3 months ago

introducing the Open AI agent cloud 🚀 unlike OpenAI Assistants / Responses API, Mistral Agents API, etc, you have complete control and can freely move state + memory from local <-> cloud (using agent file!)

Kevin Lin 林冠言 retweeted
yizhongwyz Profile Picture

Yizhong Wang @yizhongwyz

3 months ago

Thrilled to announce that I will be joining @UTAustin @UTCompSci as an assistant professor in fall 2026! I will continue working on language models, data challenges, learning paradigms, & AI for innovation. Looking forward to teaming up with new students & colleagues! 🤠🤘

Kevin Lin 林冠言 retweeted
cursor_ai Profile Picture

Cursor @cursor_ai

3 months ago

A conversation on the optimal reward for coding agents, infinite context models, and real-time RL

Kevin Lin 林冠言 retweeted
Letta_AI Profile Picture

Letta @Letta_AI

3 months ago

🚀 Just launched: The first benchmark for LLMs on agentic memory management! Memory (or context) management is the final bottleneck. As models get smarter, the agents that win will be those that can intelligently manage what they remember, forget, and retrieve.

Kevin Lin 林冠言 retweeted
chai_research Profile Picture

CHAI AI. William Beauchamp. @chai_research

4 months ago

Interesting idea of “sleep-time compute” - Activates deep thinking in idle periods - Reduces response latency - 6.4x fewer tokens per query - 5.2x speedup in response time

Kevin Lin 林冠言 retweeted
Letta_AI Profile Picture

Letta @Letta_AI

4 months ago

Our next Stateful Agents meetup will be hosted in SF by @IndexVentures 5/15! Come by to learn about stateful agents and MCP🔌(Model Context Protocol) - integrating MCP and stateful agents is a powerful way to connect external context and tools to your agents.

Kevin Lin 林冠言 retweeted
Letta_AI Profile Picture

Letta @Letta_AI

5 months ago

MemGPT 2.0 with sleep-time compute is now available in @Letta_AI ! 💤🧠 You can now create agents with “sleep-time” enabled - they are just like normal agents, but have one or more sleep-time agents continuously re-writing their in-context memory, separate from the main process.

Kevin Lin 林冠言 retweeted
jiayi_pirate Profile Picture

Jiayi Pan @jiayi_pirate

5 months ago

We explore a new dimension in scaling reasoning models in Adaptive Parallel Reasoning APR lets LMs learn to orchestrate both serial & parallel compute E2E via supervised training + RL — w/ better efficiency and scalability than long CoT on Countdown 🧵 arxiv.org/abs/2504.15466

Kevin Lin 林冠言 retweeted
Letta_AI Profile Picture

Letta @Letta_AI

5 months ago

We're excited to release our latest paper, “Sleep-time Compute: Beyond Inference Scaling at Test-Time”, a collaboration with @sea_snell from UC Berkeley and @Letta_AI advisors / UC Berkeley faculty Ion Stoica and @profjoeyg letta.com/blog/sleep-tim…

Kevin Lin 林冠言 retweeted
_akhaliq Profile Picture

AK @_akhaliq

5 months ago

Sleep-time Compute Beyond Inference Scaling at Test-time

Kevin Lin 林冠言 retweeted
kalomaze Profile Picture

kalomaze @kalomaze

5 months ago

the most underused, underappreciated, and underrated trick that is basically only used by Anthropic; context distillation

Kevin Lin 林冠言 retweeted
belindazli Profile Picture

Belinda Li @belindazli

6 months ago

Past work has shown that world state is linearly decodable from LMs trained on text and games like Othello. But how do LMs *compute* these states? We investigate state tracking using permutation composition as a model problem, and discover interpretable, controllable procedures🧵

Kevin Lin 林冠言 retweeted
cathychen23 Profile Picture

Catherine Chen @cathychen23

a year ago

How does the human brain represent semantic information from different languages? Our new preprint suggests that bilingual language comprehension relies on shared semantic representations that are systematically modulated by each language! 1/n

biorxivpreprint Profile Picture

bioRxiv @biorxivpreprint

a year ago

How does the human brain represent semantic information from different languages? Our new preprint suggests that bilingual language comprehension relies on shared semantic representations that are systematically modulated by each language! 1/n https://t.co/pSdQaeYaAs

Kevin Lin 林冠言 retweeted
orionweller Profile Picture

Orion Weller @orionweller

a year ago

LLMs can use complex instructions - why can’t retrieval models? We build FollowIR, a training/test set of real-world human retrieval instructions. Our FollowIR-7B is the best IR model for instruct-following, even beating @cohere @OpenAI retrievers 🤯 📝 arxiv.org/abs/2403.15246

Kevin Lin 林冠言 retweeted
FrancesDing Profile Picture

Frances Ding @FrancesDing

2 years ago

Protein language models (pLMs) can give protein sequences likelihood scores, which are commonly used as a proxy for fitness in protein engineering. But what do likelihoods encode? In a new paper (w/ @JacobSteinhardt) we find that pLM likelihoods have a strong species bias! 1/

Trends for United States

Hobi
25,7 B posts
5sos
17,5 B posts
WILDCARD
5.553 posts
Rand Paul
19,5 B posts
Rakai
4.451 posts
Jonas Brothers
17,5 B posts
Dodgers
52,2 B posts
NFT-based
UCLA
8.560 posts
Stanford
10,4 B posts
#HawaiiFB
#GoCougs
1.298 posts
Yamamoto
22,1 B posts
Mateer
12,9 B posts
UNLV
5.262 posts
Sam Houston
Jamo
2.763 posts
GM CT
11 B posts
Underwood
10,6 B posts
Weber State

You might like

@xAI | Prev. 🦙MetaAi; MSFTResearch, allen_ai, GoogleDeepMind; PhD @berkeley_ai
Sheng Shen
@shengs1123
2K Followers
Computer Agents and RL post-training @GoogleDeepMind, prev: Stanford CS PhD student @StanfordNLP
Shikhar
@ShikharMurty
3K Followers
Researcher @AnthropicAI Prev:🦙 @aiatmeta, @allen_ai PhD @uwcse + @uwnlp
Sachin Gururangan
@ssgrn
6K Followers
Assistant prof. @LTIatCMU @SCSatCMU. Working on NLP: language interfaces, applied pragmatics, language-to-code, grounding.
Daniel Fried
@dan_fried
4K Followers
Luke Zettlemoyer
@LukeZettlemoyer
10K Followers
@stanfordnlp PhD student. tweets auto-deleted periodically.
Nelson Liu
@nelsonfliu
4K Followers
Assistant Professor @CSatUSC | Previously Visiting Researcher @facebookai | Stanford CS PhD @StanfordNLP
Robin Jia
@robinomial
4K Followers
safety, robustness, alignment @openai
Eric Wallace
@Eric_Wallace_
10K Followers
Research Scientist at Meta Fundamental AI Research (FAIR)
Scott Yih
@scottyih
2K Followers
Asst prof @UUtah · Ex @allen_ai @uwnlp postdoc @HD_NLP PhD · she/her 🇭🇷
Ana Marasović
@anmarasovic
5K Followers
Assistant Professor at UCSD CSE. NLP, ML, AI. I’m recruiting PhD students.
Lianhui Qin
@Lianhuiq
6K Followers
NLP at Tel-Aviv University and Google DeepMind
Jonathan Berant
@JonathanBerant
3K Followers
Asst Prof @UChicagoCS & @DSI_UChicago, leading Conceptualization Lab https://t.co/BVCT3zdaNV Minting new vocabulary to conceptualize generative models.
Ari Holtzman
@universeinanegg
3K Followers
AI Researcher at Google DeepMind
Omer Levy
@omerlevy_
4K Followers
Building/analyzing NLP and vision models. PhD student @berkeley_ai. Formerly: @allen_ai, @penn
Sanjay Subramanian
@sanjayssub
895 Followers