JHU CLSP (@jhuclsp) header
JHU CLSP (@jhuclsp) profile picture

Center for Language and Speech Processing at @JohnsHopkins #NLProc #MachineLearning #AI https://nitter.cf/t.co/6IXR5OSQtw @jhuclsp@sigmoid.social

Baltimore, MD
Joined June 2016
Photo by JHU CLSP (@jhuclsp)Photo by JHU CLSP (@jhuclsp)Photo by JHU CLSP (@jhuclsp)Photo by JHU CLSP (@jhuclsp)Photo by JHU CLSP (@jhuclsp)Photo by JHU CLSP (@jhuclsp)Photo by JHU CLSP (@jhuclsp)Photo by JHU CLSP (@jhuclsp)Photo by JHU CLSP (@jhuclsp)Photo by JHU CLSP (@jhuclsp)
A beautiful sunny day here on @HopkinsEngineer campus! ☀️
1
1
31
9,041
JHU CLSP retweeted
🤖 AI is now writing its own research papers, resolving decades-old Erdős problems, and even producing a claimed proof of Navier–Stokes. But we only ever see the final results, never the ideation behind them. How does an LLM actually search its way to a final idea, and where does it get stuck during ideation? We turn this into a game of Twenty Questions: an LLM asks yes/no questions to rediscover the core idea of a published paper. Introducing 𝗥𝗲-𝗦𝗲𝗮𝗿𝗰𝗵, which makes every step of ideation in LLM autoresearch observable and measurable. Paper: arxiv.org/abs/2611.14487
12
20
3
137
7,516
What information capabilities do #AI systems need to help us, informavores, navigate diverse, multicultural narratives in our current & future information ecosystems? Do current models support them? No. Introducing NarrativeBench. isle-dev.github.io/Narrative… #NeurIPS #NLProc #HCI
3
16
2
45
2,741
JHU CLSP retweeted
Excited to share what we've been working on since this summer- started in my internship at @MSFTResearch! Modern DNNs predict neural activity surprisingly well, but are black boxes. We introduce interpretable weight-space programming (iWSP) and ask: can LLM agents build predictive models of the brain directly in code, with no training or pretrained weights, so every feature stays transparent? 🧵 (1/6)
11
75
4
417
26,808
📢 Applying to a PhD program in Speech and Language Processing? Johns Hopkins CLSP's Applicant Support Program offers Fall 2026 applicants a one-time review of their statements and CV by current PhD students. 🗓️ Sign up by November 20, 2026! Learn more: clsp.jhu.edu/apply-for-phd/
6
36
1,759
🚨 CISS 2027 Call for Papers! Submit your original research by November 30 for the 60th Annual Conference on Information Science and Systems, hosted by Johns Hopkins ECE. 📅 March 24–26, 2027 | Baltimore Submit: ciss.jhu.edu #CISS2027
3
553
My second LessWrong piece is out - a curated set (w/ commentary) of @allTheYud and @So8res posts. LLMs think of Yud/Soares as dismissive of welfare concerns while they actually take the welfare of current models seriously, alongside p(doom). Link below 👇
4
6
79
3,379
If you’re applying to JHU’s CS Ph.D. program and don’t have access to application support, we’d be happy to help! Please submit your application below.
✨ Interested in applying to Johns Hopkins’ Computer Science Ph.D. program? ✨ JHU CLSP is offering a student-run application mentoring program for those seeking additional feedback on their application materials. ...
6
24
124
6,171
An agent says “task complete.” How did it get there? Did it test its changes, recover from errors, or revise its plan after human feedback? We will be presenting our work on Act*ONOMY, a taxonomy for understanding and analyzing AI agent behavior, at NeurIPS 2026 in Sydney next month. Developed through Grounded Theory, Act*ONOMY provides a shared vocabulary of 10 actions, 46 subactions, and 120 behavior categories. It links behavior labels to exact quotes in agent traces, helping researchers compare agents, inspect what agent does during runtime, and investigate potential failure patterns. Explore the resources on our website(actonomy-agents.github.io/): 🔎 A living and extensible taxonomy that can be expanded as AI agents evolve. 🧩 Annotated trajectories and reports. Follow agent runs and a human–agent coding session turn by turn, explore behavioral profiles, and inspect the evidence behind each annotation. 🛠️ Automated skills. Analyze your own agent traces and human–agent interactions, discover behaviors in new papers, and maintain or extend the taxonomy. 📄 Paper: arxiv.org/abs/2605.13625 🤗 Hugging Face dataset: huggingface.co/datasets/Soph… 💻 GitHub (code, datasets, and extension protocol): github.com/gaojie058/Act-ono… We’re continuing to update the repository and taxonomy, and welcome feedback on the tools and codebook. Many thanks to all my coauthors for their contributions to this work! @KaiserWhoLearns @JentseHuang @ktvank @SijieCesium @YuanCosette @shi_weiyan Zhuoran Lu @ZiangXiao @DanielKhashabi @mdredze If you will be at NeurIPS in Sydney, I’d love to chat about how you study agent behavior and hear about your work!
11
25
3
136
8,973
JHU CLSP retweeted
Need different model weights? Just change them 🙃 Our new paper VFold does exactly that to enable low-overhead cross-layer value cache compression: half the value cache, >98% of full-cache performance 🧵 📄 arxiv.org/abs/2610.12338
3
23
2
50
2,870
JHU CLSP retweeted
✨ Interested in applying to Johns Hopkins’ Computer Science Ph.D. program? ✨ JHU CLSP is offering a student-run application mentoring program for those seeking additional feedback on their application materials. ...
1
28
4
135
15,807
Claude refuses to make an exact copy of The Landlord's Game, a game made in 1904 which has been in the public domain for decades.
202
292
49
5,943
214,649
Astra with 90% on ARC-AGI-1 - superhuman no CoT perf!
This replicates in 10 high-signal tasks from the Think Fast task suite: 6.1 Terra clearly beats the no-CoT performance of GPT-6 Sol in all of them. Seems likely that Terra is also looped.
4
1
21
1,708
GPT-6.1 Terra has drastically enhanced no-COT capabilities
4
1
1
43
4,823
Excited to share that I have reached 500 citations on Google Scholar! 🌟 This milestone doesn't just belong to me, but to all my amazing co-authors as well, without whom it would not have been possible.
4
4
58
2,894
Happy Halloween from Johns Hopkins Robotics! Check out this cool demo of humanoid crawling created by my amazing colleagues
For Halloween, @geetachandra21 got our Unitree G1 to crawl like a spider! Give Geeta (co-advised w/ @HaiminHu) a follow if you're interested in more humanoid demos. We're doubling down on whole-body locomanipulation research at @HopkinsDSAI @HopkinsEngineer @JHUCompSci -- reach out if interested in working with us! (credit to Addison Johnson, who helped design the mounts for the hands, and filmed this; and to Geeta who took a full-stack approach to ensuring that no humans or humanoids were harmed during the production of this video! and to @jloganolson for last year's humanoid crawling deo) sciphylab.org/
1
21
4
128
8,629
JHU CLSP retweeted
Foundation models are trained in stages: pretraining, midtraining, and post-training. But is this separation fundamental? We introduce unified training: a single algorithm for the full pipeline. This gives ⚙️ Simpler infrastructure 🔬 A framework for better science (1/n)
53
258
51
1,600
199,897
Happy that Omnilingual MT got an Outstanding Paper Award at #EMNLP2026! It's hopefully a step towards making omnilinguality normal rather than outstanding.
4
6
53
1,952
🎤✨ Our second keynote is LIVE! We’re excited to have Prof. Berrak Sisman from @jhuclsp take the stage! 🚀 If you haven’t joined us yet, now’s the perfect time to tune in. Don’t miss out on this exciting session! 🌟
We’re live! 🎙️ The SALMA Workshop (@SALMAworkshop26) is now underway at #EMNLP2026 (@emnlpmeeting) in Room F9–F10. Stop by for an incredible slate of keynote speakers and cutting-edge papers on speech and audio language models!
1
1
5
326
Your agent solved the task at xhigh. Would low have solved it too? You can’t tell without rerunning it, and you still pay for every token. We tested models x reasoning effort level x benchmarks. On tasks low effort can already solve, top effort costs up to 42× more #LLM #AI 🧵👇
16
36
13
192
81,418
✨ Interested in applying to Johns Hopkins’ Computer Science Ph.D. program? ✨ JHU CLSP is offering a student-run application mentoring program for those seeking additional feedback on their application materials. ...
1
28
4
135
15,807
Students who may not otherwise have access to application support are especially encouraged to apply. Apply here: forms.gle/W2pXmman47CBoNZp9 Applications are due November 20. We hope to see yours!
2
14
1,515