@stephofx

phd student @mldcmu & NSF grfp fellow | prev: @penn

 Pittsburgh, PA
Joined September 2018
  • Tweets43
  • Following223
  • Followers147
3Media
1/ When LLMs answer questions with retrieved webpages, several plausible answers may appear. What if one appears much more often and fits much of what the question asks, but still isn't the right answer? Excited to share our #EMNLP2026 paper on this! 📄arxiv.org/abs/2610.27114
1
15
1
46
3,321
Stephan Xie retweeted
Daman Arora
@amuseddamanOct 16
1/ RL for improving task success has made LLMs remarkably capable. But what happens when the model fails? Turns out, training for graceful failure can be surprisingly tricky! Very excited to share our new paper: Training Language Models to Fail Gracefully 🧵
23
89
19
479
57,892
Stephan Xie retweeted
Ameet Talwalkar
@atalwalkarOct 15
Excited to introduce Scout 1.0, the model powering Bits Investigation, Datadog’s AI SRE agent for complex troubleshooting and late-night alerts. We post-trained GLM-5.3 Flash to match frontier models at much lower cost on our internal AlertEval benchmark. 🧵
8
27
7
116
8,414
Stephan Xie retweeted
dinos
@din0s_Oct 2
wake up babe the freaking pope got schmidhuber'd
55
316
37
3,897
177,789
Stephan Xie retweeted
Samuel Sokota@ssokotaSep 30
In our Nature paper, we introduce the first superhuman Stratego AI, which we built using general techniques that we developed for RL & test-time compute under imperfect information. 1/N
43
188
29
1,420
287,458
the difference being that you may receive viewers not named astra or opus at submission time
will brown
@willcbSep 25
in 2024 i stopped writing papers and started writing tweets and that was such a good call honestly
1
8
1,474
Stephan Xie retweeted
Tim Dettmers
@Tim_DettmersSep 22
Releasing our runtime dynamic compression framework that achieves 1.5-2.0 bit compression at high quality. This is integrated into the bitsandbytes2 library, which starts as a private beta today. Paper: timdettmers.com/papers/runti… Private beta signup: forms.gle/Pwtp8CVRULdoEnY79
20
48
3
295
31,946
live RL metricwatching might be the next gen of streaming
9
159
Let us know if you’d like to give a talk or sponsor!
Lawrence Jang
@JangLawrenceKSep 11
We are running a weekly Agents + RL + Envs seminar at CMU this semester. We are happy to field external speakers, please DM me if you are interested in giving a talk! We are also looking for sponsors - please DM or email me. You can also reach out to @kohjingyu @PranjalAggarw16 @stephofx @ianwu97!
8
2,630
Stephan Xie retweeted
Lawrence Jang
@JangLawrenceKSep 11
We are running a weekly Agents + RL + Envs seminar at CMU this semester. We are happy to field external speakers, please DM me if you are interested in giving a talk! We are also looking for sponsors - please DM or email me. You can also reach out to @kohjingyu @PranjalAggarw16 @stephofx @ianwu97!
1
10
2
33
5,313
Can an AI agent surface why an ML paper might be hard to reproduce — just by reading it, without running any code? We build ReproRepo, a framework for auditing reproducibility with agents. Across 1,149 recent papers, the best agent surfaced a semantically related, human-reported reproducibility blocker for ~90% of them. 🧵👇
2
9
3
37
9,154
Check out our blog post on ARFBench here!
ML@CMU@mlcmublogApr 27
blog.ml.cmu.edu/2026/04/27/a… How good are AI systems at time-series Q&A? On ARFBench, top models hit ~63% on real incident data. But they miss different things than humans; combine both and accuracy jumps to 87%. Read more in our latest blog post!

Introducing ARFBench: A time series question-answering benchmark based on real incidents

How well do AI systems answer questions about time series data? On ARFBench, the best models reach ~63% accuracy on real incident data. But models and human experts fail on different questions:...

blog.ml.cmu.edu
9
1,132
How well do AI systems (LLMs, VLMs, time series FMs) answer questions about time series data📈? On ARFBench, the best models achieve ~63% accuracy on real incident data. But models and human experts fail in different areas: combining them achieves 87% accuracy. 🧵1/
1
15
1
42
3,986
more replies
9/ This was a very fun and insightful collaboration between Datadog AI Research and collaborators at CMU, including Ben, @MononitoGoswami , @JunhongShen1 , Emaad, Chenghao, David, @ThisIsOthmane , and my advisor @atalwalkar.
1
3
182
10/ Blog post: datadoghq.com/blog/ai/introd… x-listed: blog.ml.cmu.edu/2026/04/27/a… Paper: arxiv.org/pdf/2604.21199 Dataset+Model+Leaderboard: huggingface.co/datasets/Data…

Introducing ARFBench: A time series question-answering benchmark based on real incidents | Datadog

ARFBench is a time series question-answering benchmark built from real Datadog incidents to evaluate how well AI models can reason about anomalies.

datadoghq.com
2
108
Load more