Houjun Liu
@jemoka.com
NLP & POMDPs; CS@Stanford; gradient descent enthusiast
www: jemoka.com
ac: nlp.stanford.edu/~houjun/
www: jemoka.com
ac: nlp.stanford.edu/~houjun/
Reposted by Houjun Liu
meet me at this button friends
November 11, 2025 at 9:01 PM
meet me at this button friends
Reposted by Houjun Liu
LMAO, openreview down point 9pm UCT when ICLR is supposed to be releasing. Coincidence?
November 11, 2025 at 9:01 PM
LMAO, openreview down point 9pm UCT when ICLR is supposed to be releasing. Coincidence?
Reposted by Houjun Liu
Introducing 𝘁𝗵𝗼𝘂𝗴𝗵𝘁𝗯𝘂𝗯𝗯𝗹𝗲𝘀: a *fully unsupervised* LM for input-adaptive parallel latent reasoning
✅ Learn yourself a reasoning model with normal pretraining
✅ Better perplexity compared to fixed thinking tokens
No fancy loss, no chain of thought labels 🚀
✅ Learn yourself a reasoning model with normal pretraining
✅ Better perplexity compared to fixed thinking tokens
No fancy loss, no chain of thought labels 🚀
October 2, 2025 at 3:54 PM
Introducing 𝘁𝗵𝗼𝘂𝗴𝗵𝘁𝗯𝘂𝗯𝗯𝗹𝗲𝘀: a *fully unsupervised* LM for input-adaptive parallel latent reasoning
✅ Learn yourself a reasoning model with normal pretraining
✅ Better perplexity compared to fixed thinking tokens
No fancy loss, no chain of thought labels 🚀
✅ Learn yourself a reasoning model with normal pretraining
✅ Better perplexity compared to fixed thinking tokens
No fancy loss, no chain of thought labels 🚀
Introducing 𝘁𝗵𝗼𝘂𝗴𝗵𝘁𝗯𝘂𝗯𝗯𝗹𝗲𝘀: a *fully unsupervised* LM for input-adaptive parallel latent reasoning
✅ Learn yourself a reasoning model with normal pretraining
✅ Better perplexity compared to fixed thinking tokens
No fancy loss, no chain of thought labels 🚀
✅ Learn yourself a reasoning model with normal pretraining
✅ Better perplexity compared to fixed thinking tokens
No fancy loss, no chain of thought labels 🚀
October 2, 2025 at 3:54 PM
Introducing 𝘁𝗵𝗼𝘂𝗴𝗵𝘁𝗯𝘂𝗯𝗯𝗹𝗲𝘀: a *fully unsupervised* LM for input-adaptive parallel latent reasoning
✅ Learn yourself a reasoning model with normal pretraining
✅ Better perplexity compared to fixed thinking tokens
No fancy loss, no chain of thought labels 🚀
✅ Learn yourself a reasoning model with normal pretraining
✅ Better perplexity compared to fixed thinking tokens
No fancy loss, no chain of thought labels 🚀
Reposted by Houjun Liu
New Paper Day! For EMNLP findings—in LM red-teaming, we show you have to optimize for **both** perplexity and toxicity for high-probability, hard to filter, and natural attacks!
August 20, 2025 at 7:51 PM
New Paper Day! For EMNLP findings—in LM red-teaming, we show you have to optimize for **both** perplexity and toxicity for high-probability, hard to filter, and natural attacks!
New Paper Day! For EMNLP findings—in LM red-teaming, we show you have to optimize for **both** perplexity and toxicity for high-probability, hard to filter, and natural attacks!
August 20, 2025 at 7:51 PM
New Paper Day! For EMNLP findings—in LM red-teaming, we show you have to optimize for **both** perplexity and toxicity for high-probability, hard to filter, and natural attacks!
Reposted by Houjun Liu
Reposted by Houjun Liu
You're not too dumb for Haskell, you just need a reason to practice. :)
June 21, 2025 at 8:17 AM
You're not too dumb for Haskell, you just need a reason to practice. :)
Reposted by Houjun Liu
Just published in JOSS: 'Turftopic: Topic Modelling with Contextual Representations from Sentence Transformers' https://doi.org/10.21105/joss.08183
July 3, 2025 at 1:13 PM
Just published in JOSS: 'Turftopic: Topic Modelling with Contextual Representations from Sentence Transformers' https://doi.org/10.21105/joss.08183
Reposted by Houjun Liu
OCaml @ocaml.org is in The Economist!
Jane Street is the quant shop's quant shop. The goose that lays the golden egg is its tech system, which is built rather unusually https://econ.trib.al/MPdov6Y
Jane Street’s sneaky retention tactic
It involves the use of an obscure, French programming language
econ.trib.al
June 27, 2025 at 7:32 PM
OCaml @ocaml.org is in The Economist!
Reposted by Houjun Liu
We’re proud to announce three new tenure-track assistant professors joining TTIC in Fall 2026: Yossi Gandelsman, Will Merrill, and Nick Tomlin (@nickatomlin.bsky.social). Meet them here: buff.ly/JH1DFtT
June 27, 2025 at 4:29 PM
We’re proud to announce three new tenure-track assistant professors joining TTIC in Fall 2026: Yossi Gandelsman, Will Merrill, and Nick Tomlin (@nickatomlin.bsky.social). Meet them here: buff.ly/JH1DFtT
Reposted by Houjun Liu
New paper on the generalization of Flow Matching www.arxiv.org/abs/2506.03719
🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn *can only generate training points*?
w @quentinbertrand.bsky.social @annegnx.bsky.social @remiemonet.bsky.social 👇👇👇
🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn *can only generate training points*?
w @quentinbertrand.bsky.social @annegnx.bsky.social @remiemonet.bsky.social 👇👇👇
June 18, 2025 at 8:08 AM
New paper on the generalization of Flow Matching www.arxiv.org/abs/2506.03719
🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn *can only generate training points*?
w @quentinbertrand.bsky.social @annegnx.bsky.social @remiemonet.bsky.social 👇👇👇
🤯 Why does flow matching generalize? Did you know that the flow matching target you're trying to learn *can only generate training points*?
w @quentinbertrand.bsky.social @annegnx.bsky.social @remiemonet.bsky.social 👇👇👇
Reposted by Houjun Liu
New Paper Day! For ACL 2025 Findings:
You should **drop dropout** when you are training your LMs AND MLMs!
You should **drop dropout** when you are training your LMs AND MLMs!
June 2, 2025 at 1:22 AM
New Paper Day! For ACL 2025 Findings:
You should **drop dropout** when you are training your LMs AND MLMs!
You should **drop dropout** when you are training your LMs AND MLMs!
New Paper Day! For ACL 2025 Findings:
You should **drop dropout** when you are training your LMs AND MLMs!
You should **drop dropout** when you are training your LMs AND MLMs!
June 2, 2025 at 1:22 AM
New Paper Day! For ACL 2025 Findings:
You should **drop dropout** when you are training your LMs AND MLMs!
You should **drop dropout** when you are training your LMs AND MLMs!
Reposted by Houjun Liu
I'm excited to announce that I’ll be joining the Computer Science department at Johns Hopkins as an Assistant Professor this Fall! I’ll be working on large language models, computational social science, and AI & society—and will be recruiting PhD students. Apply to work with me!
May 30, 2025 at 3:56 PM
I'm excited to announce that I’ll be joining the Computer Science department at Johns Hopkins as an Assistant Professor this Fall! I’ll be working on large language models, computational social science, and AI & society—and will be recruiting PhD students. Apply to work with me!
May 21, 2025 at 4:11 AM
@radbuglet.bsky.social since when did you become a user of bluesky?
May 17, 2025 at 5:52 AM
@radbuglet.bsky.social since when did you become a user of bluesky?
Reposted by Houjun Liu
Jax has a debugger now! This changes everything: docs.jax.dev/en/latest/de...
Compiled prints and breakpoints — JAX documentation
docs.jax.dev
May 5, 2025 at 11:35 PM
Jax has a debugger now! This changes everything: docs.jax.dev/en/latest/de...
Reposted by Houjun Liu
Inspiring celebration of David Attenborough from Kate Winslet and many others - and a much-needed reminder to stand with science www.theguardian.com/tv-and-radio...
Happy birthday, David Attenborough! 99 ways he has inspired us, by Barack Obama, Billie Eilish, Morgan Freeman – and many more
This week the presenter turns 99. To celebrate, we asked 99 nature lovers – including Margaret Atwood, Jane Fonda, Bono, Kate Winslet and Michael Palin – how he has helped us see the world with fresh ...
www.theguardian.com
May 4, 2025 at 2:23 PM
Inspiring celebration of David Attenborough from Kate Winslet and many others - and a much-needed reminder to stand with science www.theguardian.com/tv-and-radio...
Is it just me or is the latest style guide of ChatGPT's IFT is like terribly sarcastic? I don't need 👉 finger guns after every single message.
April 11, 2025 at 6:14 PM
Is it just me or is the latest style guide of ChatGPT's IFT is like terribly sarcastic? I don't need 👉 finger guns after every single message.
I'm to this day still confused about why people are so hyped about softmax-bottlenecked "deep research" approaches; idk about you but I don't usually need to compose a 10 page treatise on first order logic to decide whether or not an if statement is backwards....
March 1, 2025 at 8:58 PM
I'm to this day still confused about why people are so hyped about softmax-bottlenecked "deep research" approaches; idk about you but I don't usually need to compose a 10 page treatise on first order logic to decide whether or not an if statement is backwards....
cs theory seems like an outlier within theoretical sciences where we are living in the era of Euler, Gauss, etc. for math with substantial results being continuously developed live as a part of frontier research. what a time to be alive.
New paper: Simulating Time With Square-Root Space
people.csail.mit.edu/rrw/time-vs-...
It's still hard for me to believe it myself, but I seem to have shown that TIME[t] is contained in SPACE[sqrt{t log t}].
To appear in STOC. Comments are very welcome!
people.csail.mit.edu/rrw/time-vs-...
It's still hard for me to believe it myself, but I seem to have shown that TIME[t] is contained in SPACE[sqrt{t log t}].
To appear in STOC. Comments are very welcome!
people.csail.mit.edu
February 22, 2025 at 8:02 AM
cs theory seems like an outlier within theoretical sciences where we are living in the era of Euler, Gauss, etc. for math with substantial results being continuously developed live as a part of frontier research. what a time to be alive.
the world if I could spell "causal interventions" correctly on the first try
February 14, 2025 at 11:23 PM
the world if I could spell "causal interventions" correctly on the first try
Reposted by Houjun Liu
Ever dreamed of AI agents learning through interacting with the open world unsupervisedly? Our latest preprint introduces NNetNav-Live which collects training data through exploration on real websites and hindsight labeling, which produces a SOTA OSS agent.
February 6, 2025 at 7:22 PM
Ever dreamed of AI agents learning through interacting with the open world unsupervisedly? Our latest preprint introduces NNetNav-Live which collects training data through exploration on real websites and hindsight labeling, which produces a SOTA OSS agent.