Honghua Zhang

72 posts

Honghua Zhang banner
Honghua Zhang

Honghua Zhang

@HonghuaZhang2

prev agentic coding @xAI, Ph.D. from @UCLA StarAI Lab

Katılım Nisan 2020
222 Takip Edilen784 Takipçiler
Honghua Zhang retweetledi
Po-Nien Kung
Po-Nien Kung@P_N_Kung·
🌟 Excited to share Ctrl-R, accepted as an ICML 2026 Spotlight! Specify a target reasoning structure, Ctrl-R can enforce it during rollout, and retain accurate importance-sampling weights for principled policy optimization. paper: arxiv.org/abs/2603.01641 🧵[1/5]
GIF
English
3
23
100
8.8K
Honghua Zhang
Honghua Zhang@HonghuaZhang2·
It’s really been a great pleasure.
English
23
0
48
5.4K
Honghua Zhang retweetledi
Harold
Harold@LiLiunian·
Extremely honored to be part of this incredible team ❤️! Here is one of my favorite samples ; )
Bill Peebles@billpeeb

it's an honor to work with such an incredible group of researchers. massive shoutout to @LiLiunian, @DmytroOk , avi, @ezhang887 , @ryw999 @troyluhman and eric, eric mintun, @bram_wallace @JureZbontar @mmmbchang @gabriel1 @dangengdg and @cmikeh2 for their incredible work on sora 2. the sora research team is very special.

English
66
26
784
905.8K
Honghua Zhang
Honghua Zhang@HonghuaZhang2·
It's been a great pleasure contributing to grok-code-fast-1 as part of this amazing team. It's a super fast yet super capable agentic model for your daily coding needs. Try it out in your favorite IDEs and let us know what to improve on!
xAI@xai

Introducing Grok Code Fast 1, a speedy and economical reasoning model that excels at agentic coding. Now available for free on GitHub Copilot, Cursor, Cline, Kilo Code, Roo Code, opencode, and Windsurf. x.ai/news/grok-code…

English
1
0
25
2.7K
Honghua Zhang retweetledi
Thang Luong
Thang Luong@lmthang·
Yes, there is an official marking guideline from the IMO organizers which is not available externally. Without the evaluation based on that guideline, no medal claim can be made. With one point deducted, it is a Silver, not Gold.
Mikhail Samin@Mihonarium

🚨 According to a friend, the IMO asked AI companies not to steal the spotlight from kids and to wait a week after the closing ceremony to announce results. OpenAI announced the results BEFORE the closing ceremony. According to a Coordinator on Problem 6, the one problem OpenAI couldn't solve, "the general sense of the IMO Jury and Coordinators is that it was rude and inappropriate" for OpenAI to do this. OpenAI wasn't one of the AI companies that cooperated with the IMO on testing their models, so unlike the likely upcoming Google DeepMind results, we can't even be sure OpenAI's "gold medal" is legit. Still, the IMO organizers directly asked OpenAI not to announce their results immediately after the olympiad. Sadly, OpenAI desires hype and clout a lot more than it cares about letting these incredibly smart kids celebrate their achievement, and so they announced the results yesterday.

English
14
55
568
122.6K
Honghua Zhang retweetledi
Rota
Rota@pli_cachete·
Terence Tao on the supposed Gold from OpenAI at IMO
Rota tweet mediaRota tweet mediaRota tweet media
English
91
526
6.1K
622.4K
Honghua Zhang retweetledi
Meihua Dang
Meihua Dang@meihuadang·
#CVPR2025 "Personalized Preference Fine-tuning of Diffusion Models". We extend DPO to align text-to-image diffusion models with individual user preferences. At test time, it generalizes to unseen users from just few-shot examples — moving toward pluralistic alignment.
Meihua Dang tweet media
English
3
55
378
42.8K
Honghua Zhang retweetledi
Igor Babuschkin
Igor Babuschkin@ibab·
Yesterday we discovered a bug that caused many people to get answers from Grok 2 instead of Grok 3 even though it said Grok 3 in the app 😬 Please give Grok 3 another chance if you found it to be underwhelming the last time. Now available for free.
xAI@xai

This is it: The world’s smartest AI, Grok 3, now available for free (until our servers melt). Try Grok 3 now: x.com/i/grok X Premium+ and SuperGrok users will have increased access to Grok 3, in addition to early access to advanced features like Voice Mode

English
192
220
4.3K
547K
Honghua Zhang retweetledi
Zihang Dai
Zihang Dai@ZihangDai·
Let's go!!!!!
Arena.ai@arena

BREAKING: @xAI early version of Grok-3 (codename "chocolate") is now #1 in Arena! 🏆 Grok-3 is: - First-ever model to break 1400 score! - #1 across all categories, a milestone that keeps getting harder to achieve Huge congratulations to @xAI on this milestone! View thread 🧵 for more insights into Grok-3's performance after ~8K votes in the Arena.

English
26
30
243
37.7K
Honghua Zhang
Honghua Zhang@HonghuaZhang2·
Ctrl-G helps with writing #NeurIPS2024 announcements! Interactive text editing tool based on Ctrl-G released at billkunghappy.github.io/Ctrl-G/ @P_N_Kung
Honghua Zhang@HonghuaZhang2

Proposing Ctrl-G, a neurosymbolic framework that enables arbitrary LLMs to follow logical constraints (length control, infilling …) with 100% guarantees. Ctrl-G beats GPT4 on the task of text editing by >30% higher satisfaction rate in human eval. arxiv.org/abs/2406.13892

English
1
4
16
3.1K
Honghua Zhang
Honghua Zhang@HonghuaZhang2·
So excited to present Ctrl-G **Adaptable Logical Control for Large Language Models** TODAY at #NeurIPS2024 West Ballroom 4:30 - 7:30 pm. Ctrl-G is THE solution to LLM fill-in-the-middle generation, numerical planning and structured output. Stop by to discuss more!
Honghua Zhang@HonghuaZhang2

Proposing Ctrl-G, a neurosymbolic framework that enables arbitrary LLMs to follow logical constraints (length control, infilling …) with 100% guarantees. Ctrl-G beats GPT4 on the task of text editing by >30% higher satisfaction rate in human eval. arxiv.org/abs/2406.13892

English
0
2
17
1.7K
Honghua Zhang
Honghua Zhang@HonghuaZhang2·
It has been a great pleasure working with (and playing poker with) @zhezeng0908 for the last 6 years! She is not only a math genius but also a really considerate person. Apply to her Ph.D. and it will be a fun journey!
Zhe Zeng@zhezeng0908

📢 I’m recruiting PhD students @CS_UVA for Fall 2025! 🎯 Neurosymbolic AI, probabilistic ML, trustworthiness, AI for science. See my website for more details: zzeng.me 📬 If you're interested, apply and mention my name in your application: engineering.virginia.edu/department/com…

English
0
0
2
457
Honghua Zhang
Honghua Zhang@HonghuaZhang2·
@far__away__ Roughly speaking, if we remove our HMM then our method degenerates to the ones you mentioned.
English
0
0
0
141
Honghua Zhang
Honghua Zhang@HonghuaZhang2·
Proposing Ctrl-G, a neurosymbolic framework that enables arbitrary LLMs to follow logical constraints (length control, infilling …) with 100% guarantees. Ctrl-G beats GPT4 on the task of text editing by >30% higher satisfaction rate in human eval. arxiv.org/abs/2406.13892
GIF
English
20
98
601
109K
Honghua Zhang
Honghua Zhang@HonghuaZhang2·
@far__away__ We did not perform a comprehensive comparison since constraint-based decoding (e.g., Guidance-AI and outlines) are not suitable for tasks like text-infilling and keyphrase inclusion: they do not have any probabilistic information to solve these tasks. See paper Sec 3.3 for more.
English
0
0
0
147
Honghua Zhang
Honghua Zhang@HonghuaZhang2·
We hope that Ctrl-G could supplement a new paradigm for querying LLMs in addition to prompting. In the long term, by leveraging tractable generative models like HMMs, we aim to lift the limitations of LLMs due to their intractability for marginalization.
English
1
1
26
2.1K
Honghua Zhang
Honghua Zhang@HonghuaZhang2·
@atlantis__labs @teknium The point of Ctrl-G is not about replacing prompting but improving it. In fact we show that we can combine both prompting and Crtl-G to achieve both soft control (e.g., style) and logical control. Stay tuned for the upcoming user interface for text editing where you can do both.
English
1
0
0
42

Keşfet

@veggie_eric @ChiragCX @P_N_Kung @zhezeng0908 @far__away__ @elonmusk @BarackObama @taylorswift13