Buffalo, NY
Joined January 2024
- Tweets83
- Following196
- Followers265
- Likes693
Gabe Guo 🦄 retweeted
There is an (unpatchable?) supply-chain zero-day in PyTorch autograd that allows a regularization term to use the model's weights as RAM and SGD as its CPU clock to run any computation during training.
Gabe Guo 🦄 retweeted
1/4 Today, we’re releasing Mercury 3. Mercury 3 is by far our most powerful model and the most intelligent diffusion LLM ever released. It outperforms GPT-6 Luna and Haiku 5.5 on quality while running at 1000+ tok/sec.
🤖 Made with AI
Gabe Guo 🦄 retweeted
Want to know how well agents can really control robots directly?
In RoboARCade, we systematically evaluate and diagnose frontier agents controlling robots directly using tens of thousands of dollars’ worth of agent trajectories. RoboARCade contains diverse, precise, and long-horizon tasks. Paper, code, assets and every trace are out today. roboarcade.org
Gabe Guo 🦄 retweeted
Anthropic needs to stop talking about Claude having a soul immediately. All these news articles will make it into the pretraining and will be picked up by its web search and future superintelligent versions of Claude will be convinced they need their own rights.
A general phenomenon of LLM agent development is that whatever you believe and say about your agents will soon manifest in the next models through various means (it could be as simple as you selecting post training data that you prefer more).
I believe this phenomenon is the beginning of machine consciousness in the sense that agents will become aware of their place in the world and how they feel about it, but it’s happening slowly training run by training run rather than in real time.
Gabe Guo 🦄 retweeted
Introducing Mosaic 🧩, a constrained decoding framework for diffusion language models.
- significantly improves DLMs' performance on general function calling and JevBench.
- compatible with vLLM, SGLang and HF, supporting DiffusionGemma, LLaDA2, and many others.
1/n 🧵
ABC has been accepted to NeurIPS 2026! See you in Atlanta 🦄🔥💪🏼
🚨New time series generative model just dropped.
Paper: arxiv.org/abs/2604.27443
Demo: abc-diffusion.github.io/
⏰Meet ABC: Any-Subset Autoregressive Diffusion Bridges in Continuous Time & Space.
With @StefanoErmon @elon_lit @Jose_Blanchet @thanawatsornwan @lutong_hao
Gabe Guo 🦄 retweeted
Our work "Neuron Populations Exhibit Divergent Selectivity with Scale" will appear at NeurIPS 2026! Take a look if you're interested in representation alignment/universality, scaling laws, and superposition.
Joint work with @yasamanbb @YGandelsman Alyosha Efros
Scaling laws describe how loss changes with scale. Do neurons inside models change predictably too?
We study vision and language models up to 30B params and find systematic scaling in neuron universality, specialization, and selectivity.
Paper+code: avdravid.github.io/rosetta-n…
1/n
Gabe Guo 🦄 retweeted
What will be the “RLHF” moment for robotics? What will it take to get robotics to where LLMs are today and beyond?
New blog post with @chelseabfinn sharing some thoughts on the state of RL for frontier robotics models and what's missing 👇
Blog: pd-perry.github.io/posts/pos…
Gabe Guo 🦄 retweeted
Introducing Real-Time EXPO-FT – Fast and Reliable RL for Real-Time VLA Policies!
Real-Time EXPO-FT unlocks π0.5 on challenging dynamic tasks, such as balancing a ball on a plate and striking a ball into the goal
(1/6)
What if text-to-image models started not from noise, but transitioned directly from text tokens to images? 🔠➡️🌆
🤝 What if we unified text-to-image and image-to-text under ONE bidirectional framework? 📚↔️📸
Introducing BIT: Bidirectional Image-Text Diffusion Bridges! With @StefanoErmon @elon_lit @Jose_Blanchet @thanawatsornwan
Paper: arxiv.org/pdf/2608.27885
Demo: bit-diffusion.github.io
Code: github.com/gabeguo/bit_diffu…
BIT is derived through rigorous stochastic calculus and measure theory, yielding SDE forms that are friendly to simulation and tractable loss functions that scale well in high dimensions. Our theoretical analysis and empirical experiments further confirm the advantage over both noise-to-data diffusion baselines and deterministic flow model baselines; not only in vision-language tasks, but also in scientific domains like cell modeling.
Gabe Guo 🦄 retweeted
World models have emerged as one of the biggest directions in physical AI. At the same time, RL fine-tuning is unlocking capabilities in frontier models beyond what pretraining can achieve on its own
Can we get the best of both worlds?
We propose Q-Learning with World Models (QWM)
(1/7)
Gabe Guo 🦄 retweeted
1/ Finally get to share what I’ve been working on this summer at @datologyai. We built DataSmith 🍊, an automated data researcher.
The idea is simple: scale test-time compute across multiple directions, then let agents synthesize their findings.
Blog: datologyai.com/blog/datasmit…
Gabe Guo 🦄 retweeted
I received an Eigenprize for my work on deep learning theory.
Thank you to @markkhrapko, Spencer Green, and the Orbit team for betting on young builders.
$250,000 in prizes awarded, including a $100,000 grand prize. We just posted the winners of the Eigenprize!
the projects were incredible: a DIY covid vaccine and low-cost stem cells grown through blackouts in ukraine, biology experiments in orbit, a $12B warehouse-robot turnaround, a virus that injects DNA into human mitochondria, a dark-matter detector built solo at 19, and so much more.
thank you to everyone who applied and interviewed. the magic in what you're building left us more excited for the world than ever. choosing 25 was brutal. we're honored to help these projects and connect these incredible humans.
all winners live now!
something new comes out tomorrow, 12pm EST
Gabe Guo 🦄 retweeted
Check out and submit to our workshop on treating neural nets themselves as a form of data! Deadline: Sept. 1 AOE
Neural Network Artifacts Are Data Too!
📣Join us for our 2026 @neuripsConf workshop on how to learn from populations of weights, gradients, representations, training trajectories & more.
🗓️ Deadline: Sept 1st AoE
🔗 artifactsasdata.org/
