PhD student, NSF/NRT BIOE Fellow @ UCSB ECE that can still dunk 🏀. Building BisQue Ultra — a distributed agentic system that runs real science.
Santa Barbara, CA
Joined July 2026
- Tweets64
- Following14
- Followers6
- Likes34
I’m assuming this means Astra 6.1 is delayed until further notice.
How we think about securing frontier RL training runs: openai.com/index/towards-saf…
Our paper is ACCEPTED to NeurIPS, crazy to say that. My first first-author paper is a NeurIPS paper.
Astra is so much better to read than Sol. It was so hard to make it through a full response. Fable 5.1 feels two generations behind after using Astra.
Apple should drop a Mac Studio that can run these >1T parameter models. There’s no reason for them to develop their own model, they are perfectly positioned in the AI landscape to continue doing what they do best—hardware.
Lakers new owners, Grok 4.6, DeepSeek V4 PRO is out, Qwen 3.7 MAX 2.4T, and we might get another reset today from Tibo. If Astra drops this week, I’ll finish my PhD in 32 minutes using everything.
BisQue Ultra is running DeepSeek V4 Flash 0731, our platform is in Research Preview for students and faculty at research institutions. I spent way too much time engineering the backend using my almost decade of experience in distributed systems.
ALT BisQue Ultra Performance serving V4 Flash 0731, 13B active / 284B total, 786,432-token window.
It might be more expedient to pay NeurIPS and other top conference reviewers. If you reviewed, you get 50% off registration. If the AC rates you a good reviewer (good feedback, discussion, responsive), the reviewer should get paid, idk $200.
As soon as Qwen 3.8 27B drops on HuggingFace, I am downloading it. Then, immediately pushing it to production.
The training signal, not the harness nor the prompt, is what will differentiate these frontier models (>1T params). 95B activated is insane.
📢Meet Qwen3.8-Max — our most capable model to date.
Next week, the open weights of Qwen3.8-Max will be released, and Qwen3.8-27B is also going open-weights to meet you all!🎉
Qwen3.8-Max, a new bar for coding and cowork at 2.4T parameters:
- Autonomous coding: 10+ days of self-evolving development, from empty folder to production without hand-holding, complete project trace in the GitHub：github.com/qwen-code-dev-bot…
- Real work, real results: Production-quality deliverables across hundreds of professions.
- Long-horizon mastery: System-level autonomous planning with closed-loop adaptive learning, driving 500+ turns of chip design optimization and 365 days of e-commerce strategy.
- Native multimodal intelligence: Vision isn't just input — it's a continuous feedback loop for planning, execution, and self-correction.
đź’°Pricing:
Input: $2.0 / M tokens
Output: $6.0 / M tokens
Implicit Caching: $0.25 / M tokens
Start building with Qwen3.8-Max! 🚀
đź“– Blog: qwen.ai/blog?id=qwen3.8
✅ Qwen Studio: chat.qwen.ai/?models=qwen3.8…
⚡ API: qwencloud.com/models/qwen3.8…
I was at the Stanford AI Summit this week, let’s connect if we didn’t get a chance to talk! Maybe next year I’ll be one of the investors.
Amil retweeted
An internal version of Astra, @OpenAI’s next major model family, solved 10 major open problems in mathematics, quantum complexity, and theoretical computer science.
We believe it will be a major step for scientific reasoning. openai.com/index/ten-advance…
Spent some time building out the Lens Viewer on Ultra for Time Series data. I tested the viewer on the Human Connectome Project (HCP) dataset. Input is a .dtseries.nii and the sample here is rfMRI_REST1_LR_Atlas_hp2000_clean.dtseries.nii
BisQue Ultra can detect small objects (Prairie dogs and burrows) from aerial imagery. This project is with our collaborators at the Smithsonian Institute. Model developed at UCSB Vision Research Lab, and BisQue Ultra orchestrates.
I spent way too much time on my NeurIPS rebuttal. What happens if the reviewers don't followup? Has anyone graduated without a NeurIPS paper?
One of the main reasons I started this account is every year that I teach, I have many amazing students asking for advice like applying to grad school, what's it like being a PhD student in AI/ComputerVision/DeepLearning. I hope I can answer more broadly on here.
Mark Zuckerberg @finkd, Jensen Huang @JensenHuang, and Amil Khan @Amilworks now all on here. Current and future AI giants.