
Participate in design @PalantirTech @Betteromics Currently serving @radiantai_inc Python and JS are my destiny, but changing requirements is my calamity.
program,ca
Joined April 2010
- Tweets1.1K
- Following107
- Followers74
- Likes34
Pinned Tweet
We build
@radiantai_inc
exactly for this moment when you realize you might want to rely on more than just one vendor. It takes one click to migrate production AI apps -- you get our cost management, security, and governance capabilities out of the box.
radiantai.com/blog/letter-fr…
Allen Cai retweeted
DeepSeek-V3-0324 is insane at UI 🤯
Things are getting wild someone prompted this 👀
Build your app now: hf.co/spaces/enzostvs/deepsi… (no login required)
Allen Cai retweeted
Google Gemini 2.5 is the first public AI model to definitively beat the level of performance of human PhDs with access to Google on hard multiple choice problems inside their field of expertise (around 81%).
All AI tests are flawed, but GPQA Diamond has been a pretty good one.
Allen Cai retweeted
How capable are AI agents of replicating complex AI research? @OpenAI introduces PaperBench, a new benchmark evaluating the coding ability of AI agents to replicate state-of-the-art AI research. Agents must replicate 20 ICML 2024 Spotlight and Oral papers from scratch.
Benchmark, Agents:
1️⃣ Receive an ICML 2024 Spotlight/Oral paper and addendum (no rubric shown).
2️⃣ Understand the paper's core empirical contributions.
3️⃣ Develop a complete codebase *from scratch* to implement the experiments.
4️⃣ Create a submission repository including a reproduce script as the execution entry point.
5️⃣ The reproduce script is executed in a sandbox environment (VM/Docker with GPU).
6️⃣ The result is graded against a detailed, hierarchical, author-co-developed rubric using an automated LLM-based judge.
Insights
- 💡 From tested models Claude 3.5 Sonnet performed best but only with 21.0% accuracy
- 🧠 OpenAI's o1 jumped from 13.2% to 24.4% with improved prompting
- 👨🎓 Human ML PhDs outperform AI agents (41.4% vs 26.6%) on a subset after 48 hours.
- 🧪 LLM judge (SimpleJudge with o3-mini-high) is used for grading
- 📉 Tested models struggle with long-horizon planning, execution
- 🛠️ Released a lighter version, PaperBench Code-Dev, focusing only on development
Allen Cai retweeted
New Short Course: Getting Structured LLM Output!
Learn how to get structured outputs from your LLM applications in this course, built in partnership with @dottxtai, and taught by @willkurt, a Founding Engineer, and @just_cameron , Developer Relations Engineer.
It's challenging for software to automatically parse through an LLM's freeform text outputs. Structured outputs—like JSON—solve this by converting natural language into consistent, clear, data that a machine can read and process. This course teaches you how to generate structured outputs while building several use cases, including a social media analysis agent.
You’ll learn about structured outputs and efficient ways to generate outputs in your defined schema or format. You’ll begin by using structured output APIs, then use re-prompting libraries like “instructor” to generate structured output. Finally, you’ll learn how constrained decoding works; this is a very clever technique in which constraints are applied on each subsequent token generated, blocking any tokens that don’t fit your defined schema.
In detail, you’ll:
- Learn why structured outputs are important, how they allow for scalable software development, and the different approaches to generate them, including vendor-provided APIs, re-prompting libraries, and structured generation.
- Build a simple social media agent using OpenAI’s structured output API, learn how to define a model's desired structured output using Pydantic, and perform basic programming with your outputs, such as importing structured data into a data frame using pandas.
- Learn how to use the open-source library "instructor," which checks the structured output of the model and re-prompts the model until it validates the desired output, and explore the limitations of this approach.
- Understand how structured generation by the “outlines” library works by modifying LLM logits, on a per-generated-token basis based on the desired format, to give a particular output structure.
- Learn how regular expressions, which outlines works with, are represented as finite-state machines, and how they can be used to develop a range of structured outputs beyond JSON.
By the end of this course, you’ll have broadened your knowledge of the approaches you can use to get structured outputs from your LLM applications.
Please sign up here: deeplearning.ai/short-course…
Allen Cai retweeted
Days like today are why JD Vance was selected. The entire trajectory of MAGA since 2016 has been towards selecting for sycophancy. The people who are there now are there because they’re the least likely to question “the president’s desires,” as Vance puts it. We’re in trouble.
Allen Cai retweeted
Another huge week in AI Agents 🧵
I summarized everything announced by AgentOps, Browser Use, LangChain, CoPilotKit, Triple Whale, Replit, Outshift by Cisco, Mastra AI, and more.
Here's everything you need to know and how to make sense of it:
(save for later)
Allen Cai retweeted
Here is a fully Agentic example on how to use MCP servers with @GoogleDeepMind Gemini. It shows how to use the Airbnb MCP to find rooms in Paris this weekend for a booking.
> Define MCP server
> Convert MCP tools to Gemini
> Initial Request
> Agentic Loop for automatic tool calling
> Final Response
code ⬇️
Allen Cai retweeted
wrote about the CSS scroll-snap techniques to do this last year 🤙
even easier with some of the newer APIs like onscrollsnapchange and scroll animation
might be time to drop the more polished version 🧑🍳
Allen Cai retweeted
Replying to @SpaceXSI
@xAI has acquired @X in an all-stock transaction. The combination values xAI at $80 billion and X at $33 billion ($45B less $12B debt).
Since its founding two years ago, xAI has rapidly become one of the leading AI labs in the world, building models and data centers at unprecedented speed and scale.
X is the digital town square where more than 600M active users go to find the real-time source of ground truth and, in the last two years, has been transformed into one of the most efficient companies in the world, positioning it to deliver scalable future growth.
xAI and X’s futures are intertwined. Today, we officially take the step to combine the data, models, compute, distribution and talent. This combination will unlock immense potential by blending xAI’s advanced AI capability and expertise with X’s massive reach. The combined company will deliver smarter, more meaningful experiences to billions of people while staying true to our core mission of seeking truth and advancing knowledge. This will allow us to build a platform that doesn’t just reflect the world but actively accelerates human progress.
I would like to recognize the hardcore dedication of everyone at xAI and X that has brought us to this point. This is just the beginning.
Thank you for your continued partnership and support.
Allen Cai retweeted
Ableton MCP from @sidahuj shows the real future of AI creativity partnership—AI handles the tedious tasks while you maintain creative control. No more workflow disruptions; just pure creative experimentation unleashed.
And it's also just super cool.
Volume Up 🔊
Allen Cai retweeted
I added a new agent to our AI hedge fund.
The Phil Fisher agent.
This new agent:
1 • seeks long-term R&D growth
2 • demands consistent profit margins
3 • evaluates management quality
4 • will pay a premium for high growth
5 • performs "scuttlebutt" research
Phil Fisher was a legendary growth investor.
Warren Buffett credits Fisher for significantly influencing his own investment philosophy.
The Phil Fisher agent only says “buy” when it finds companies with outstanding future potential AND proven operational excellence.
Here is how it works:
1. Long-Term Growth: loves companies investing 3-15% of revenue in R&D with multi-year revenue growth above 40%.
2. Consistent Profits: Identifies stable or improving operating margins with low volatility.
3. Excellent Management: Requires high ROE (>20%), low debt, and consistent positive free cash flow.
4. Fair Valuation: Willing to pay premium prices for quality growth.
5. Scuttlebutt Research: Monitors executive buying patterns and news sentiment for confirmation of business strength.
Allen Cai retweeted
Today at 10 a.m. at the San Jose Civic Center: Join NVIDIA CEO Jensen Huang and industry leaders from D-Wave, @AWS, and more for an exciting discussion on the current landscape and future of #quantumcomputing. #GTC25.
🔗 nvda.ws/3RdhWqR
Allen Cai retweeted
Connect any AI Agent to any Data Source and Tools.
AgentIQ by Nvidia lets you connect your agent to any data source with simple function calls. Reuse existing agents and tools across different workflows.
Works with LangChain, Llama Index, Crew AI, and more.
100% opensource.
Allen Cai retweeted
🦜🔨LangGraph Builder
Build the "cognitive architecture" of your agent in a UI, and generate code stubs!
Video tutorial: youtu.be/iwPeT_I_GEc
Try it out here: build.langchain.com/
OSS as well:
Frontend: github.com/langchain-ai/lang…
CLI: github.com/langchain-ai/lang…
Allen Cai retweeted
📣 NVIDIA Isaac for Healthcare
Announced at #GTC25, this is an AI-powered medical robotics development platform. Developers can use a digital twin and robotic learning framework to build, test, and model smart healthcare robotic systems.
Allen Cai retweeted
Watch NVIDIA CEO Jensen Huang unveil the latest breakthroughs in #AI, robotics, accelerated computing, and more in the #GTC25 keynote.
Now LIVE ➡️ nvda.ws/4iyuzc5
📣 Keep up with all the announcements on @nvidianewsroom

























