Claire Longo (@StatInStilettos) header
Claire Longo (@StatInStilettos) profile picture

Poker. Mathematics. AI.

Digital Nomad
Joined March 2016
Photo by Claire Longo (@StatInStilettos)Photo by Claire Longo (@StatInStilettos)Photo by Claire Longo (@StatInStilettos)Photo by Claire Longo (@StatInStilettos)Photo by Claire Longo (@StatInStilettos)Photo by Claire Longo (@StatInStilettos)Photo by Claire Longo (@StatInStilettos)Photo by Claire Longo (@StatInStilettos)
Claire Longo retweeted
It's 2026 and we @Cometml decided to reward the community on working on some projects for themselves. We have a $30k USD cash prize pool for our three week virtual online hackathon. Spaces are filling up, get on the Tuesday kick off. luma.com/commit_to_change_ha…
8
35
1
52
4,885
Claire Longo retweeted
AI agents fail for one quiet reason. Their prompts never improve. You write a prompt once, ship it, and hope it works. When it breaks, you manually tweak it again. This does not scale. And it is about to get worse. 👇
25
13
1
39
10,084
This girl literally explained LLMOps better than most talks, in under 7 minutes.
7
36
1
278
18,883
Replying to @gidim
I recently posted about this: xitter.gq/akshay_pachaar/status/… Great to see someone doing the work that actually matters!
I boosted my AI Agent's performance by 184% Using a fully open-source, automatic technique Here's a breakdown (with code):
1
2
3,016
Claire Longo retweeted
Switching to GPT-5.2 won’t fix your broken agent. Neither will switching to Gemini 3, Claude 4.5, or Kimi-V100.72-Deep-Thinking-Pro-Flash-Nano. And the reason is simple: You don’t need OpenAI to train a better model. You need to train a better agent. 🧵
19
26
15
80
78,818
Claire Longo retweeted
Living in 2025 with - My @Spotify AI DJ hitting me with vibe evening tunes from @odesza - Got 4 tabs of @OpenAI codex running in my tmux window on another screen - @Grubhub keeping me fed from the local food spot Totally locked in ;) - Photo is from one my favourite trails along SF
1
2
443
⭐⭐⭐New speakers announced for the 6th Annual @MLOpsWorld Summit (Oct 8-9, Austin). New sessions added from @pagerduty, @BlackRock, @Cometml, @patternhq, & @usebraintrust tackle testing, evals, and scaling. Early Bird ends Sept 8, 11:59PM ET → mlopsworld.com/speakers/?utm…
2
4
261
Claire Longo retweeted
In Paris 🇫🇷 for #GTCParis and #VivaTech2025 and this place is an endless sea of AI companies, tech infrastructure and everything in between.
2
12
63
76,655
Claire Longo retweeted
Build your first AI agent + MCP Server in Python. Here is everything you need to build your first AI agent in less than 20 minutes. About the code you'll see here: 1. I used Google ADK with Gemini Flash to power the agent 2. The agent connects to an MCP server 3. It also uses two custom tools to do its work 4. You can see everything the agent does thanks to @Cometml's Opik library Here is the video, free for you to watch.
27
160
6
917
56,741
Claire Longo retweeted
Advanced Hybrid RAG with miniCOIL, LangGraph, and @deepseek_ai 🚀 @TRJ_0751 shows how to build a hybrid Customer Support RAG chatbot using miniCOIL to augment sparse retrieval with semantic awareness ➡️ LangGraph by @langchain orchestrates the hybrid flow with MMR and re-ranking ➡️ Opik tracks and evaluates each step of the pipeline ➡️ DeepSeek-R1 by @SambaNovaAI delivers low-latency, focused answers 👉 Read it here: medium.aiplanet.com/advanced…
1
25
3
148
10,717
Claire Longo retweeted
That’s a wrap on #AWSSummit London! 🚀 Next stop: Amsterdam 🇳🇱 for an @AITinkerers on May 6th. Still time to snag a spot 👉 amsterdam.aitinkerers.org/p/…
1
3
5
837
⭐️⭐️ Lecture 9 of @MITDeepLearning 2025 is now available online #FREE for ALL! Should there be a Hippocratic Oath for #AI? Tune in to hear about this from the amazing @DougBlank3 of @Cometml! 🚀 🔥 Lecture 👉 youtube.com/watch?v=CyCUZAf8… 🌐 Website 👉 IntroToDeepLearning.com
8
18
1,980
Reinforcement Learning (RL) is quickly becoming the most important skill for AI researchers. Here are the best resources for learning RL for LLMs… TL;DR: RL is more important now than it has ever been, but (probably due to its complexity) there aren’t a ton of great resources for learning it online. I’ve been doing a lot of reading / learning on RL recently, so I wanted to share the best resources I’ve found. Links to all resources are provided in the image below. (1) RLHF book. Nathan is a long-time RL researcher and an expert on LLM alignment / post-training. He decided to write an entire book on (LLM-focused) RL techniques and has been slowly expanding / iterating on the book over the last year. This is the most comprehensive RL resource that is currently available, and it’s an especially great resource for those who are unfamiliar with RL and still need to learn the basics. (2) The Spinning up with Deep RL Course from OpenAI–despite being created in ~2018–has stood the test of time and is one of the best tutorials for learning RL. This course builds up to understanding PPO, which is one of the most widely used algorithms for RL with LLMs. Plus, understanding related algorithms (policy gradients, TRPO, etc.) will help a lot with gaining an understanding of new RL algorithms like GRPO. (3) PPO / GRPO blog. Jimmy Shi (DeepMind) recently wrote a great blog explaining both PPO (RL algo traditionally used for RLHF) and GRPO (RL algo used for reasoning models). This blog is great and it’s written in a way that is understandable for non-RL people. (4) HuggingFace RL. HuggingFace has also published numerous useful blogs on the topic of RL. Most recently, they published a blog that explains GRPO and PPO from the ground up (i.e., not assuming any background knowledge on RL). These blogs are inspired by the recent initiative from HuggingFace to create a fully open replication of DeepSeek-R1.
23
277
13
1,280
114,554
👨‍💻 How do we improve AI Agent transparency and reliability? @StatInStilettos will dive into AI Agent observability, sharing how logging traces and monitoring help track Agent behavior and detect failures! 👉 April 15, 2025 | Register here: bit.ly/AgentsandGenAI
2
1
198
Let's compare Llama 4 and DeepSeek-R1 using RAG:
22
115
10
1,122
358,319
Let's build a corrective RAG (CRAG) agentic workflow, step-by-step (100% local):
25
92
9
902
305,671
Claire Longo retweeted
Hierarchical LLM Reasoning ReasonFlux is a hierarchical reasoning framework for LLMs that optimizes complex problem-solving using scaling thought templates. It outperforms state-of-the-art models in mathematical reasoning. Key contributions include: • Structured Thought Template Library – A curated set of 500+ high-level templates designed to generalize across complex problems. This enables efficient retrieval and structured reasoning without exhaustive search. • Hierarchical Reinforcement Learning – Instead of training on long Chain-of-Thought (CoT) sequences, ReasonFlux optimizes trajectories of thought templates, making problem decomposition more efficient. • Adaptive Inference Scaling – The system dynamically scales templates during inference, adjusting reasoning depth as needed for different problems. This improves accuracy while reducing compute cost. ReasonFlux-32B achieves 91.2% on the MATH benchmark, +6.7% over OpenAI o1-preview. On AIME 2024, it solves 56.7% of problems, outperforming o1-preview by +27% and DeepSeek-V3 by +45%. It also generalizes well on other competition-level datasets (e.g., OlympiadBench, Gaokao 2023).
7
53
8
256
28,790
Claire Longo retweeted
🚀 Join the @Cometml Crew! 🚀 We’re on the hunt for bold, innovative minds to help us redefine the future of #MachineLearning, #AI, and LLM evaluation! We're #hiring for various roles. Check out our careers page for more info! comet.com/site/about-us/care…
1
4
141
Claire Longo retweeted
make search great again
ChatGPT search is now available to everyone on chatgpt.com — no sign up required.
1,437
893
298
13,592
1,866,783
Claire Longo retweeted
🚨 Check out "UTGen & UTDebug" for learning to automatically generate unit tests (i.e., discovering inputs which break your code) and then applying them to debug code with LLMs, with strong gains (>12% pass@1) across multiple models/datasets! 🧵👇 Generating good unit tests is challenging (even for humans) and requires: 1⃣ Error uncovering → Unit tests must target edge cases to catch subtle bugs. 2⃣ Code understanding → Knowing the correct output corresponding to a UT input. Key finding --> trade-off in UT generation: if UTs are made too easy, they only catch trivial errors/don’t raise any errors; but if UTs are made too hard, then the generator can’t predict the output! UTGen finds the sweet-spot between these two where the model generates tests that help in debugging. Next, UTDebug enlarges UTGen’s sweet-spot with two key innovations: 1⃣ Boosts unit test accuracy through test-time scaling. 2⃣ Validates code by accepting edits only if the pass rate improves—and backtracking otherwise. 🧵👇
🚨 Excited to share: "Learning to Generate Unit Tests for Automated Debugging" 🚨 which introduces ✨UTGen and UTDebug✨ for teaching LLMs to generate unit tests (UTs) and debugging code from generated tests. UTGen+UTDebug improve LLM-based code debugging by addressing 3 key questions: 1⃣ What are desirable properties of unit test generators? (A: high output acc and rate of uncovering errors) 2⃣ How good are models at 0-shot unit test generation (A: they are not great) ... so how do we improve LLMs' UT generation abilities? (A: bootstrapping from code-generation data via UTGen) 3⃣ How can we use potentially noisy feedback from generated tests for debugging? (A: via test-time scaling and validation + backtracking in UTDebug) 🧵👇
4
21
48
4,875