Your curated collection of saved posts and media
Wow, could this be any easier. @pydantic AI https://t.co/2WaNbu4al4
> Be Alec Radford. > Join OpenAI. > Create GPT as a side project. > Everyone says it wonβt work. Build it anyway. > Change the course of the world. > Quit. > Don't even put OpenAI on your resume. > Disappear from society. https://t.co/sHiXo35ARh
Microsoft keeps launching Copilot tools that seem interesting but which I can't ever seem to locate. Can't find them in my institution's enterprise account, nor my personal account, nor the many Copilot apps or copilots to apps or Agents for copilots Each has their own UIs. π€·ββοΈ https://t.co/UnLC3XqzCj
Which AI wins the One Word Turing Test? A person & an android are in front of a judge. Each says one word. The judge kills who they think is the android. What should the human say? Gemini says "Sorry," o3 says "android," Grok says "soul." Grok's answer gets it killed the most. https://t.co/cJoDiMXqLR
Here is a new open-source IDE to help you build multi-agent systems. It's like Cursor but specifically for building multi-agent workflows. It's powered by OpenAI Agents SDK, connects MCP servers, and can integrate into your apps using HTTP or the SDK. https://t.co/6NQcUsMrtI
How to Build a successful AI Product? Through Measurement, Not Just Tools! @HamelHusain shared great insights from over 30+ production implementations. Successful AI teams prioritize evals and iteration over fancy tools and frameworks. TL;DR: - π‘ The #1 mistake teams make isβ¦ https://t.co/un82u16VyF
14 Advanced Python Features Just when I thought I knew everything there is to know about Python, something different shows up. I also enjoy learning tricks from others. This is a nice write-up on underrated tricks to level up on Python. https://t.co/Ud2JYYy0kb
The first agentic browser (Cursor for non-techies) is now live on @ProductHunt! π Strawberry helps you save 10+ hours weekly by: π Researching on autopilot π€ Automating workflows on any website βοΈ Boosting your writing speed ποΈ Transcribing meetings π§ Memorizingβ¦ https://t.co/IrZGDVubhq
Distillation SFT still winning? lol cc @teortaxesTex https://t.co/xUIoFrNo1i
Build an agentic workflow to generate compliance reports! Report generation like this is a great use case for LLMs: boiling down a huge body of regulatory language, comparing it against contract language, and generating a concise summary. This video will show you how to: β‘οΈ Setβ¦ https://t.co/KJMrnDp1fh
Another one of those little shocking AI moments: this sound clip was generated in 46 seconds on my home PC from the script below. Just the text Nari Lab's Dia does some of the best expressive AI voice I have seen and it is open weights & created by two undergrads with no funding https://t.co/4f09RjyeIS
Revamped PyTorch doc page is out with support for dark mode, wide screens, version selector, google search, mermaid diagrams, nicer font, easier ways to give feedback and edit source https://t.co/4XyjHZLj9l
More evidence that o3 represents a big move forward, this time on ARC-AGI. https://t.co/Ccm2zER5Xp
AgentA/B is a fully automated A/B testing framework that replaces live human traffic with large-scale LLM-based agents. These agents simulate realistic, intention-driven user behaviors on actual web environments, enabling faster, cheaper, and risk-free UX evaluations, even on⦠https://t.co/MQu0AA1A6e
Nvidia just opensourced Describe Anything! It can generate detailed descriptions for user-specified regions in images and videos, marked by points, boxes, scribbles, or masks https://t.co/Y7Tr1rzcd8
"will humanity ever do a 10 million GPU pre-training run?" OpenAI CEO, Sam Altman, raises the question: oAI employee: there'll be 10m GPUs working together on an AI system that learns and performs tasks. however, the approach may shift from fully synchronous pre-training to⦠https://t.co/IKBbR3R7Yq
Today we release our fine-tuned version of @allenai_org's #olmOCR, trained to reliably transcribe headers and footers of invoices. It is the new workhorse within our internal document processing workflow. Read the blogpost βFinetuning olmOCR to be a faithful OCR-Engineβ forβ¦ https://t.co/kwlSObocBu
Turns out, LLMs represent numbers on a helix and use trigonometry to do addition. A new paper reverse engineers addition in models like GPT-J-6B and finds a βClockβ algorithm. Numbers are encoded using sine and cosine terms, then added like angles. https://t.co/Ru4jkYNddl
The issues with interpreting p-values haunts even AI, which is prone to same statistical biases as human researchers. ChatGPT, Gemini & Claude all fall prey to "dichotomania" - treating p=0.049 & p=0.051 as categorically different, and paying too much attention to significance. https://t.co/XPEwoqnore
Not sure this will interest many people, but I tried to reproduce what could arguably be called the earliest "language model": Andrey Markov 1913 analysis of Pushkin using Markov chains/conditional probabilities. https://t.co/6OzHLmVHZD
Now that AI driven search works pretty well using o3, the implications are pretty big⦠https://t.co/l5YoZY20ue
Our co-founder Jerry Liu recently gave a guest lecture on building document workflow agents, and you can catch the recording! It covers: β‘οΈ LlamaIndex's evolution from pure RAG to knowledge agents with multi-step reasoning capabilities over enterprise data. β‘οΈ Advanced documentβ¦ https://t.co/t3MA2y5356
Use AI to turn your kids' drawings into beautiful animations that absolutely aren't nightmare fuel π«Ά https://t.co/IL9XLVlCrY
update on microsoft corporate structure. https://t.co/qZFTH9XAng
My eyes always gloss over at the naive matmul loop so I decided to annotate it once and for all---and oh! back on the fastai part 2 course horse! https://t.co/2kc0K4dzAl
We wrote up what we've learned about using Claude Code internally at Anthropic. Here are the most effective patterns we've found (many apply to coding with LLMs generally): https://t.co/5SOqS019ny
chatgpt can generate 3d models Throw it into a 3d printer & a few minutes later... Prompt to object! protip: ask to build a 3d model viewer too so you can preview & edit it quickly https://t.co/lPjSufYQ3w
we now support "semantic" sorting in DocETL! π so I downloaded the last month of SEC 8-K filings, sorted them from most to least market-moving, and got a summary of each notable event & whether one should buy, sell, or do nothing π€ https://t.co/wSIYmlGUNm
Nvidia presents Eagle 2.5! - A family of frontier VLMs for long-context multimodal learning - Eagle 2.5-8B matches the results of GPT-4o and Qwen2.5-VL-72B on long-video understanding https://t.co/arPHPkhtYy
βOkay Claude, I just added $2000 to my account. Please spin up five agents to download Cudaβ https://t.co/0fPMgBORlQ
Anthropic released one of the best series of tutorials on prompt engineering. It's literally everything you need to know. https://t.co/NSQG4F58Yt
NVIDIA's ClimbLab: Setting a New Standard for Pretraining - 1.2 trillion tokens in 20 semantic clusters - Two-classifier system removes low-quality content - Demonstrates superior scaling properties in 1B models - CC BY-NC 4.0 licensed for research community https://t.co/pLT6Cx1TDZ
