Jeremy Howard
@jeremyphoward
🇦🇺 Co-founder: @AnswerDotAI & @FastDotAI ; Prev: professor @ UQ; Stanford fellow; @kaggle president; @fastmail/@enlitic/etc founder https://jeremy.fast.ai/
In my opinion, it's ok for Anthropic to be a business and act like one. But this reinforces the need for open science and open-source AI to avoid concentration of power and control in the hands of a few of these businesses, otherwise we'll be in big trouble!
SCOOP: Leaked memo from Anthropic CEO Dario Amodei outlines the startup's plans to seek investment from the United Arab Emirates and Qatar. “Unfortunately, I think ‘no bad person should ever benefit from our success’ is a pretty difficult principle to run a business on.”
Best model with an OSI-approved license: 🇨🇳: R1, Qwen3 🇪🇺: Mistral Small 🇺🇸: IBM Granite
sorry my verdict on Grok-4 is that it is not better than Opus for coding, and not better for o3 for reasoning. I don't think it has been trained on benchmarks, but I think its brain is deep friend into a problem-solution mindset that doesn't extend to real-world situations...…
>>> Qwen3-Coder is here! ✅ We’re releasing Qwen3-Coder-480B-A35B-Instruct, our most powerful open agentic code model to date. This 480B-parameter Mixture-of-Experts model (35B active) natively supports 256K context and scales to 1M context with extrapolation. It achieves…
✨Announcing: tinyio! A tiny barebones event loop library for Python. Born out of my frustration with asyncio... GitHub: github.com/patrick-kidger… It's nothing too fancy, just a little library that does one thing well. 🔥
This will assuredly ruffle feathers since it involves money (and nuance) I think we, as founders, need to say no to partial acquisitions Otherwise, less people will go into startups because the social contract of "Ride (or die) Together" gets fractured And then everyone loses
NEW: How partial acquisitions are killing startup culture — and what we can do about it. Earlier this month, after Google poached Windsurf’s braintrust and licensed its IP (Cognition later acquired what was left), @JustJake coined the term “shell-qui-hires” to describe…
I believe that lesscss.org was replaced by sass/scss in @getbootstrap 4 due to social pressure applied in the after-work SF tech scene rather than technical merit
What's a conspiracy theory you believe with your whole heart?
Opposition to the moon mission is a *great* example how we memory hole pessimism in ways that distorts our understanding of history and progress buff.ly/1q248ce
Also it's actually openly licensed - nvidia have been really improving their model licensing game recently :D
Wait NVIDIA has just released new SOTA open source models?! Available in 4 sizes 1.5B, 7B, 14B and 32B that you can run 100% locally. - OpenReasoning-Nemotron - SOTA scores across many benchmarks - Tailored for math, science, code How to run it on your laptop and details below
TL;DR: Qwen series finetuned on 5M reasoning traces from DeepSeek R1 0528 671B, i.e. hard distillation.
Wait NVIDIA has just released new SOTA open source models?! Available in 4 sizes 1.5B, 7B, 14B and 32B that you can run 100% locally. - OpenReasoning-Nemotron - SOTA scores across many benchmarks - Tailored for math, science, code How to run it on your laptop and details below
Our IMO gold model is not just an "experimental reasoning" model. It is way more general purpose than anyone would have expected. This general deep think model is going to be shipped so stay tuned! 🔥
So happy to see this incredible achievement. Huge congrats to @lmthang, @quocleix, @YiTayML and the IMO team on the result. This was a great collaboration across teams to build a general Gemini DeepThink model that can also get gold at IMO.
Kimi K2 paper dropped! describes: - MuonClip optimizer - large-scale agentic data synthesis pipeline that systematically generates tool-use demonstrations via simulated and real-world environments - an RL framework that combines RLVR with a self- critique rubric reward mechanism…
Officially validated IMO gold medal, purely via search in token space, achieved in 4.5 hrs (unclear at what compute cost). The solutions read nicely as well deepmind.google/discover/blog/…
MIRI had just published the most Orwellian proposal possible: · Government surveillance of datacenters via hardware-enabled mechanisms and software-based tools. · Tracking personnel: Surveillance of key AI researchers and their locations, computers, and research activities. WTF
The majority of COVID-19 infections occur via near-field transmission (aerosols travelling 1-2m from infectious person). Sitting below a ceiling fan reduces aerosol concentration in the breathing zone by 87%. @ukhadds sciencedirect.com/science/articl…
I debated 20 far-far-far-right conservatives on Jubilee's 'Surrounded.' Below are some brief highlights of them unable to answer my basic questions or rebut the simplest of points. Not sure whether to laugh or cry. Here's the full eye-opening 'debate': youtube.com/watch?v=2S-WJN…
56 years ago today, America landed on the moon. Everyone forgets most of America opposed the Apollo Program: newsletter.pessimistsarchive.org/p/most-america…
The sad robot in matharena.ai/imo/ is Grok 4. This shows again how careful one has to be with overblown claims from closed releases saying the usual "it's so over". Test contamination that cannot be checked makes benchs look great, but on novel problems, the crash comes.
developer.nvidia.com/blog/cutlass-p… marks the start of a short series of blogposts about CUTLASS 3.x and CuTe that we've been meaning to write for years. There are a few more parts to come still, hope you enjoy!