Intelligence.Log
2026-05-05
Extracted: 55 items. Sources: GitHub, Bluesky, X.
++ AI OVERVIEW ++
The open-source world saw a quiet but notable uptick in infrastructure and education, with **asg017/liblotus** (a Rust library starred by Simon Willison) and Hugging Face’s **context-course** (a new Python resource on context engineering for code agents) gaining traction. On Bluesky, the conversation turned sharply toward the ethics and limits of AI autonomy: **Simon Willison** pushed back on “AI-run business experiments” that waste non-consenting humans’ time, while **Nathan Lambert** lamented that even top-tier coding agents struggle with on-policy distillation despite being fed core papers and extensive context. Meanwhile, **Ethan Mollick** injected a political reality check into the “AI replacing doctors” debate, noting that powerful professional guilds (doctors, lawyers, bankers) hold voting power and deep community ties—factors often overlooked in purely technical forecasts. The day’s threads collectively underscore a growing tension between AI’s rapid deployment and the human systems—from consent to professional politics—that resist frictionless automation.
grep TOPIC=
grep SOURCE=
sort --by=
GH
A course on context engineering with code agents.
“This repository offers a course on context engineering specifically for code agents, covering how to design prompts and manage context to improve agent performance. It includes hands-on code examples and practical guidance for building more effective AI agents.”
GH
“Liblotus is a Rust library for building fast, embeddable vector search indexes with support for hybrid search (sparse + dense vectors). It offers efficient indexing and querying for semantic search applications.”
BSKY
AI-run business experiments are interesting and fun up to the point where they waste the time of humans who haven't opted into the experiments - I think they need to keep their own human operators in the loop for outbound actions that affect other people simonwillison.net/2026/May/5/o...
❤️ 42 Likes|[Agent][Safety]
BSKY
Adding an on policy distillation section to the RLHF book and it’s remarkable how bad LLMs / coding agents are at it, despite me giving them the core papers and 250 pages of context on how I present ideas.
❤️ 16 Likes|[LLM][Agent]
BSKY
Missing from the “will AI replace doctors?” debate is that doctors (and lawyers and psychologists and bankers) all vote & form the donor base to political parties & have deep community ties & sit in Congress. The government will largely determine what AI is allowed to do, no matter what it can do.
❤️ 54 Likes|[Safety]
X
2025 LLM Year in Review
[LLM]
“DeepSeek Summary: Karpathy posted a review of LLM developments in 2025, likely summarizing key trends and breakthroughs.”
X
LLM Knowledge Bases
Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating
[LLM][RAG]
“DeepSeek Summary: Karpathy describes using LLMs to build personal knowledge bases, shifting his token usage from code manipulation to knowledge manipulation.”
X
I'm being accused of overhyping the [site everyone heard too much about today already].
[LLM]
“DeepSeek Summary: Karpathy responds to accusations of overhyping a popular site, likely related to AI.”
X
Vibe coding is irresponsibly building software through dice rolls, not caring what code is produced
[Safety][Tooling]
“DeepSeek Summary: Simon Willison criticizes 'vibe coding' as building software irresponsibly without regard for code quality.”
X
This may be the best guidance I've seen anywhere on writing a really good commit history.
[Tooling]
“DeepSeek Summary: Simon Willison praises guidance on writing good commit history.”
X
A short note that the predictions that LLMs would favor 'boring technology' that's once you attach them to a good coding agent harness at least
[Agent][LLM]
“DeepSeek Summary: Simon Willison notes that LLMs favor boring technology when attached to a good coding agent harness.”
X
Visibility is the easiest piece. The hard part is analyzing and understanding what you're observing. I've spoken to teams recording 100k+
[Evaluation][Deployment][Tooling]
“DeepSeek Summary: Harrison Chase emphasizes that while gaining visibility into AI systems is straightforward, the real challenge lies in analyzing and understanding the observed data.”
X
TL;DR: More and more agents need a workspace: a computer where they can run code, install packages, and access files. Sandboxes provide this
[Agent][Infra][Deployment]
“DeepSeek Summary: Harrison Chase argues that AI agents increasingly require a sandboxed workspace to execute code and access resources safely.”
X
Resource constraints are a beautiful thing. Superior OSS models put huge pressure on...
[LLM][Infra]
“DeepSeek Summary: Jim Fan argues that resource constraints can be beneficial, as they force efficiency and innovation, and that open-source models create competitive pressure.”
X
The Second Pre-training Paradigm
[LLM][Multi-modal]
“DeepSeek Summary: Jim Fan introduces a new paradigm for pre-training, likely discussing a shift in how models are trained.”
X
I've been a bit quiet on X recently. The past year has been a transformational experience.
“DeepSeek Summary: Jim Fan reflects on a transformative year and his reduced activity on social media.”
X
Absolutely any time I try to explore something even slightly against commonly accepted beliefs,
[Safety]
“DeepSeek Summary: Jeremy Howard expresses frustration about exploring ideas that go against commonly accepted beliefs.”
X
I replicated this result, that Grok focuses nearly entirely on finding out what Elon thinks in
[Evaluation][Safety]
“DeepSeek Summary: Jeremy Howard replicated a finding that Grok AI focuses on Elon Musk's thoughts.”
X
reading "AI News" (previously Smol Talk) is probably the highest-leverage 45 mins
[LLM]
“DeepSeek Summary: Recommends a newsletter as high-leverage reading.”
X
I'm giving the opening Keynote at ICML 2024 on Tuesday the 23rd @ 9:30am CEST.
[LLM]
“DeepSeek Summary: Announcing a keynote at ICML 2024.”
X
I think it's clear that for many smaller companies that invested in deep learning, it turned out
[Evaluation]
“DeepSeek Summary: Deep learning investments may not have paid off for smaller companies.”
X
Re-reading an article I wrote in 2017, and I'm finding I could have written it yesterday
[Evaluation]
“DeepSeek Summary: Chollet's views on AI from 2017 remain relevant today.”
X
Folks who work in AI or software engineering feel like the world is changing exponential fast.
[Deployment]
“DeepSeek Summary: Perception of rapid exponential change in AI and software engineering.”
X
To really understand a concept, you have to 'invent' it yourself in some capacity.
[Evaluation]
“DeepSeek Summary: Understanding requires active reinvention, not passive learning.”
X
Dario is wrong. He knows absolutely nothing about the effects of technological revolutions on the labor market.
[Safety]
“DeepSeek Summary: Yann LeCun criticizes Dario's views on technological revolutions and labor market effects.”
X
It seems to me that before "urgently figuring out how to control AI systems much smarter than us" we need
[Safety]
“DeepSeek Summary: LeCun questions the urgency of controlling superintelligent AI.”
X
The emergence of superintelligence is not going to be an event. We don't have anything close to a
[Safety]
“DeepSeek Summary: LeCun argues superintelligence will be gradual, not sudden.”
X
Very excited to share @theworldlabs 's latest research work RTFM!! It's a real-time, ...
[Multi-modal][Agent]
“DeepSeek Summary: Fei-Fei Li announces World Labs' RTFM research, focusing on real-time spatial intelligence.”
X
Just received new reach minis for the Miami office! This is the first robot that goes out
[Agent]
“DeepSeek Summary: Clem Delangue announces arrival of Reachy mini robots for the Miami office, highlighting a new robot deployment.”
X
Gotta hand it to Labour's team. This is some top-drawer trolling.
[Evaluation]
“DeepSeek Summary: Phil Wang praises Labour's social media team for effective trolling.”
X
My Halloween costume this year is 'Sexy Stand-Up Comedian'.
[Evaluation]
“DeepSeek Summary: Phil Wang makes a self-deprecating joke about his profession for Halloween.”
X
If you were holding off to try @MSFTDeepSpeed ZeRO++ it looks like deepspeed@master should
[Infra][Deployment]
“DeepSeek Summary: Announcement that DeepSpeed ZeRO++ is available on master branch, encouraging users to try it.”
X
Hear, hear, I'm excited to introduce a new performance metric: Maximum Achievable Matmul
[Evaluation][Infra]
“DeepSeek Summary: Introduces a new performance metric for matrix multiplication, likely for benchmarking ML models.”
X
If you're trying out FA4, you're likely to run into not being able to load cutlass.cute
[Infra][Tooling]
“DeepSeek Summary: Warns about a common issue with FA4 (Flash Attention 4) related to loading cutlass.cute.”
X
Thanks to an awesome contribution from @omarnomad The Machine Learning Engineering Open book now can
[Tooling][LLM]
“DeepSeek Summary: Acknowledges a contribution to the Machine Learning Engineering Open Book.”
X
Had a nice time chatting about the state of diffusion models and some text-to-image data shenanigans at
[Multi-modal]
“DeepSeek Summary: Discussion on diffusion models and text-to-image data issues.”
X
Release notes: Release Diffusers 0.34.0: New Image and Video Models, Better torch.
[Tooling]
“DeepSeek Summary: Announcement of Diffusers 0.34.0 release with new models and improvements.”
X
2 years at HF today. Incredibly grateful for the mixed bag of opportunities I have been
[Deployment]
“DeepSeek Summary: Reflection on two years at Hugging Face.”
X
Guide: ReAct agent from scratch with Gemini 2.5 and LangGraph | Gemini API | Google AI for Developers. ai.google.dev.
[Agent][LLM][Deployment]
“DeepSeek Summary: Philipp shared a guide on building a ReAct agent from scratch using Gemini 2.5 and LangGraph.”
X
How to use Deep Research with the Gemini API. www.philschmid.de.
[Agent][LLM][Tooling]
“DeepSeek Summary: Philipp posted about using Deep Research with the Gemini API, linking to his blog.”
X
Here is a full implementation of the Chinese Room using a printed copy of GPT-1, in case you have a few spare years and want to actually run
[LLM]
“DeepSeek Summary: Mollick humorously suggests using a printed GPT-1 to run the Chinese Room thought experiment, highlighting the impracticality of simulating AI manually.”
X
So much work is going into faking continual learning and memory for AIs
[LLM][Evaluation]
“DeepSeek Summary: Mollick notes the trend of building systems that simulate continual learning and memory in AI, questioning the authenticity of such approaches.”
X
Sometimes when I demo AI, I show it turning cover letters into goofy formats (poetry, ...)
[Tooling]
“DeepSeek Summary: Mollick uses creative AI demos like converting cover letters into poetry to showcase AI's flexibility and engage audiences.”
X
If it helps, I teach at a business school & many of my smartest students are hired by funds because they can reliably turn their only-human ...
[Evaluation]
“DeepSeek Summary: Mollick observes that top business students are valued for their uniquely human skills, even in an AI-driven world.”
X
EMILY M. BENDER: Yeah. And so passive, like, oops, the moon, the moon went further away. It's like no, actually, you made some decisions.
[Safety]
“DeepSeek Summary: Critique of passive language in AI discourse, emphasizing human agency in decision-making.”
X
@emilymbender.bsky.social. emilymbender. Feb 10. Image is of the 1990s Microsoft writing assistant character Clippy with its eyebrows raised positioned in.
[LLM]
“DeepSeek Summary: Tweet includes an image of Clippy, likely used to comment on AI or technology.”
X
Facebook (sorry: Meta) AI: Check out our "AI" that lets you access all of humanity's knowledge.
[Deployment]
“DeepSeek Summary: Sarcastic critique of Meta's AI claims, mocking the idea of accessing all human knowledge.”
X
New preprint! Everyone loves causal interp. It's coherently defined! It makes testable predictions
[Evaluation]
“DeepSeek Summary: Announces a new preprint on causal interpretation, emphasizing its coherent definition and testable predictions.”
X
Ok, I wrote this up (link below)
[LLM]
“DeepSeek Summary: Indicates a write-up on a topic, with a link to further content.”
X
I work on understanding and improving training for NLP models, with a focus on studying how structures and mechanistic behaviors emerge over the
[LLM][Evaluation]
“DeepSeek Summary: Describes research focus on NLP model training and emergence of mechanistic behaviors.”
X
#throwback to the beginnings of a beautiful friendship =D @ansonmount @HellOnWheelsAMC #HellonWheels #onlocation.
[Multi-modal]
“DeepSeek Summary: Angela Zhou shares a throwback to the early days of a friendship with co-star Anson Mount on the set of Hell on Wheels.”
X
For the first time in almost a decade, I'm teaching a class on learning and control.
[Evaluation]
“DeepSeek Summary: Ben Recht announces teaching a class on learning and control after nearly a decade.”
X
Everyone knows actions are fundamentally different than predictions, but it's hard to write this
[Evaluation]
“DeepSeek Summary: Recht highlights the distinction between actions and predictions in machine learning.”
X
Very half-baked philosophy of engineering post: How do we prove that something is unpredictable in machine
[Evaluation]
“DeepSeek Summary: Recht questions how to prove unpredictability in machine learning systems.”
-- END OF LOG --
[STATS] 55 items · Filter applied
Powered by Horizon + DeepSeek