Intelligence.Log
2026-05-21
Extracted: 59 items. Sources: GitHub, Bluesky, X, Blogs.
++ AI OVERVIEW ++
Today's discourse is split between the environmental cost of frontier AI and a playful pushback against its homogenization. Ethan Mollick sparked a concrete debate by calculating the resource footprint of solving the Erdős problem, estimating it consumed 0.6–6.3 kWh of electricity and 3–31L of water—a stark reminder of the physical toll behind abstract intelligence. In a lighter but pointed vein, Naomi Saphra announced a new literary award with a deliberate barrier to entry: commercial frontier LLMs are disqualified by requiring 10% of each submission to be smut, a clever provocation about the sanitized, risk-averse nature of current models. The contrast highlights a growing tension between celebrating AI’s capabilities and critiquing its unsustainable scale and cultural blandness.
grep TOPIC=
grep SOURCE=
sort --by=
GH
Open Source Robotic Arm for All Developers
“reBot-DevArm is an open-source robotic arm designed for developers, offering a platform for experimentation and learning in robotics. It provides hardware schematics, firmware, and software tools to enable custom control and integration with AI systems.”
GH
Universal Rust multiplexer with a typed SDK — drive any CLI or TUI app from code. Native on Linux, macOS, and Windows.
“rmux is a universal Rust multiplexer that allows you to programmatically drive any CLI or TUI application via a typed SDK. It supports native execution on Linux, macOS, and Windows, making it a cross-platform tool for automating terminal interactions.”
GH
AI agent skills published by NVIDIA
“NVIDIA's curated collection of reusable AI agent skills, designed to accelerate development of agentic workflows. Provides modular, production-ready building blocks for common agent tasks.”
GH
Pi coding agent extension: llama.cpp provider with dynamic model + context window discovery
“Pi-llama is a coding agent extension that integrates llama.cpp as a provider, enabling dynamic model and context window discovery. It allows users to leverage local LLMs for coding tasks with flexible model selection and automatic context size adjustment.”
GH
“Teich is a Python library for building and managing AI agents with a focus on modularity and extensibility. It provides tools for agent orchestration, memory management, and tool integration, aiming to simplify the development of complex AI workflows.”
BSKY
We can estimate the resource cost of solving the Erdos problem. The calculations below seem reasonable, so using the best public estimates we have, it took 0.6–6.3 kWh of electricity & 3–31L of water Equivalent to < 3 almonds worth of water & the electricity equivalent of 2-20 miles of EV driving
❤️ 94 Likes|[Infra]
BSKY
my new literary award cannot be won by a commercial frontier LLM because I will require that 10% of each submission is smut
❤️ 15 Likes|[Evaluation]
BSKY
I released the first alpha of Datasette Agent - a conversational AI assistant for Datasette that can answer questions about data in SQLite databases, and can be extended with plugins to add extra tools and features You can watch a demo video or try it out yourself on agent.datasette.io
❤️ 38 Likes|[LLM][Agent][Tooling]
BSKY
US government has pit on hold plans to evaluate AI systems before their release. Cites competition with China www.nytimes.com/2026/05/21/t...
❤️ 6 Likes|[Safety][Evaluation]
BSKY
BSKY
BSKY
Seems GPT-5.2 reaches expert level in peer review: 45 scientists took 469 hours evaluating human & AI reviews on 82 papers. "Surprisingly, current AI reviewers are competitive even with the top-rated reviewers in Nature’s official peer review..." though not without weaknesses, so use AI + humans.
❤️ 66 Likes|[Evaluation][LLM]
BSKY
There has been a lot of speculation that AI companies were unprofitable, but Anthropic will have an operating profit of $559M this quarter. “In the first quarter, Anthropic spent 71 cents on computing power for every dollar it made. In the current quarter, it expects to spend 56 cents per dollar…”
❤️ 112 Likes|[Infra][Deployment]
BSKY
I have been thinking about this in light of Anthropic’s recent verbalization interp paper. It had no evidence convincing me that their verbalizations are faithful, but they are convincingly useful. Even wrong output can stimulate human creativity and increase the entropy of exploration.
❤️ 22 Likes|[Safety]
BSKY
Maybe it is a good day to go to the Whitney in NYC and look at The Rose. It is very big. A human spent 8 years painting it. She made it too big and couldn't get it back out her door. They had to cut a hole in the wall and forklift it out.
❤️ 28 Likes|
BSKY
❤️ 6 Likes|
X
Drafted a blog post - Used an LLM to meticulously improve the argument over 4 hours.
[LLM]
“DeepSeek Summary: Karpathy used an LLM to refine a blog post over 4 hours, showcasing iterative AI-assisted writing.”
X
Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around
[LLM]
“DeepSeek Summary: Karpathy observes a widening gap in public understanding of AI capabilities and identifies a key issue.”
X
Everything about the LLM stack is different (neural architecture, training data, training algorithms, and especially optimization pressure) so
[LLM][Infra]
“DeepSeek Summary: Karpathy emphasizes the distinctiveness of the LLM stack across multiple dimensions.”
X
A few random notes from claude coding quite a bit last few weeks. Coding workflow. Given the latest lift in LLM
[LLM][Tooling]
“DeepSeek Summary: Karpathy shares notes on using Claude for coding, focusing on workflow improvements from recent LLM advances.”
X
Excited to share that I am starting an AI+Education company called Eureka Labs. The announcement: --- We are
[LLM]
“DeepSeek Summary: Karpathy announces the launch of Eureka Labs, an AI+Education company.”
X
Vibe coding is irresponsibly building software through dice rolls, not caring what code is produced
[Agent][Tooling]
“DeepSeek Summary: Simon Willison critiques 'vibe coding' as an irresponsible approach to software development.”
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 might favor boring technology when paired with a good coding agent harness.”
X
I'm beginning to suspect that a key skill in working effectively with coding agents is developing an intuition for when you don't need to
[Agent][Tooling]
“DeepSeek Summary: Simon Willison suggests that intuition for when not to intervene is key for coding agent effectiveness.”
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][Tooling]
“DeepSeek Summary: Visibility is easy, but analyzing observations is hard.”
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][Safety]
“DeepSeek Summary: Agents require sandboxed workspaces for code execution.”
X
Today we're launching LangChain Labs, a new applied research effort focused on Continual Learning. Our goal is to advance open,
[LLM][Fine-tuning]
“DeepSeek Summary: LangChain Labs launched for continual learning research.”
X
Everyone wants to ship agents. The best organizations have figured out how to do it repeatedly, safely, and systematically.
[Agent][Deployment][Safety]
“DeepSeek Summary: Top organizations ship agents repeatedly and safely.”
X
Resource constraints are a beautiful thing. Survival instinct in a cut-throat AI competitive land
[Agent]
“DeepSeek Summary: Jim Fan reflects on how resource constraints drive innovation in AI.”
X
It gives me a lot of comfort knowing that we are the last generation without advanced robots everywhere.
[Agent]
“DeepSeek Summary: Jim Fan expresses comfort in being the last generation before widespread advanced robots.”
X
In this context, I define world modeling as predicting the next plausible world state (or a longer duration of states) conditioned on an action.
[Multi-modal]
“DeepSeek Summary: Jim Fan defines world modeling as predicting future world states given an action.”
X
We are scientists, engineers, and builders behind some of the most widely used AI products and libraries, including ChatGPT.
[Infra][LLM]
“DeepSeek Summary: Soumith Chintala announces Thinking Machines Lab, highlighting their team's role in creating widely-used AI products like ChatGPT.”
X
Many people assume that LRM reasoning breaks down past a certain 'complexity' or 'number of steps'
[LLM][Evaluation]
“DeepSeek Summary: Chollet comments on a common assumption about reasoning in large reasoning models (LRMs).”
X
It's surprisingly easy to do 'hard' things -- for the most part, you need to get started and keep at it
[Tooling]
“DeepSeek Summary: Chollet shares a motivational thought about tackling hard tasks.”
X
Dario is wrong. He knows absolutely nothing about the effects of technological revolutions on the labor market.
[Safety]
“DeepSeek Summary: LeCun criticizes Dario's understanding of technological revolutions and labor market effects.”
X
I love Geoff. But he understands even less than Dario about the effects of technological revolutions on
[Safety]
“DeepSeek Summary: LeCun critiques Geoff Hinton's understanding of technological revolutions on labor, while expressing personal affection.”
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, suggesting a different priority.”
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 a sudden event, and we lack the foundations.”
X
LOL. Remove the code in the algorithm that boosts the tweets of Elon by elvodqa · Pull Request #160 ·... github.com.
[Evaluation]
“DeepSeek Summary: Max Woolf finds humor in a GitHub pull request that removes code boosting Elon Musk's tweets.”
X
Wager established. Jonathan Frankle (@jefrankle) stepped up to my Transformer long bet.
[LLM]
“DeepSeek Summary: Sasha Rush established a long bet with Jonathan Frankle about Transformers.”
X
I have been compiling LLM/VLM training logbooks/chronicles. This is the one of the best sources to ...
[LLM][Fine-tuning][Infra]
“DeepSeek Summary: Stas Bekman compiles LLM/VLM training logbooks, providing a valuable resource for training insights.”
X
Thanks to an awesome contribution from @omarnomad The Machine Learning Engineering Open book now can ...
[Tooling][Infra]
“DeepSeek Summary: Acknowledges a contribution to the Machine Learning Engineering Open Book, expanding its capabilities.”
X
This is a long overdue section of the ML Engineering Understanding Training Loss Patterns ...
[LLM][Fine-tuning][Evaluation]
“DeepSeek Summary: Introduces a new section on understanding training loss patterns in ML engineering.”
X
Modern art. Artist: PyTorch memory profiler Model: Llama-8B The piece on the left is the ...
[Infra][Tooling]
“DeepSeek Summary: Uses PyTorch memory profiler output as a form of modern art, showing memory patterns of Llama-8B.”
X
1. Read the post. 2. Contemplate. 3. Repeat 1.
[LLM]
“DeepSeek Summary: Sayak Paul shares a simple three-step process for engaging with content: read, contemplate, repeat.”
X
Release notes: Release Diffusers 0.34.0: New Image and Video Models, Better torch.
[Multi-modal][Deployment]
“DeepSeek Summary: Announcement of Diffusers 0.34.0 release with new image and video models and improved PyTorch support.”
X
While I was in SF, I had a chance to present all the things in the diffusion community enabled by PyTorch at the
[Multi-modal][Infra]
“DeepSeek Summary: Sayak Paul presented diffusion community advancements enabled by PyTorch during a visit to San Francisco.”
X
I broke my own rule to never post about AI detection as it is fraught in many ways. The problem is that if you use AI a lot, you know AI writing on sight, which makes the difficulty of objectively proving that AI use to others very frustrating
[Evaluation]
“DeepSeek Summary: Mollick argues that heavy AI users can recognize AI writing intuitively, but struggle to prove it objectively, highlighting the limitations of AI detection.”
X
In 1980, the philosopher John Searle proposed a thought experiment: a person locked in a room, manipulating Chinese characters according to a
[LLM]
“DeepSeek Summary: Mollick references Searle's Chinese Room argument, likely to discuss implications for AI understanding and consciousness.”
X
Image is of the 1990s Microsoft writing assistant character Clippy with its eyebrows raised positioned in.
[LLM][Safety]
“DeepSeek Summary: Bender posts an image of Clippy, likely to comment on AI nostalgia or critique.”
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][Evaluation]
“DeepSeek Summary: Bender critiques passive language in AI discourse, emphasizing human agency.”
X
Look what @alexhanna and I got to do! (Hang out with the cool kids ...) We're talking about the Turing Test, the grandmother of all tests for AI sentience. Joining us are AI researchers Alex Hanna and Emily M. Bender
[Evaluation][Safety]
“DeepSeek Summary: Bender promotes a discussion on the Turing Test and AI sentience.”
X
New preprint! Phase transitions! We love to see them during LM training.
[LLM][Fine-tuning]
“DeepSeek Summary: Announces a new preprint about phase transitions in language model training.”
X
Life update: I'm starting as faculty at Boston University in 2026! BU has SCHEMES for LM interpretability & analysis, so I couldn't be more pumped to join a
[Evaluation][Safety]
“DeepSeek Summary: Announces new faculty position at Boston University focusing on LM interpretability.”
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 a long hiatus.”
X
Building a theory of the architecture of organizing machines and people.
[Agent]
“DeepSeek Summary: He is working on a theory for organizing both machines and people.”
X
On unquantifiable costs and inherent tradeoffs in decision theory.
[Safety]
“DeepSeek Summary: He discusses the challenges of unquantifiable costs and tradeoffs in decision theory.”
BLOG
<p>We just <a href="https://datasette.io/blog/2026/datasette-agent/">announced the first release of Datasette Agent</a>, a new extensible AI assistant for Datasette. I've been working on my <a href="https://llm.datasette.io/">LLM</a> Python library for just over three years now, and Datasette Agent...
“Datasette Agent is a new extensible AI assistant for Datasette, built on the LLM Python library. It enables natural language querying of databases and can be extended with plugins for custom workflows. This release marks a significant step in combining AI with data exploration.”
-- END OF LOG --
[STATS] 59 items · Filter applied
Powered by Horizon + DeepSeek