Intelligence.Log
2026-05-18
Extracted: 48 items. Sources: GitHub, Bluesky, X.
++ AI OVERVIEW ++
The AI research community is navigating a fragmented landscape, as Ethan Mollick notes that his Bluesky feed has grown quieter—not because tensions have eased, but because automated blocking lists have effectively siloed conversations. Meanwhile, a prescient two-month-post-ChatGPT prediction is being recirculated as a daily reminder of how the field’s trajectory remains eerily on track. On GitHub, the star charts reflect a continued focus on open-weight model fine-tuning and agentic tooling, with several new repositories for structured output parsing and long-context inference gaining traction. Naomi Saphra’s lighthearted but pointed observation about always reading YouTube comments underscores a broader sentiment: the most raw and unfiltered feedback on AI tools often comes from unexpected corners of the web. Overall, the day’s signals point to a community both self-aware of its echo chambers and hungry for ground-level signals amid the noise.
◆ Signal
Co-Starred · Last 7 days
Repos independently starred by multiple AI leaders in the week ending 2026-05-18. Stronger signal = more overlap.
antirez/ds4
×2 starrers▲ 8/10★ 9.0k
DeepSeek 4 Flash local inference engine for Metal and CUDA
grep TOPIC=
grep SOURCE=
sort --by=
GH
“This repository provides examples for mounting Hugging Face model caches, enabling efficient reuse of downloaded models across environments. It focuses on Shell scripts for setup and configuration.”
BSKY
BlueSky AI conversations have gotten less heated recently* * because much of this site has blocked me via automated lists so I have no contact with large parts of this social network, which isn’t necessarily a good thing, though it does make for nice echo chambers, which are pleasant, at least.
❤️ 85 Likes|
BSKY
Most prophetic tweet of all time (2 months post ChatGPT release by a member of the AI technical staff). And you can safely repost it every day and it will still be prophetic for the future. This is the least the world will care about AI.
❤️ 78 Likes|[LLM]
BSKY
BSKY
Musk loses court battle with OpenAI on the grounds that the statute of limitations had passed. The 🍿 was good while it lasted www.cnbc.com/2026/05/18/m...
❤️ 6 Likes|
BSKY
In a Turing Test of sorts, it looks like a 100% AI generated story just won the Commonwealth Prize for the Caribbean region "for its lyrical precision and haunting atmosphere, the story stood out for the confidence and restraint of its voice." Published in Granta: granta.com/the-serpent-...
❤️ 127 Likes|[LLM][Evaluation]
BSKY
One thing to watch for with Claude & GPT is that the models expose too much irrelevant history in their outputs. Slides are given footers saying things like "Better, more targeted version" if you asked for a better version, documents make references to how they are improved, etc.
❤️ 92 Likes|[LLM]
BSKY
Absolutely horrified at this planned study from my own institution: www.404media.co/researchers-... >>
❤️ 396 Likes|[Safety]
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][Tooling]
“DeepSeek Summary: Using LLMs to build personal knowledge bases for research topics, shifting token usage from code to knowledge manipulation.”
X
A few random notes from claude coding quite a bit last few weeks.
Coding workflow. Given the latest lift in LLM coding capability, like many others I rapidly went from about 80% manual+autocomplete coding and 20% agents in November to 80% agent coding and 20% edits+touchups in
[Agent][Tooling][Deployment]
“DeepSeek Summary: Shift from 80% manual coding to 80% agent coding due to improved LLM coding capabilities.”
X
Judging by my tl there is a growing gap in understanding of AI capability. The first issue I think is around recency and tier of use. I think a lot of people tried the free tier of ChatGPT somewhere last year and allowed it to inform their views on AI a little too much. People on X are the first to know.
[LLM][Evaluation]
“DeepSeek Summary: Highlights a gap in AI understanding due to outdated or limited use of AI tools.”
X
It's interesting how "better at code" has become the defining goal of almost every AI lab over the
[LLM][Evaluation]
“DeepSeek Summary: Simon observes that AI labs are prioritizing code generation as the primary benchmark for model improvement.”
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]
“DeepSeek Summary: Agents require a sandboxed workspace to execute code and manage dependencies.”
X
Your harness, your memory ... The “best” way to build agentic systems has changed dramatically over the past three years. When ChatGPT came out,
[Agent][LLM]
“DeepSeek Summary: The approach to building agentic systems has evolved significantly since ChatGPT's release.”
X
Today we're launching LangChain Labs, a new applied research effort focused on Continual Learning. Our goal is to advance open,
[Agent][Tooling]
“DeepSeek Summary: Announcement of LangChain Labs, an applied research initiative for continual learning.”
X
The Second Pre-training Paradigm
[Multi-modal][Fine-tuning][Agent]
“DeepSeek Summary: Jim Fan discusses a new pre-training paradigm, likely related to robotics or AI.”
X
Here's a complete unedited video of asking Grok for its views on the Israel/Palestine situation. It first searches twitter for what Elon thinks.
[Safety][Evaluation]
“DeepSeek Summary: Jeremy Howard demonstrates Grok's behavior by asking about Israel/Palestine, showing it searches for Elon Musk's views first.”
X
My 4yo daughter (and her stuffed animals) have just started learning to code. I'm amazed at how great the learning tools are for pretty much everything she wants to learn. And she's having such a great time!
[Tooling]
“DeepSeek Summary: Jeremy Howard shares that his 4-year-old daughter is learning to code with effective tools.”
X
@Mojo_flyin @dwarkesh_sp i'd be excited to have Google and AMD create real ecosystems. I fail to see why Jensen has a burden there. There should be multiple strong American AI ecosystems at the hardware layer. Historically, I struggled to make a dent in conversations around ecosystems with many
[Infra]
“DeepSeek Summary: Chintala advocates for multiple strong American AI hardware ecosystems, including Google and AMD, and dismisses the idea that Jensen (Nvidia) has a burden to support them.”
X
Thinky's secret plan:
1: Increase Human<->AI bandwidth
2: Raise ceiling of human+AI intelligence
3: Help humans continue as main-characters in the new world
We are at Step 1.
Interaction Models are great real-time collaborative tools for humans.
Here's a preview:
[Multi-modal][Agent]
“DeepSeek Summary: Chintala outlines Thinking Machines Lab's vision: increasing human-AI bandwidth, raising human+AI intelligence ceiling, and keeping humans central. They are at Step 1 with Interaction Models.”
X
Current AI is a librarian of existing knowledge. Science requires an explorer of the unknown.
[Evaluation]
“DeepSeek Summary: Chollet contrasts current AI's role as a librarian of existing knowledge with the need for an explorer of the unknown in science.”
X
Folks who work in AI or software engineering feel like the world is changing exponential fast.
[Deployment]
“DeepSeek Summary: Chollet notes the perception among AI and software engineers that the world is changing at an exponential pace.”
X
The 3rd edition of my book Deep Learning with Python is being printed right now, and will be in bookstores within 2 weeks. The problem with Facebook is not *just* the loss of your privacy and the fact that it can be used as a totalitarian panopticon.
[Fine-tuning]
“DeepSeek Summary: Chollet announces the upcoming 3rd edition of his book and comments on Facebook's issues beyond privacy loss.”
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 RTFM, a real-time research work from World Labs.”
X
We're launching the agentic robotics app store today. Let's democratize AI robotics for all! 300+ apps shipped. 10,000 robots in the wild. It used to take weeks from a robotics engineer to build apps, now everyone can do it in hours with ML intern or your favorite neighborhood
[Agent][Deployment][Tooling]
“DeepSeek Summary: Hugging Face launches an app store for agentic robotics, enabling rapid app development by non-experts.”
X
Local open-weight AI on a laptop has been improving more than twice as fast as Moore's Law! Between May 2024 and May 2026, the most expensive MacBook Pro you could buy stayed at 128 GB of unified memory. The hardware ceiling barely moved. But the smartest open-weight model
[LLM][Infra][Deployment]
“DeepSeek Summary: Open-weight AI models on laptops are improving at over twice the rate of Moore's Law, outpacing hardware constraints.”
X
congrats to OpenAI on winning the Turing Test
[LLM][Safety]
“DeepSeek Summary: Sarcastic or ironic congratulations to OpenAI.”
X
Wager established. Jonathan Frankle (@jefrankle) stepped up to my Transformer long bet.
[LLM]
“DeepSeek Summary: Sasha Rush made a long bet with Jonathan Frankle about Transformers.”
X
Some personal news: I recently joined Cursor. Cursor is a small, ambitious team, and they've created...
[Tooling]
“DeepSeek Summary: Sasha Rush announced joining Cursor, a small ambitious team.”
X
Been reflecting a bit on the Harvard news. This paper from 2017 was ... Didn't realize at the time how lucky for us Americans to work with incredible people from around the world.
[Safety]
“DeepSeek Summary: Reflection on Harvard news and gratitude for international collaboration.”
X
When dealing with tight gpu memory situations try to change a PyTorch version to both newer and older and it might just do the trick. I get massively different memory usage patterns with pt 2.4 to 2.8. e.g. today I tried last night's nightly and it was using 6GB less memory
[Infra][Fine-tuning]
“DeepSeek Summary: Changing PyTorch version can significantly reduce GPU memory usage, with differences of up to 6GB between versions.”
X
I have been compiling LLM/VLM training logbooks/chronicles. This is the one of the best sources to ...
[LLM]
“DeepSeek Summary: Stas Bekman compiles training logbooks for LLM/VLM, providing valuable resources.”
X
Thanks to an awesome contribution from @omarnomad The Machine Learning Engineering Open book now can ...
[Tooling][LLM]
“DeepSeek Summary: The ML Engineering Open Book receives contributions to expand its content.”
X
This is a long overdue section of the ML Engineering Understanding Training Loss Patterns ...
[Fine-tuning][Evaluation]
“DeepSeek Summary: New section on understanding training loss patterns added to ML Engineering resources.”
X
The kernels project at Hugging Face has been growing! We want it to be the go-to place for kernel devs and kernel users. We're looking to work w/ folks who're interested in doing agentic kernel dev, providing real optim value to real models. Reach out if interested :)
[Infra][Deployment][Tooling]
“DeepSeek Summary: Sayak Paul announces growth of Hugging Face kernels project and seeks contributors for agentic kernel development.”
X
1. Read the post. 2. Contemplate. 3. Repeat 1.
[Evaluation]
“DeepSeek Summary: Sayak Paul shares a brief, reflective post about reading and contemplation.”
X
Google DeepMind and Korea Partner to Accelerate Scientific Discovery. deepmind.google.
[Deployment]
“DeepSeek Summary: Announcement of partnership between Google DeepMind and Korea to accelerate scientific discovery.”
X
I think the adaptive thinking requirement in Claude Opus 4.7 is bad in the ways that all AI effort routers are bad, but magnified by the fact that there is no manual override like in ChatGPT. It regularly decides that non-math/code stuff is "low effort" & produces worse results.
[LLM][Evaluation]
“DeepSeek Summary: Ethan Mollick criticizes Claude Opus 4.7's adaptive thinking for misclassifying non-math/code tasks as low effort, leading to worse results.”
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][Safety]
“DeepSeek Summary: Mollick humorously references the Chinese Room thought experiment with a physical implementation using GPT-1.”
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
[Safety][Evaluation]
“DeepSeek Summary: Bender and Hanna discuss the Turing Test as the foundational test for AI sentience.”
X
For those playing along at home, here's a "AI is sentient!" argument bingo card.
[Safety]
“DeepSeek Summary: Bender creates a bingo card for common arguments claiming AI sentience.”
X
what a perfect space for scientific discourse! I'll start off with a few images of myself
[Safety]
“DeepSeek Summary: Sarcastic comment about using images of oneself for scientific discourse.”
X
Life update: I'm starting as faculty at Boston University in 2026! BU ...
[Evaluation]
“DeepSeek Summary: Announces starting as faculty at Boston University in 2026.”
X
Revisiting Sutton's Bitter Lesson in the wake of GPT-5.
[LLM]
“DeepSeek Summary: Ben Recht revisits Sutton's Bitter Lesson in the context of GPT-5, likely discussing the implications of scaling and general methods over specialized knowledge.”
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
I have a recommended reading list for Artificial Intelligence, and it hasn't changed since 2019.
[LLM]
“DeepSeek Summary: Ben Recht notes that his recommended AI reading list remains unchanged since 2019.”
-- END OF LOG --
[STATS] 48 items · Filter applied
Powered by Horizon + DeepSeek