Intelligence.Log
2026-05-12
Extracted: 55 items. Sources: GitHub, Bluesky, X, Blogs.
++ AI OVERVIEW ++
Today's discourse was notably shaped by Simon Willison's reflections on GitLab's latest restructuring and workforce reduction, prompting a deep dive into the public employee handbooks of both GitLab and 37signals. This sparked a broader conversation about transparency in corporate downsizing and the cultural signals sent by version-controlled HR documents. Meanwhile, Willison's more cryptic post—"old woman, possibly damp, faster than an old woman should be"—has the AI community speculating about a possible reference to edge-case behavior in large language models or a playful nod to unexpected performance benchmarks. On the GitHub front, trending repos continue to emphasize developer tooling and infrastructure resilience, though no single project has yet dominated the charts today. The overall sentiment leans toward introspection on organizational scaling, with a side of whimsical technical ambiguity.
◆ Signal
Co-Starred · Last 7 days
Repos independently starred by multiple AI leaders in the week ending 2026-05-12. Stronger signal = more overlap.
antirez/ds4
×3 starrers▲ 8/10★ 8.1k
DeepSeek 4 Flash local inference engine for Metal and CUDA
grep TOPIC=
grep SOURCE=
sort --by=
GH
DeepSeek 4 Flash local inference engine for Metal and CUDA
“ds4 is a lightweight, high-performance local inference engine for DeepSeek 4 Flash, supporting both Metal (Apple Silicon) and CUDA (NVIDIA GPUs). It offers efficient model execution with minimal dependencies, making it ideal for on-device AI applications.”
GH
26m function call model that runs on incredibly small devices
“A compact 26M parameter model optimized for function calling on resource-constrained devices, enabling on-device AI execution. It leverages Gemma architecture and is designed for low-latency, privacy-preserving inference.”
GH
The `npx serve` of Apple Simulators.
“Serve-sim provides a simple CLI to run Apple Simulators headlessly, similar to `npx serve` for web servers. It enables launching iOS simulators without the Simulator.app GUI, useful for CI/CD and automated testing.”
GH
How FastFast can you pull from Hugging Face?
“A simple Python script to benchmark download speeds from Hugging Face Hub, measuring how fast you can pull models and datasets. Useful for diagnosing network performance and optimizing CI/CD pipelines.”
GH
“This plugin enables Tailscale authentication for Datasette, allowing users to restrict access to their Datasette instances to Tailscale network members. It leverages Tailscale's identity and access controls for seamless, secure sharing.”
BSKY
BSKY
Wrote about today's GitLab restructuring / "workforce reduction" announcement, and ended up digging around in version control for both the GitLab and the 37signals public employee handbooks to help illustrate my thoughts simonwillison.net/2026/May/11/...
❤️ 35 Likes|
BSKY
How are universities feeling about the agreement that Instructure reached with cyberhacker including a pinkie-swear not to use the data exfiltration for further extortion of universities, students, and faculty?
❤️ 43 Likes|[Safety]
BSKY
Open software lowered deployment cost. Open AI lowers development cost. E.g. developing a bespoke model for an enterprise use case. We’re early in companies figuring out how to leverage this successfully. www.interconnects.ai/p/how-open-m...
❤️ 20 Likes|[Deployment]
BSKY
I sat down with @jonhernandezia.bsky.social in Madrid to discuss the growing risks and impacts of AI and the urgent need to improve our social, political, and technical safeguards. Thanks for an excellent conversation! www.youtube.com/watch?v=_-Cu...
❤️ 9 Likes|[Safety]
BSKY
Expect your feed to look more and more like this in the coming weeks and months.
❤️ 71 Likes|[Agent][LLM]
BSKY
BSKY
People who are concerned about the good of humanity are always kissing up THIS HARD to fascists like Trump, right? Right??? (Also, Bernie bros, please get your guy.)
❤️ 59 Likes|
BSKY
Missed Ghost in the Machine at SIFF? You can catch it via Kinema! And, through tomorrow, participate in a fundraiser for @dairinstitute.bsky.social while you're at it! notaidoc.com/dairxghost
❤️ 9 Likes|[Safety]
BSKY
Some reflections on frequently unasked questions about stochastic parrots (the phrase and the paper): medium.com/@emilymenonb...
❤️ 76 Likes|[LLM][Safety]
BSKY
X
My most amusing interaction was where the model (I think I was given some earlier version with a
[LLM][Safety]
“DeepSeek Summary: Karpathy shares an amusing interaction with an AI model, possibly an earlier version.”
X
I'm starting to get into a habit of reading everything (blogs, articles, book chapters,…)
[Tooling]
“DeepSeek Summary: Karpathy mentions adopting a habit of extensive reading.”
X
Very interested in what the coming era of highly bespoke software might look like.
[Agent][Infra]
“DeepSeek Summary: Karpathy expresses interest in the future of customized software.”
X
2025 LLM Year in Review
[LLM][Evaluation]
“DeepSeek Summary: Karpathy shares a review of LLM developments in 2025.”
X
I've published video, slides and a detailed annotated transcript from my talk at this week's AI Engineer World's
[LLM]
“DeepSeek Summary: Simon Willison shared a talk about the last year six months in LLMs, illustrated with pelicans on bicycles.”
X
It's interesting how 'better at code' has become the defining goal of almost every AI lab over the
[LLM][Tooling]
“DeepSeek Summary: Simon Willison observes that AI labs are increasingly focused on improving code generation capabilities.”
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][Tooling]
“DeepSeek Summary: Agents require a workspace to execute code, install packages, and access files, and sandboxes fulfill this need.”
X
When building agents, you need to iterate on production data much more than when building traditional software. You need to iterate on how
[Agent][Evaluation][Deployment]
“DeepSeek Summary: Agent development requires more iteration on production data compared to traditional software.”
X
Traditional Application Performance Monitoring (APM) tools focus on metrics like latency, traffic, errors, and saturation. They track HTTP
[Agent][Evaluation][Tooling]
“DeepSeek Summary: Traditional APM tools track latency, traffic, errors, and saturation, but may not suffice for agent observability.”
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.
[Agent][Multi-modal]
“DeepSeek Summary: Defines world modeling as predicting future world states conditioned on actions.”
X
Absolutely any time I try to explore something even slightly against commonly accepted beliefs,
[Evaluation]
“DeepSeek Summary: Jeremy Howard expresses frustration that exploring ideas against common beliefs is met with resistance.”
X
hi, i'm a sole proprietor/founder in Austria and i earn many many multiples of what i'd earn as an employee, despite "predatory income tax". in fact, i opt out
[Deployment]
“DeepSeek Summary: Highlights the benefits of being a sole proprietor over an employee, even with high taxes.”
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. Here's a preview:
[Agent][Multi-modal][Tooling]
“DeepSeek Summary: Soumith outlines Thinking Machines' three-step strategy: increasing human-AI bandwidth, raising intelligence ceiling, and keeping humans central.”
X
There's a big difference between solving a problem from first principles vs applying a solution
[Evaluation]
“DeepSeek Summary: Chollet emphasizes the distinction between solving problems from first principles versus applying existing solutions.”
X
Very excited to share @theworldlabs 's latest research work RTFM!! It's a real-time,
[Multi-modal]
“DeepSeek Summary: Fei-Fei Li announces RTFM, a real-time research work from The World Labs.”
X
Just received new reach minis for the Miami office! This is the first batch was a bit late (sorry it's hard to build and ship open-source robots) but we shipped 1,000 of them last week, 1,000 this week and
[Deployment][Tooling]
“DeepSeek Summary: Clem Delangue announces receiving Reachy Mini robots for the Miami office, noting that building and shipping open-source robots is challenging, but they are ramping up production.”
X
LOL. Remove the code in the algorithm that boosts the tweets of Elon by elvodqa · Pull Request #160 ·... github.com.
[Safety]
“DeepSeek Summary: Max Woolf finds humor in a GitHub pull request that removes code boosting Elon Musk's tweets.”
X
I got to cover for the excellent @HadleyFreeman in the Guardian today so
[Agent]
“DeepSeek Summary: Phil Wang wrote a guest column for The Guardian, covering for Hadley Freeman.”
X
today i woke up to a living version of a phd student's nightmare. a new paper in my inbox: a detailed reproduction of a paper i wrote
[Evaluation]
“DeepSeek Summary: Sasha Rush woke up to a detailed reproduction of his own paper, a common PhD student nightmare.”
X
Oh it looks like I only made videos for the first half... If people start doing it I'll add more videos.
[Tooling]
“DeepSeek Summary: Sasha Rush notes he only made videos for the first half of a series and will add more if there is interest.”
X
(Thanks to @AntonAbilov who led a lot of this work)
[LLM]
“DeepSeek Summary: Sasha Rush thanks Anton Abilov for leading a significant portion of a project.”
X
Classical Jensen math. Unidirectional bandwidth is topped at 450GB/s, and then there comes a protocol overhead of two digit percentage. 1.
[Infra]
“DeepSeek Summary: Discusses bandwidth limits and protocol overhead in the context of Jensen's math.”
X
Hear, hear, I'm excited to introduce a new performance metric: Maximum Achievable Matmul
[Infra][Tooling]
“DeepSeek Summary: Introduces a new performance metric for matrix multiplication.”
X
Thanks to an awesome contribution from @omarnomad The Machine Learning Engineering Open book now can
[Tooling]
“DeepSeek Summary: Acknowledges a contribution to the Machine Learning Engineering Open Book.”
X
My presentation at the @PyTorch Conf EU is now live. It's an exciting piece given its emphasis on how we make Diffusers play quite well w/ `torch.compile`
[Deployment][Infra]
“DeepSeek Summary: Sayak Paul presented at PyTorch Conf EU about optimizing Diffusers with torch.compile.”
X
How to use Deep Research with the Gemini API
[LLM][Tooling]
“DeepSeek Summary: Philipp Schmid shares a guide on using Deep Research with the Gemini API.”
X
Guide: ReAct agent from scratch with Gemini 2.5 and LangGraph | Gemini API | Google AI for Developers
[Agent][LLM][Tooling]
“DeepSeek Summary: Philipp Schmid provides a guide to building a ReAct agent from scratch using Gemini 2.5 and LangGraph.”
X
This is going to get even worse as people realize that careful tuning in their prompts can make AI writing seem not like AI writing to readers. We expect word counts to align, in some way, with thinking & value. Writing took effort. We are not mentally ready for the alternative.
[LLM][Safety]
“DeepSeek Summary: Careful prompt tuning can make AI writing indistinguishable from human writing, challenging our assumptions about effort and value.”
X
So much work is going into faking continual learning and memory for AIs
[LLM][Evaluation]
“DeepSeek Summary: Critiques the focus on simulating continual learning and memory in AI systems.”
X
On the plus side with Opus 4.7, if it does decide to think it produces BY FAR the best
[LLM][Agent]
“DeepSeek Summary: Notes that Opus 4.7 can produce excellent results when it engages in reasoning.”
X
@emilymbender.bsky.social. emilymbender. Feb 10. Image is of the 1990s Microsoft writing assistant character Clippy with its eyebrows raised positioned in.
[Safety]
“DeepSeek Summary: Emily Bender posted an image of Clippy, the Microsoft writing assistant, with raised eyebrows.”
X
Life update: I'm starting as faculty at Boston University in 2026! BU ...
[LLM][Fine-tuning]
“DeepSeek Summary: Naomi Saphra announces she will join Boston University as faculty in 2026.”
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 her research focus on training dynamics and mechanistic interpretability in NLP models.”
X
what a perfect space for scientific discourse! I'll start off with a few images of myself
[Safety]
“DeepSeek Summary: Sarcastic comment about scientific discourse, possibly related to AI.”
X
For the first time in almost a decade, I'm teaching a class on learning and control.
[Evaluation]
“DeepSeek Summary: Ben Recht is 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 developing a theoretical framework for organizing both machines and people.”
X
With more equations than usual, I explain how policy gradient gives you a framework to randomly search for
[Fine-tuning]
“DeepSeek Summary: He explains policy gradient as a random search framework with mathematical detail.”
BLOG
Further reflections on China's high-participation, open-first AI ecosystem.
“The post explores how open model ecosystems, particularly in China, create compounding benefits through high participation and open-first strategies. It argues that openness accelerates innovation and leads to more robust AI development compared to closed systems.”
-- END OF LOG --
[STATS] 55 items · Filter applied
Powered by Horizon + DeepSeek