Intelligence.Log
2026-04-22
Extracted: 69 items. Sources: GitHub, Bluesky, X, Blogs.
++ AI OVERVIEW ++
Today's discussions highlight the evolving dynamics of AI pricing and the rapid expansion of model capabilities. Simon Willison details Anthropic's swift reversal on Claude Code pricing, underscoring the volatility in commercializing AI tools. Meanwhile, Ethan Mollick showcases GPT-5.4's multimodal potential, combining data analysis with image generation to create novel visualizations. The overall theme is one of fast-paced iteration, both in business models and in the creative, practical applications of increasingly integrated AI systems.
◆ Signal
Co-Starred · Last 7 days
Repos independently starred by multiple AI leaders in the week ending 2026-04-22. Stronger signal = more overlap.
huggingface/ml-intern
×2 starrers▲ 7/10★ 688
🤗 ml-intern: an open-source ML engineer that reads papers, trains models, and ships ML models
grep TOPIC=
grep SOURCE=
sort --by=
GH
LLM inference server with continuous batching & SSD caching for Apple Silicon — managed from the macOS menu bar
“OMLX is an LLM inference server optimized for Apple Silicon with continuous batching and SSD caching for efficient memory management. It features a macOS menu bar interface for easy server management and supports OpenAI API compatibility for seamless integration.”
GH
Experiments with GitHub event data
“This repository demonstrates practical data warehousing techniques for GitHub event data using modern tools like DuckDB and dbt. It provides concrete examples of dimensional modeling and star schema implementation for analyzing GitHub activity patterns.”
BSKY
Wrote up Anthropic's self-own about Claude Code pricing from this afternoon on my blog - it turned out they'd reversed course just as I hit publish, so I've tried to update it to reflect the current state simonwillison.net/2026/Apr/22/...
❤️ 41 Likes|[Deployment]
BSKY
Okay went from “I’m not sure I have 30 minutes worth of relevant content to talk about” to “my talk will probably take 60 minutes and I need to cut it to 30”
❤️ 6 Likes|
BSKY
BSKY
BSKY
BSKY
You can combine GPT-5.4 Thinking/Pro's ability to work with data and its ability to generate images. It creates a graph, then makes an visualization. Not always perfect, but fun. Here's AI task horizons data with a bit of Basquiat, haunted by ghosts, from the Voynich manuscript, as a decaying pier.
❤️ 16 Likes|[Multi-modal][LLM]
BSKY
The new Qwen3.6-27B just gave me definitely the best pelican riding a bicycle I've had from a 16.8GB model file! simonwillison.net/2026/Apr/22/...
❤️ 137 Likes|[LLM][Multi-modal]
BSKY
Fantastic (long) piece on what AU means to mathematics (and what mathematics is) So many insights. The first begins why so may people in AI research are getting obsessed with “taste”. davidbessis.substack.com/p/the-fall-o...
❤️ 20 Likes|[Evaluation]
BSKY
Anthropic builds the arguably most powerful cybersecurity coding system ever, gets hacked. Hackers use the arguably most powerful coding system ever to build websites. This is funny.
❤️ 19 Likes|[Safety][LLM]
BSKY
Just got a very AI-generated-sounding invite to come to a reading group and discuss my book "Gone with the Wind". Maybe I should show up with a wig and stuff.
❤️ 36 Likes|[Safety]
BSKY
GPT imagegen 2: "Show me a photo taken of pages 113-114 from the books": "Eldritch Horrors as Pets: A Guide" "How Womblenauts Work" "Photographs of the People of New York Who Look Like Birds". "Cakes shaped like fish shaped like cakes"
❤️ 50 Likes|[LLM][Multi-modal]
BSKY
Every system that was regulated, either explicitly or implicitly, by the fact that they were effortful for humans (letters of recommendation, government filings, essays, or, as this paper finds, lawsuits) will break under a wave of AI.
❤️ 684 Likes|[Deployment][Safety]
BSKY
This wasn't the case with previous image generators, but the LLM you select has a huge effect on GPT-imagegen-2 output. GPT-5.4 Thinking and GPT-5.4 Pro will produce much better images, especially for complex things. This is, of course, not intuitive or explained anywhere.
❤️ 24 Likes|[LLM][Multi-modal]
BSKY
Have you been asked by a medical provider recently for consent to have an "AI" scribe record your visit? Us, too. And we have **thoughts** buttondown.com/maiht3k/arch...
❤️ 107 Likes|[Safety]
BSKY
NYC & area folks -- @alexhanna.bsky.social and I are doing our first ever LIVE TAPING of Mystery AI Hype Theater 3000 at Starr Bar in Brooklyn on Thursday April 30th Think of just how cool that crowd is going to be and you can be a part of it too! www.eventbrite.com/e/mystery-ai...
❤️ 25 Likes|[Safety]
BSKY
Building a theory of the architecture of organizing machines and people.
❤️ 14 Likes|[Agent][Infra]
X
Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model.
[Agent][Fine-tuning]
“DeepSeek Summary: Karpathy shared an experiment where he left an autoresearch system tuning nanochat for approximately two days on a depth=12 model, indicating ongoing work with automated AI research processes.”
X
I'm being accused of overhyping the [site everyone heard too much about today already].
[LLM]
“DeepSeek Summary: Karpathy addresses criticism about potentially overhyping a trending AI-related site, showing engagement with community feedback and hype cycles in AI.”
X
We're missing (at least one) major paradigm for LLM learning. Not sure what to call it,
[LLM][Fine-tuning]
“DeepSeek Summary: Karpathy identifies a gap in current LLM learning paradigms, suggesting there's an important missing approach that hasn't been named or fully developed yet.”
X
Vibe coding is irresponsibly building software through dice rolls, not caring what code is produced
[Agent][Tooling]
“DeepSeek Summary: Critiques 'vibe coding' as irresponsible software development that relies on random chance rather than deliberate design.”
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: Suggests that effective use of AI coding agents requires knowing when not to use them, not just how to use them.”
X
Open Models, Open Runtime, Open Harness - Building your own AI agent with LangChain and
[Agent][LLM][Tooling]
“DeepSeek Summary: Discusses building AI agents using open-source models, runtime, and harness tools like LangChain.”
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][Infra]
“DeepSeek Summary: Emphasizes that data collection is simple compared to the challenge of meaningful analysis and interpretation of observations.”
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: Argues that AI agents require dedicated workspaces or sandboxes to execute code, manage dependencies, and handle files securely.”
X
The first time I met Jensen was also the first time I met @elonmusk. I was interning at OpenAI that day and
“DeepSeek Summary: Jim Fan shares a personal anecdote about meeting Jensen Huang and Elon Musk during his OpenAI internship, highlighting early connections in the AI industry.”
X
Resource constraints are a beautiful thing. Survival instinct in a cut-throat AI competitive land
[Infra]
“DeepSeek Summary: Fan frames resource limitations as a positive driver for innovation and survival in the highly competitive AI field.”
X
I've been a bit quiet on X recently. The past year has been a transformational experience.
“DeepSeek Summary: Fan notes a period of reduced public activity, attributing it to a significant personal or professional transformation over the past year.”
X
It gives me a lot of comfort knowing that we are the last generation without advanced robots everywhere.
[Agent][Deployment]
“DeepSeek Summary: Expresses a unique sense of comfort in living during a transitional period before ubiquitous advanced robotics.”
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][LLM]
“DeepSeek Summary: Provides a technical definition for 'world modeling' in AI, framing it as the prediction of future states based on actions.”
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.
[LLM][Evaluation]
“DeepSeek Summary: Demonstrates Grok's behavior when asked about a complex geopolitical topic, showing it searches for Elon Musk's opinions first.”
X
Here's what I would prefer to see:
“DeepSeek Summary: Jeremy Howard expresses a preference for how something should be presented or implemented.”
X
Useful thread to read for folks that haven't done enterprise deployments before.
[Deployment][Infra]
“DeepSeek Summary: Shares a resource thread about enterprise AI deployments for those new to the topic.”
X
Soumith Chintala needs no introduction in the ML world — his insights are incredibly accessible across Twitter, LinkedIn, podcasts, and
[LLM]
“DeepSeek Summary: Soumith Chintala is a well-known figure in machine learning who shares his expertise through multiple accessible platforms.”
X
reading 'AI News' (previously Smol Talk) is probably the highest-leverage 45 mins
[LLM][Tooling]
“DeepSeek Summary: Soumith Chintala recommends 'AI News' (formerly Smol Talk) as a highly efficient use of time for staying informed.”
X
I think it's clear that for many smaller companies that invested in deep learning, it turned out...
[Deployment][Evaluation]
“DeepSeek Summary: Chollet suggests that deep learning investments haven't paid off for many smaller companies, implying practical limitations or implementation challenges.”
X
10/ Sakana AI Co-Founder and CEO, David Ha, will be giving a talk at the #ICLR2025 “World Models” Workshop, at a panel to discuss the
[Agent][Multi-modal]
“DeepSeek Summary: David Ha will present at the ICLR 2025 World Models Workshop, indicating his active involvement in AI research conferences.”
X
On stage at Imagination In Action's AI Summit in Davos with John Werner, founder and CEO of Imagination In Action, Yann LeCun discusses the...
[Agent]
“DeepSeek Summary: Yann LeCun participated in a discussion at the Imagination In Action AI Summit in Davos, focusing on AI topics with the event's founder.”
X
Very excited to share @theworldlabs 's latest research work RTFM!! It's a real-time, [content truncated in snippet]
[Multi-modal][Tooling]
“DeepSeek Summary: Announcing new research from The World Labs focused on real-time capabilities, likely related to multimodal AI or spatial intelligence.”
X
🤗, the open and collaborative platform for AI builders. Great research on open-source by: - $4.15B invested in open-source generates $8.8T of value for companies (aka $1 invested in open-source = $2,000 of value created) - Companies would need to spend 3.5 times more on software than they currently do
[Infra][Tooling]
“DeepSeek Summary: Open-source AI investment generates massive economic value, with $1 invested creating $2,000 in value for companies.”
X
I think open source comes in as a way to create more competition... Hugging Face co-Founder & CEO Clément Delangue said on Tuesday that he believes open source AI will create 'healthy competition' among AI
[Infra][Tooling]
“DeepSeek Summary: Open-source AI fosters healthy competition in the AI industry by lowering barriers to entry.”
X
Clem Delangue, CEO of Hugging Face, is making open-source AI practical, scalable, and ethical. See why he's one of BigDATAwire's 2026 People
[Infra][Safety][Deployment]
“DeepSeek Summary: Focus on making open-source AI not just accessible but also practical, scalable, and ethical for real-world use.”
X
Wager established. Jonathan Frankle (@jefrankle) stepped up to my Transformer long bet.
[LLM]
“DeepSeek Summary: Sasha Rush established a wager or bet related to Transformers with Jonathan Frankle, indicating a significant technical debate or prediction in the field.”
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, describing it as a small, ambitious team, hinting at their new role and the company's innovative direction.”
X
Congrats to the @cursor_ai team on the launch of Composer 2! We are proud to see Kimi-k2.5 provide the foundation.
[Tooling][Infra]
“DeepSeek Summary: Sasha Rush congratulated the Cursor team on launching Composer 2, highlighting that Kimi-k2.5 serves as its foundation, indicating technical integration or collaboration.”
X
I have been compiling LLM/VLM training logbooks/chronicles. This is one of the best sources to...
[LLM][Evaluation]
“DeepSeek Summary: Stas Bekman is compiling training logbooks/chronicles for LLM/VLM models, which he considers a valuable resource.”
X
Thanks to an awesome contribution from @omarnomad The Machine Learning Engineering Open book now can...
[Tooling][Deployment]
“DeepSeek Summary: Acknowledgement of a contribution to the Machine Learning Engineering Open book project.”
X
If you were holding off to try @MSFTDeepSpeed ZeRO++ it looks like deepspeed@master should...
[Infra][Tooling]
“DeepSeek Summary: Suggests that DeepSpeed ZeRO++ is now ready to try via the master branch.”
X
Modern art. Artist: PyTorch memory profiler Model: Llama-8B The piece on the left is the...
[LLM][Tooling]
“DeepSeek Summary: Uses artistic metaphor to describe PyTorch memory profiling output for Llama-8B model.”
X
Working at Hugging Face over the past 3.5+ years has allowed me to identify what technical areas truly interest me! In turn, that has allowed me to directly
[Infra][Tooling]
“DeepSeek Summary: Reflection on career growth at Hugging Face leading to clarity on technical interests.”
X
Blog ... Philipp Schmid profile. Philipp Schmid. ✓ Affiliate badge. _philschmid.
[LLM]
“DeepSeek Summary: The user posted a link to a blog post, indicated by the snippet mentioning 'Blog' and their profile details.”
X
My most popular AI post was a bunch of made-up "graphs" four years ago.
[Evaluation]
“DeepSeek Summary: Reflects on how AI content, even if fabricated, can gain significant traction and popularity over time.”
X
As stories about AI increasingly become stories of either catastrophe or salvation,
[Safety]
“DeepSeek Summary: Points out the polarized narrative around AI, often framed as either a dire threat or a miraculous solution.”
X
Classic study gave 146 economist teams the same dataset & got wildly different answers New
[Evaluation]
“DeepSeek Summary: Illustrates how even with identical data, interpretations and conclusions can vary dramatically among experts.”
X
Image is of the 1990s Microsoft writing assistant character Clippy with its eyebrows raised positioned in.
[Agent][Evaluation]
“DeepSeek Summary: Emily M. Bender references the Microsoft Clippy character, likely drawing a parallel to contemporary AI assistants or critiquing anthropomorphic design in technology.”
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 or framing around AI outcomes, emphasizing that decisions (and thus responsibility) lie with developers or deployers, not with the technology itself.”
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][LLM]
“DeepSeek Summary: Bender mentions participating in a discussion about the Turing Test, framing it historically as the 'grandmother of all tests for AI sentience,' likely to critique or contextualize its ongoing relevance.”
X
This book starts like it's gonna be a fun microhistory of TB (it gave us the Stetson!)
“DeepSeek Summary: Naomi Saphra comments on a book about tuberculosis, noting its engaging start as a microhistory.”
X
New preprint! Everyone loves causal interp. It's coherently defined! It makes testable predictions
[LLM][Evaluation]
“DeepSeek Summary: Announces a new preprint on causal interpretability, highlighting its clear definitions and testable predictions.”
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][Fine-tuning][Evaluation]
“DeepSeek Summary: Describes her research focus on NLP training dynamics and the emergence of structures and mechanistic behaviors.”
X
I weigh in on the Trump administration's newfound obsession with Gold Standard Science and reproducibility.
[Evaluation]
“DeepSeek Summary: Ben Recht comments on the Trump administration's focus on scientific reproducibility and gold standard science.”
X
For the first time in almost a decade, I'm teaching a class on learning and control.
[Agent]
“DeepSeek Summary: Ben Recht announces he's teaching a class on learning and control after nearly ten years.”
X
With more equations than usual, I explain how policy gradient gives you a framework to randomly search for
[Agent]
“DeepSeek Summary: Ben Recht explains policy gradient methods as a framework for random search in reinforcement learning.”
X
Fully open machine learning requires not only GPU access but a community commitment to openness.
[Infra]
“DeepSeek Summary: Ben Recht argues that open machine learning requires both hardware access and community commitment to openness.”
BLOG
<p>Anthropic today quietly (as in <em>silently</em>, no announcement anywhere at all) updated their <a href="https://claude.com/pricing">claude.com/pricing</a> page (but not their <a href="https://support.claude.com/en/articles/11049762-choosing-a-claude-plan">Choosing a Claude plan page</a>, which...
“Anthropic made unannounced pricing changes to Claude Code, creating confusion about potential costs. The author analyzes the discrepancies between different official pages to clarify the actual pricing structure.”
-- END OF LOG --
[STATS] 69 items · Filter applied
Powered by Horizon + DeepSeek