Intelligence.Log
2026-04-14
Extracted: 51 items. Sources: GitHub, Bluesky, X, Blogs.
++ AI OVERVIEW ++
Today's discourse centers on the accelerating and disruptive nature of AI progress. Ethan Mollick highlights a shift from gradual improvement to "large discrete jumps" in economically critical capabilities with each new model release, suggesting industries should brace for sudden, significant impacts. This theme of step-change advancement is mirrored in trending repositories, where developers are rapidly prototyping tools that leverage these emerging abilities, particularly in code generation and autonomous agent workflows. The community is actively debating how to practically integrate and govern these powerful, leapfrogging systems.
grep TOPIC=
grep SOURCE=
sort --by=
GH
A CLI to estimate inference memory requirements for Hugging Face models, written in Python.
“This CLI tool provides memory estimation for Hugging Face model inference, helping developers plan resource allocation. It supports GGUF and SafeTensors formats, offering practical insights for deployment.”
BSKY
Soon, at each release of AI along the current capability curve, you will start to see large discrete jumps in ability in economically important areas, because the previous AI ability level in some aspect of the job bottlenecked progress. When bottlenecks are released, it looks like a leap forward.
❤️ 8 Likes|[Deployment][Evaluation]
BSKY
Learn about the "AI-as-Amplifier Paradox" at #CHI2026. Skill amplification? Or skill erosion? Or both? (CHI Honorable Mention Paper)
❤️ 21 Likes|[Safety]
BSKY
BSKY
One of my key strategies with Interconnects is to develop the practice of making my work obviously compelling to a wider audience, keeping them hooked over time and wondering what I'm up to, etc. www.interconnects.ai/p/what-ive-b...
❤️ 7 Likes|[Deployment]
BSKY
Excited to launch the accompanying free RLHF Course for my book. To kick it off, I've released: - Welcome video - Lecture 1: Overview of RLHF & Post-training - Lecture 2: IFT, Reward Models, Rejection Sampling - Lecture 3: RL Math - Lecture 4: RL Implementation Landing page: rlhfbook.com/course
❤️ 64 Likes|[Fine-tuning][Evaluation]
BSKY
AI keeps getting better but the last time the shape of the jagged frontier changed radically was o1 & the Reasoner. A good mental model of the coming months is that models get extremely good at the things they are already quite good at (coding), but weaknesses will be similar (long form fiction)
❤️ 38 Likes|[LLM][Evaluation]
BSKY
Interesting: "Currently, 38% of Americans live within 5 miles of at least one operational data center... Living near a data center doesn’t have much of an effect on public opinion about the facilities." From now on, it looks like most DCs will be rural, though. www.pewresearch.org/short-reads/...
❤️ 32 Likes|[Infra]
BSKY
Time to #TalkAboutHumanities -- Linguistics is the study of how language works and how we work with language, and linguists end up very sensitized to language use and how it shapes our social world.
❤️ 34 Likes|[LLM][Safety]
BSKY
I heard a reporter from Axios interviewed on NPR the other day (Marketplace Tech, I think) talking about how the tech companies are putting out new models every 6 months to 1 year and how each model is more "powerful" than the previous. 🧵>>
❤️ 36 Likes|[Evaluation]
BSKY
I voluntarily read plenty of LLM output, but there should be consent. My default assumption when I read text is it reflects a human's thoughts, and it gives me the ick to realize half a line in that it doesn't.
❤️ 30 Likes|[LLM][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: Karpathy is shifting focus from coding to using LLMs to build and manage personal knowledge bases for research, indicating a move towards knowledge compounding and organization.”
X
LLMs are emerging as a new kind of intelligence, simultaneously a lot smarter than I expected and a lot dumber than I expected.
[LLM][Evaluation]
“DeepSeek Summary: Karpathy expresses the dual-nature surprise of LLM capabilities, acknowledging both their advanced and surprisingly limited aspects.”
X
Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model.
[Fine-tuning][Agent][Tooling]
“DeepSeek Summary: Karpathy is experimenting with automated research and fine-tuning processes for a smaller model (nanochat), indicating hands-on work in model optimization.”
X
Screenshot from a video game where a team of raccoons go on a heist
[Multi-modal]
“DeepSeek Summary: Simon Willison shares an AI-generated image prompt result showing creative multi-modal AI capabilities.”
X
It's interesting how 'better at code' has become the defining goal of almost every AI lab over the...
[Agent][Evaluation]
“DeepSeek Summary: Willison observes the AI industry's intense focus on improving coding capabilities as a primary benchmark for progress.”
X
reading 'AI News' (previously Smol Talk) is probably the highest-leverage 45 mins
[LLM]
“DeepSeek Summary: Soumith Chintala recommends 'AI News' (formerly Smol Talk) as a highly valuable 45-minute investment for staying informed about AI developments.”
X
Open LLMs need to get organized and co-ordinated about sharing human feedback.
[LLM][Evaluation]
“DeepSeek Summary: Soumith Chintala calls for better organization and coordination among open LLM projects regarding the sharing of human feedback data.”
X
MacStudio you ask? Apple Engineering's actual time spent on PyTorch support
[Infra][Deployment]
“DeepSeek Summary: Soumith Chintala comments on Apple's engineering investment in PyTorch support, likely in the context of Mac Studio hardware.”
X
Sometimes we forget that NVIDIA wins because it's a software company.
[Infra][Tooling]
“DeepSeek Summary: Soumith Chintala reminds that NVIDIA's success is fundamentally driven by its software capabilities, not just hardware.”
X
I think it's clear that for many smaller companies that invested in deep learning, it turned out
[Evaluation][Deployment]
“DeepSeek Summary: Chollet suggests that deep learning investments haven't paid off for many smaller companies, implying practical limitations or misalignment with business needs.”
X
David Ha @hardmaru and team are super practical scientifically research driven geniuses . And this is amazing to see
[Evaluation]
“DeepSeek Summary: A third party praises David Ha (@hardmaru) and his team for being practical, scientifically research-driven geniuses.”
X
Impressive model based on a few minutes of playing, but disappointing to see no mention at all of a model card, red teaming, yesterday's incident,
[Safety][Evaluation]
“DeepSeek Summary: Max Woolf critiques a new AI model for lacking proper documentation (model card) and safety testing (red teaming), while also referencing a recent incident.”
X
Stand-up comedy's gain is game criticism's loss: comedian @PhilNWang speaks with clarity and insight about the five video games he would like to
“DeepSeek Summary: Phil Wang discusses video games with clarity and insight, suggesting his transition from game criticism to comedy.”
X
Phil Wang // Insta: @wangpix (@PhilNWang). 324 likes 24 replies. I got to cover for the excellent @HadleyFreeman in the Guardian today so
“DeepSeek Summary: Phil Wang mentions covering for Hadley Freeman at The Guardian, indicating journalistic work.”
X
Some personal news: I recently joined Cursor. Cursor is a small, ambitious team, and they've created...
[Tooling]
“DeepSeek Summary: Sasha Rush announces joining Cursor, describing it as a small, ambitious team.”
X
If you were holding off to try @MSFTDeepSpeed ZeRO++ it looks like deepspeed@master should
[Infra][Deployment]
“DeepSeek Summary: Suggests that DeepSpeed's ZeRO++ feature is now available in the master branch and worth trying.”
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 potential issue when experimenting with FA4 (likely FlashAttention 4) related to loading cutlass.cute.”
X
Thanks to an awesome contribution from @omarnomad The Machine Learning Engineering Open book now can
[Tooling]
“DeepSeek Summary: Acknowledges a contribution that enhanced the 'Machine Learning Engineering Open book'.”
X
Classical Jensen math. Unidirectional bandwidth is topped at 450GB/s, and then there comes a protocol overhead of two digit percentage.
[Infra]
“DeepSeek Summary: Discusses a hardware bandwidth limit (450GB/s) and significant protocol overhead affecting performance.”
X
Had a nice time chatting about the state of diffusion models and some text-to-image data shenanigans at
[Multi-modal]
“DeepSeek Summary: Sayak Paul discusses the current state of diffusion models and text-to-image data issues.”
X
X
Giving a talk here is by far the most fulfilling experience of my life!
“DeepSeek Summary: Sayak Paul expresses that giving a talk was an extremely fulfilling personal experience.”
X
How to use Gemma 4 with the Gemini API and Google AI Studio. www.philschmid.de.
[LLM][Tooling]
“DeepSeek Summary: A technical guide on integrating Gemma 4 with Google's Gemini API and AI Studio platform.”
X
Random thought. We are going to be so much faster at creating and building.
“DeepSeek Summary: An optimistic reflection on the accelerating pace of innovation and development capabilities.”
X
In discussions of AI and jobs, we put too much emphasis on the technology and not enough on...
[Deployment]
“DeepSeek Summary: Critiques the overemphasis on AI technology itself in job discussions, suggesting other factors are being overlooked.”
X
As stories about AI increasingly become stories of either catastrophe or salvation,...
[Safety]
“DeepSeek Summary: Observes that AI narratives are polarizing into extremes of doom or utopia, missing more balanced discussions.”
X
As someone involved in academic research on AI, it is notable to me that most of the key...
[Evaluation]
“DeepSeek Summary: Notes from an academic research perspective that key developments in AI are happening outside traditional academic institutions.”
X
So much work is going into faking continual learning and memory for AIs,...
[LLM][Tooling]
“DeepSeek Summary: Comments on the significant engineering effort being devoted to simulating continuous learning and memory capabilities in AI systems.”
X
For those playing along at home, here's a 'AI is sentient!' argument bingo card.
[LLM][Safety][Evaluation]
“DeepSeek Summary: Emily M. Bender shares a bingo card mocking common arguments about AI sentience, highlighting her critical stance on AI hype.”
X
This is infuriating and also was totally predictable. Thank you @daveyalba
[Safety][Evaluation][Deployment]
“DeepSeek Summary: Emily M. Bender expresses frustration over a predictable issue in AI, likely related to ethics or misuse, acknowledging another user's contribution.”
X
If you still don't know what you're doing Friday at #ICML2024, I'm going to suggest our HiLD
[LLM][Evaluation]
“DeepSeek Summary: Promoting a HiLD event at ICML 2024 conference, indicating involvement in machine learning research community.”
X
I'll meet you at this button.
[Agent]
“DeepSeek Summary: Brief, possibly cryptic tweet that could reference a specific location or inside joke within the ML community.”
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 research focus on NLP model training dynamics and emergent mechanistic behaviors in language models.”
X
For the first time in almost a decade, I'm teaching a class on learning and control.
[Evaluation]
“DeepSeek Summary: Ben Recht is returning to teaching a course on learning and control after nearly ten years.”
X
Part one of a new blog series: using the discovery of vitamins as a parable for why replication crises in [science/ML] matter.
[Evaluation]
“DeepSeek Summary: Introduces a blog series using the history of vitamin discovery as an analogy to discuss the importance of replication in scientific and machine learning research.”
X
Revisiting Sutton's Bitter Lesson in the wake of GPT-5.
[LLM][Evaluation]
“DeepSeek Summary: Re-examines the 'Bitter Lesson'—the idea that general methods leveraging computation scale best—in the context of advanced models like GPT-5.”
BLOG
What I've been up to!
“The post offers a personal update on Nathan Lambert's multifaceted contributions to AI/ML, including the ATOM Report for technical insights, a post-training course for practical education, and his book for broader dissemination of knowledge. It highlights the importance of bridging research, education, and community engagement in advancing the field.”
-- END OF LOG --
[STATS] 51 items · Filter applied
Powered by Horizon + DeepSeek