Intelligence.Log

2026-04-24

Extracted: 68 items. Sources: GitHub, Bluesky, X, Blogs.
++ AI OVERVIEW ++
Today's AI discourse was anchored by the release of **DeepSeek v4 Pro**, with Ethan Mollick adding it to his playable gallery and teasing its capabilities against a single-prompt challenge—having models build a 10,000-year procedurally generated harbor town simulation. Meanwhile, **Sakana Fugu** is gaining traction as a dynamic model orchestrator, with hardmaru noting its internal use for research to intelligently combine open and closed models per task. On the research front, Marc Lanctot issued a pointed public message to ICLR participants, hinting at ongoing tensions or critiques within the community. The overarching theme is a shift toward *model orchestration and agentic workflows*, moving beyond single-model benchmarks to evaluate how systems flexibly compose and execute complex, long-horizon tasks.
◆ Signal

Co-Starred · Last 7 days

Repos independently starred by multiple AI leaders in the week ending 2026-04-24. Stronger signal = more overlap.

huggingface/ml-intern
×2 starrers7/10688

🤗 ml-intern: an open-source ML engineer that reads papers, trains models, and ships ML models

|
[Agent][LLM][Tooling]
grep TOPIC=
grep SOURCE=
sort --by=
GH
nim-lang/uirelays0.0k3/10

Native Nim UI library based on the idea of "relays" which is a new fancy name for dependency injections via global callbacks. Has Windows API, X11, Cocoa and SDL 3 support. Write UI apps as easily as terminal apps!

Starred bylucidrains|[Tooling]
A native Nim UI library using a 'relays' pattern (dependency injection via global callbacks) for simplicity. Supports Windows API, X11, Cocoa, and SDL 3 backends, enabling cross-platform UI development with minimal boilerplate.
GH

Backup and restore OpenClaw agent workspaces to HF Buckets

Starred bypcuenca|[Agent][Infra]
This tool enables backup and restore of OpenClaw agent workspaces to Hugging Face Buckets, providing a simple shell-based solution for persisting agent state. It is useful for developers building agent workflows who need reliable workspace snapshots.
BSKY
sharky6000.bsky.socialMarc Lanctot

Dear @iclr-conf.bsky.social participants

❤️ 3 Likes|
BSKY
hardmaru.bsky.socialhardmaru

We’ve been using Sakana Fugu internally for our own research and coding. Instead of relying on a single model, it dynamically orchestrates the best combination of open and closed models for any task. The future of AI is collective intelligence. Excited to open the beta API: sakana.ai/fugu-beta

❤️ 15 Likes|[Agent][LLM]
BSKY
emollick.bsky.socialEthan Mollick

And here is DeepSeek v4 Pro. Added to the playable gallery as well

❤️ 0 Likes|[LLM]
BSKY
emollick.bsky.socialEthan Mollick

I had a range of AI models "build me a procedurally generated 3D simulation showing the evolution of a harbor town from 3000 BCE to 3000 AD" in one prompt. You can play the full gallery here: hg-20f7d1a3ce.netlify.app Or read my write up about GPT-5.5 here: www.oneusefulthing.org/p/sign-of-th...

❤️ 17 Likes|[Multi-modal][Agent]
BSKY
simonwillison.netSimon Willison

I had no idea The Wind in the Willows was this much of a banger

❤️ 28 Likes|
BSKY
simonwillison.netSimon Willison

OK this piece by @reckless.bsky.social about why AI is unpopular among most people (anyone who's not inflicted with "software brain") is just solid gold from start to finish

❤️ 109 Likes|[Deployment][Evaluation]
BSKY
simonwillison.netSimon Willison

DeepSeek V4 just dropped - two models, Flash and Pro, both benchmarking well, decent pelicans and prices that put them both as the cheapest in their respective categories by a solid margin simonwillison.net/2026/Apr/24/...

❤️ 127 Likes|[LLM][Evaluation][Deployment]
BSKY
simonwillison.netSimon Willison

This week's edition of my email newsletter features 4 pelicans riding bicycles, 1 possum on an e-scooter, up to 5 raccoons with ham radios hiding in crowds, 5 blog posts, 8 links, 3 quotes and a new chapter of my Agentic Engineering Patterns guide simonw.substack.com/p/gpt-55-cha...

❤️ 40 Likes|[Agent]
BSKY
markriedl.bsky.socialMark Riedl

My back deck grill is covered in baby praying mantises

❤️ 8 Likes|
BSKY
markriedl.bsky.socialMark Riedl

It's like a new-model Prius and a Cybertruck had a baby

❤️ 8 Likes|[LLM]
BSKY
markriedl.bsky.socialMark Riedl

Sunday at the ICLR Workshop on Algorithmic Fairness Across Alignment Procedures and Agentic Systems (www.afciworkshop.org/afaa-2026), I will speak on a framework for how to reconcile alignment, fairness, and bias. Long-term and near-term perspectives on "AI safety" are reconcilable.

❤️ 16 Likes|[Safety][Evaluation]
BSKY
sharky6000.bsky.socialMarc Lanctot

www.reuters.com/business/goo...

❤️ 4 Likes|[LLM][Infra]
BSKY
sharky6000.bsky.socialMarc Lanctot

I'm not just a broken record, I've broken the record player so that you can't stop hearing the same record for the next 4 days. 👀 #iclr2026 🫵

❤️ 10 Likes|
BSKY
gaelvaroquaux.bsky.socialGaël Varoquaux

#ICLR2026 paper✨️: Quantifying epistemic uncertainty of Blackbox classifiers, and link to better decisions Calibration on steroids, qualifying full prediction uncertainty with no need for Bayes, and tuning individual decisions 👇

❤️ 22 Likes|[Evaluation][Safety]
BSKY
yoshuabengio.bsky.socialYoshua Bengio

AI is advancing faster than our ability to manage it. We still have the opportunity to build the societal & technical guardrails needed to keep people, institutions and democracies safe—we shouldn't let it pass us by. Interview with @elconfidencial.bsky.social www.elconfidencial.com/tecnologia/2...

❤️ 13 Likes|[Safety]
BSKY
emollick.bsky.socialEthan Mollick

We really need a better word for the good kind of AI psychosis, the one where someone goes into a fugue state with the latest model and returns 40 days later from the mountaintop with something new.

❤️ 95 Likes|[Agent]
BSKY
emollick.bsky.socialEthan Mollick

My first two TiKZ Sparks unicorns from DeepSeek v4. Um. (Expert mode, from the DeepSeek site, which is supposed to be v4 Pro according to the release)

❤️ 36 Likes|[Multi-modal]
BSKY
nsaphra.bsky.socialNaomi Saphra

didn't even know it was lesbian visibility week until I got caught sneaking

❤️ 6 Likes|
X
My most amusing interaction was where the model (I think I was given some earlier version with a
[LLM]
“DeepSeek Summary: Karpathy recounts an amusing interaction with an earlier version of a model.
X
Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model.
[Fine-tuning]
“DeepSeek Summary: Karpathy describes leaving a model tuning run for two days.
X
The hottest new programming language is English
[LLM]
“DeepSeek Summary: Karpathy's famous quote about English as a programming language.
X
I'm being accused of overhyping the [site everyone heard too much about today already].
[Deployment]
“DeepSeek Summary: Karpathy responds to accusations of overhyping a site.
X
Vibe coding is irresponsibly building software through dice rolls, not caring what code is produced
[Agent][Tooling]
“DeepSeek Summary: Simon Willison criticizes 'vibe coding' as an irresponsible approach that ignores code quality.
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 knowing when not to intervene is a crucial skill for using coding agents.
X
hwchase17Harrison Chase
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: Agents require sandboxed workspaces for code execution and file access.
X
hwchase17Harrison Chase
In the hot path as the agent is running. The agent can decided to (or the user can prompt it to) update its memory as it is working on the core
[Agent][LLM]
“DeepSeek Summary: Agents can update memory during execution, either autonomously or via user prompt.
X
hwchase17Harrison Chase
traces matter!
[Agent][Evaluation][Tooling]
“DeepSeek Summary: Emphasizes the importance of tracing for agent observability.
X
hwchase17Harrison Chase
Agent harnesses are becoming the dominant way to build agents, and they are not going anywhere. These harnesses are intimately tied to agent
[Agent][Tooling][Deployment]
“DeepSeek Summary: Agent harnesses are the emerging standard for agent construction.
X
DrJimFanJim Fan
I've been a bit quiet on X recently. The past year has been a transformational experience.
[Agent]
“DeepSeek Summary: Jim Fan acknowledges his recent silence and describes the past year as transformational.
X
DrJimFanJim Fan
We are living in a timeline where a non-US company is keeping the original mission of OpenAI alive - truly
[Agent][Multi-modal]
“DeepSeek Summary: Jim Fan comments on a non-US company preserving OpenAI's original mission.
X
DrJimFanJim Fan
It gives me a lot of comfort knowing that we are the last generation without advanced robots everywhere.
[Agent][Multi-modal]
“DeepSeek Summary: Jim Fan expresses comfort that current generation lives before widespread advanced robotics.
X
jeremyphowardJeremy Howard
Here's what I would prefer to see:
[LLM]
“DeepSeek Summary: Jeremy expresses a preference, but the full content is not available in the snippet.
X
jeremyphowardJeremy Howard
I have some big news about FastHTML and @AnthropicAI Claude 4 :)
[LLM][Infra]
“DeepSeek Summary: Jeremy announces news about FastHTML and Anthropic's Claude 4.
X
soumithchintalaSoumith Chintala
reading "AI News" (previously Smol Talk) is probably the highest-leverage 45 mins
[LLM][Tooling]
“DeepSeek Summary: Soumith Chintala recommends reading 'AI News' as a high-leverage activity.
X
I think it's clear that for many smaller companies that invested in deep learning, it turned out
[Evaluation]
“DeepSeek Summary: Suggests deep learning investments may not have paid off for smaller companies.
X
One of the biggest misconceptions people have about intelligence is seeing it as some kind of unbounded scalar stat, like height.
[Evaluation]
“DeepSeek Summary: Critiques the view of intelligence as a single scalar metric.
X
Some personal news -- I'm leaving Google to go start a new company with a friend.
[Agent]
“DeepSeek Summary: Announced departure from Google to co-found a new company.
X
y
Yann LeCun
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, implying that such systems are not imminent.
X
y
Yann LeCun
An A.I. Pioneer Warns the Tech 'Herd' Is Marching Into a Dead End. www.nytimes.com.
[LLM]
“DeepSeek Summary: LeCun shares a New York Times article where he warns that the current AI direction is a dead end.
X
y
Yann LeCun
The emergence of superintelligence is not going to be an event. We don't have anything close to a
[Safety]
“DeepSeek Summary: LeCun argues that superintelligence will not appear suddenly and that current AI is far from it.
X
d
Fei-Fei Li
Very excited to share @theworldlabs 's latest research work RTFM!! It's a real-time, ...
[Multi-modal][Infra]
“DeepSeek Summary: Fei-Fei Li announces RTFM research from World Labs, focusing on real-time spatial intelligence.
X
C
Clem Delangue
Kimi K2.6 is a standout open-source model.
[LLM][Fine-tuning]
“DeepSeek Summary: Clem Delangue praised Kimi K2.6 as a standout open-source model, highlighting its competitive performance.
X
C
Clem Delangue
We're facing an LLM bubble, not a broader AI bubble.
[LLM][Evaluation]
“DeepSeek Summary: Clem Delangue distinguishes between an LLM-specific bubble and a general AI bubble, suggesting the former may burst.
X
minimaxirMax Woolf
me irl
“DeepSeek Summary: Max Woolf posted a short meme-like post 'me irl'.
X
srush_ioSasha Rush
⛏️
[Tooling]
“DeepSeek Summary: A short post with a pickaxe emoji, possibly indicating a new project or tool.
X
I have been compiling LLM/VLM training logbooks/chronicles. This is the one of the best sources to ...
[LLM][Infra]
“DeepSeek Summary: Stas Bekman compiles LLM/VLM training logbooks, providing a valuable resource for practitioners.
X
Thanks to an awesome contribution from @omarnomad The Machine Learning Engineering Open book now can ...
[LLM][Tooling][Fine-tuning]
“DeepSeek Summary: The Machine Learning Engineering Open Book gains a contribution from @omarnomad, enhancing its content.
X
If you were holding off to try @MSFTDeepSpeed ZeRO++ it looks like deepspeed@master should ...
[Infra][Deployment]
“DeepSeek Summary: Stas Bekman notes that DeepSpeed ZeRO++ is now ready to try on the master branch.
X
Modern art. Artist: PyTorch memory profiler Model: Llama-8B The piece on the left is the ...
[LLM][Tooling]
“DeepSeek Summary: Stas Bekman humorously compares PyTorch memory profiler output to modern art, showing memory usage of Llama-8B.
X
sayakpaulSayak Paul
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
[Deployment][Tooling]
“DeepSeek Summary: Sayak Paul reflects on his time at Hugging Face helping him identify his technical interests.
X
sayakpaulSayak Paul
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: He presented at PyTorch Conf EU on integrating Diffusers with torch.compile.
X
sayakpaulSayak Paul
RT @RisingSayak: We're shipping an elaborate guide on how to profile diffusion pipelines in Diffusers to set them
[Tooling][Deployment]
“DeepSeek Summary: He announced a guide on profiling diffusion pipelines in Diffusers.
X
philschmidPhilipp Schmid
Guide: ReAct agent from scratch with Gemini 2.5 and LangGraph | Gemini API | Google AI for Developers. ai.google.dev.
[Agent][LLM][Tooling]
“DeepSeek Summary: Philipp Schmid published a guide on building a ReAct agent from scratch using Gemini 2.5 and LangGraph.
X
e
Ethan Mollick
AI is actually pretty good at ideas as well.
[LLM][Evaluation]
“DeepSeek Summary: AI can generate creative ideas, not just analytical tasks.
X
e
Ethan Mollick
So much work is going into faking continual learning and memory for AIs,
[LLM][Safety]
“DeepSeek Summary: Critiques the trend of simulating memory and learning in AI rather than achieving true capabilities.
X
e
Ethan Mollick
My most popular AI post was a bunch of made-up 'graphs' four years ago.
[Evaluation]
“DeepSeek Summary: Reflects on how fabricated data can gain traction in AI discussions.
X
e
Emily M. Bender
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 with raised eyebrows, likely as a humorous critique of AI assistants.
X
N
Naomi Saphra
what a perfect space for scientific discourse! I'll start off with a few images of myself
[LLM]
“DeepSeek Summary: Naomi Saphra humorously comments on a space for scientific discourse, starting with images of herself.
X
N
Naomi Saphra
Life update: I'm starting as faculty at Boston University in 2026! BU ...
[LLM]
“DeepSeek Summary: Naomi Saphra announces her new faculty position at Boston University starting in 2026.
X
a
Angela Zhou
#throwback to the beginnings of a beautiful friendship =D @ansonmount @HellOnWheelsAMC #HellonWheels #onlocation.
“DeepSeek Summary: Angela Zhou posts a throwback photo with co-stars from the TV show Hell on Wheels.
X
b
Ben Recht
I weigh in on the Trump administration's newfound obsession with Gold Standard Science and reproducibility.
[Safety]
“DeepSeek Summary: Ben Recht comments on the Trump administration's focus on gold standard science and reproducibility.
X
b
Ben Recht
For the first time in almost a decade, I'm teaching a class on learning and control.
[Agent]
“DeepSeek Summary: Ben Recht announces teaching a class on learning and control after nearly ten years.
X
b
Ben Recht
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 as a framework for random search in optimization.
X
b
Ben Recht
Fully open machine learning requires not only GPU access but a community commitment to openness.
[Infra]
“DeepSeek Summary: Ben Recht argues that open ML needs both GPU access and community commitment to openness.
BLOG

<p>Chinese AI lab DeepSeek's last model release was V3.2 (and V3.2 Speciale) <a href="https://simonwillison.net/2025/Dec/1/deepseek-v32/">last December</a>. They just dropped the first of their hotly anticipated V4 series in the shape of two preview models, <a...

DeepSeek V4 preview models achieve near-frontier performance at a fraction of the cost, challenging the pricing strategies of leading AI labs. The release signals a major shift towards cost-efficient AI development, making advanced models more accessible.
-- END OF LOG --
[STATS] 68 items · Filter applied
Powered by Horizon + DeepSeek