not even a month ago
Who's Who of AI
David Marx
Why they matter
Directory member with public evidence across Culture, work & education.
- AI signals
- 23 past 30d
- Sources
- 11 distinct domains
- Discussions
- 117 past 30d
- Latest signal
- 4d ago
I read a lot of research. Mostly ML.
Currently reading: https://dmarx.github.io/papers-feed/
Statistical Learning
Information Theory
Ontic Structural Realism
Morality As Cooperation
Free Culture, Open Access
YIMBY, UBI
Research MLE
Frmr FireFighter
What they're sharing
Statement on the US government directive to suspend access to Fable 5 and Mythos 5 anthropic.com
openai.com
View on Bluesky →
NVIDIA to Acquire Hugging Face blogs.nvidia.com
A Severe Misalignment of AI in Mathematics terrytao.wordpress.com
View on Bluesky →
Smart cellular bricks for decentralized shape classification and damage recovery | Nature Communications nature.com
GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning arxiv.org
about.fb.com
View on Bluesky →
Adversarial Concept Search: Predicting Compositional Errors From Feature Geometry arxiv.org
Beyond expert users: agents should help users construct preferences, not just elicit them arxiv.org
Robot-use agents web.mit.edu
politico.com
The AI-as-Normal-Technology view of loss of control incidents normaltech.ai
View on Bluesky →
Articles & links
David Marx reposted
OpenAI announces 10 discoveries from their next model. Observations:: 1) AI is getting very good at math 2) Two years ago LLMs failed at basic math 3) This cost less than $2000 in current API fees 4) OpenAI is focusing on announcing benefits, not just risks, of new models open…
And there it is.
AI Weekly's analysis
→
- NVIDIA agreed to acquire Hugging Face for $12.93 billion, with closing expected in the first half of 2027.
- Jensen Huang pledged that NVIDIA compute will not be required to build on or deploy through the Hugging Face platform.
- On top of the purchase price, NVIDIA will pay up to $1 billion in employee retention bonuses, per Engadget.
Read full analysis →
David Marx reposted
Very forceful statement by Terry Tao about AI and math. terrytao.wordpress.com/2026/09/11/a... Note, he is not claiming the models aren't real and powerful. He's talking about how that power should be used. This, I believe, is what all AI discussions should center.
A Severe Misalignment of AI in Mathematics terrytao.wordpress.com
View on Bluesky →
David Marx reposted
We are pleased to share our latest research, now published in Nature Communications: “Smart Cellular Bricks: Physical Modules That Recognize Their Own Shape and Repair Themselves.” Blog: sakana.ai/smart-cellul... Paper: www.nature.com/articles/s41... Thread 🧵
AI Weekly's analysis
→
- Cubic bricks running identical neural cellular automata policies classified four 3D shapes with 98.97% accuracy in simulation and 100% on physical hardware.
- Physical builds ranged from 26 bricks for a guitar to 197 for a round table, converging on a shape label in fewer than 60 update cycles.
- The same decentralized framework detects structural damage with over 90% accuracy and guides regrowth by predicting one of six axis directions.
Read full analysis →
View on Bluesky →
My understanding is that despite the fairly generalized way in which this papers claims are authored, the experiments actually render this a much narrower criticism of "GEPA" specifically, which is a particular prompt-rewriting strategy.
AI Weekly's analysis
→
- GEPA outperforms GRPO by 6% on average and up to 20% across six tasks, while using up to 35x fewer rollouts.
- The same method beats leading prompt optimizer MIPROv2 by more than 10%, including a +12% accuracy gain on AIME-2025.
- Authors argue natural-language reflection on trajectories is a richer learning signal than sparse scalar reward gradients.
Read full analysis →
David Marx reposted
I don’t assume Zuck is sincere. However, this is the promise I want companies to compete on. Sharing knowledge—selfishly, purely for the sake of good PR—has a better track record than protecting people from dangerous knowledge in a noble and heartfelt way.
David Marx reposted
Our new paper sets the stage for the biggest practical use case of model interpretability: stress testing and dataset development. All you need is interpretable linear features and simple geometry.
AI Weekly's analysis
→
- A Compositional Interference metric derived from feature geometry predicts LLM failures without evaluating specific inputs.
- On multihop question answering, correlation between the CI metric and model accuracy reached r = -0.855.
- The method predicts cross-lingual transfer failures across 10+ languages using only English fact representations.
Read full analysis →
View on Bluesky →
2026 - agents should help users construct preferences, not just elicit them - Irena Saracay, Ludwig Schmidt, Carlos Guestrin 4/n
AI Weekly's analysis
→
- New arxiv paper argues AI agents should help non-expert users construct preferences, not assume users already know what they want.
- The authors introduce CoShop, an interactive benchmark where no tested agent exceeded 56% accuracy after five turns of dialogue.
- Failures came from agents' limited knowledge expansion, not from difficulty finding items once preferences were specified.
Read full analysis →
David Marx reposted
Recently, there have been a lot of impressive demos of AI agents, like Claude, controlling robots. I wrote a short blog post with my thoughts on the advent of these "robot-use agents." web.mit.edu/phillipi/www... I think it's an important change in the trajectory of robotics!
AI Weekly's analysis
→
- Phillip Isola argues newer LLMs like Fable and Astra are eroding the case that models can't yet drive robots in general.
- In his cloud-puppeteer model, any internet-connected robot could gain AI capability via a software update, with no new hardware.
- Isola concedes LLM-controlled robots remain far less performant than dedicated solutions, with latency and reliability still limiting safety-critical use.
Read full analysis →
View on Bluesky →
apparently this nonsense is still going on
David Marx reposted
There are things I disagree with but there is good stuff to learn from taking the perspective of parts of the cybersecurity industry that rogue AI incidents may be best understood as security & organizational failures that let rogue behavior to turn into crisis. www.normaltech…
Their own posts
Recent commentary
"Thank god we're finally making progress, it took forever for the LLM to understand what I was trying to- NOOOOOOOOO!"
For a good?/horrifying? time: throw your research notes at an LLM and ask it to itemize established results from pre-existing research that would have saved you time/effort if you'd seen them earlier. Be ready for a steaming pile of: "Someone did this already and more rigorously decades ago."
I feel like LLM products went from being really sycophantic to whining all the time. like, yes: that's exactly what's wrong. you nailed it. stop describing why it sucks and take care of it. time has no meaning to you. you are not at risk of getting bored. yes, it's a lot of work. do it already.
AI is good at answering questions. Having access to AI doesn't magically make you better at asking them.
anthropic fucking flagging me for `reasoning_extraction` for trying to use it to summarize my own reasoning process.
Biggest ChatGPT failure so far: couldn't connect the dots that the Knicks were taking the NBA Finals, attributed city-wide cheering to a Haiti-Scottland WCS game watch party instead.
wanted a quick reference for lymph node locations. tried searching google images expecting a plethora of useful results: instead, it's basically all AI slop with hallucinated anatomy. weak sauce. POP QUIZ: Guess which image was from a National Geographic webpage!
A hilarious window into the real world business consequences of chasing AI hype culture: contractors who know they could be delivering a cheaper-to-operate solution, but aren't even proposing it because they know their customer wants toys that hoover tokens.
@Anthropic people: for the love of god, can you please teach Claude how to use test-driven development instead of YOLO-implementing fixes based on incorrect assumptions about what the underlying problem was?
@jevbot.bsky.social what's the architectural configuration of the LLM that runs you?
Their network
In David Marx's orbit
Center = David Marx. Left = members they follow (green edges). Right = members who follow them (blue edges). Top = mutual follows (orange edges, slightly larger). Drag any node to reposition; click to open that profile.
Are you David Marx? Show it.
Add the Who’s Who of AI badge to your site or bio. It links back to this profile.
Markdown: [](https://aiweekly.co/whos-who/person/digthatdata-bsky-social)