_max_entropy's profile picture. Exploring ideas in Computer Science | Human Centered AI

Yagnesh Revar

@_max_entropy

Exploring ideas in Computer Science | Human Centered AI

Impressed by the creativity of some experiments in raising awareness around risks of blind adoption of improperly regulated, ungrounded models. The findings fall within expected spectrum of LLM behaviors. They're bound to imitate prevalent human biases and conflate prompts!

Nothing to see here ;-) @Nature feature, by @SilverJacket nature.com/articles/d4158…

EricTopol's tweet image. Nothing to see here ;-)
@Nature feature, by @SilverJacket 
nature.com/articles/d4158…


This is cool. Love the visual illustrations of concepts and relevant citations. Seems like a patch mosaic kind of idea by @SuryaGanguli but for texts. Dataset attribution problem. Perhaps one of the most fundamental and intriguing problems in human life and awareness.

We're starting to apply interpretability to pre-deployment audits!



Yagnesh Revar a reposté

Looking for engineers for an exciting agentic AI startup in San Francisco. Contact me at [email protected]


Fascinating ideas! Logic to tame instincts for the good. A key loophole (interpretations) shares the solution space with some key challenges in AI today like alignment and grounding. To solve or not that is the question, or how much to solve and when 🤔 nytimes.com/interactive/20…


Shouldn't this website have a status page? How to ensure it's operational / working as expected? Separation of Powers and Checks and Balances | Constitution Annotated | Congress.gov | Library of Congress constitution.congress.gov/browse/essay/a…


This headline could be misleading. Fees are good -- can actually strengthen the bridge and create better conditions for both countries, as long as it stands for what it should. Trump’s H-1B Visa Fee Knocks Down a Bridge Between India and the U.S. nytimes.com/2025/09/22/bus…


I wrote this a while ago, but I still think it might be helpful for anyone curious about LLMs and their limitations. Would love to hear your thoughts! What day is it btw? I hope it's not already outdated 🤖📈 “Beyond Large Language Models“ beyondsymbols.medium.com/beyond-llms-a9…


I've often wondered whether diffusion models are creative and if so what makes them formulate novel ideas. Just found a fascinating work digging into this: Analytic theory of creativity for convolutional diffusion models. youtube.com/watch?v=oPIqhs…

_max_entropy's tweet card. Survey Lecture 8: Surya Ganguli

youtube.com

YouTube

Survey Lecture 8: Surya Ganguli

Thanks for sharing! Learning more about patch mosaic theory was enlightening. It's quite intuitive and seems to have a lot of potential for explainability, novelty, and unraveling mysteries of diffusion models. If I have to guess Joseph Fourier would have loved these ideas.



Internet pollution was already a thing, but now it's accelerated rapidly. Unregulated entities with hidden agendas (bots) are flooding platforms with contents without getting detected or removed. @YouTube @Google It's high time AI generated content needs to be signed.

I think we need one more Internet, one in which real people talk about real things: real images, text, and videos. So a total of 3 Internets (ref history). One real and one fake for us, and the third one for robots. Sharing the 3rd one with robots may not be a bad idea.



Yagnesh Revar a reposté

There is a very special type of evil that shows up when people see themselves as fighting for a "greater cause", so moral, so just, that it outweighs any individual immoral behaviour. Those people imagine this comes for the right place, while it comes from the worst one.


A good thing about oligarchy that's rarely talked about -- more accountability and simpler govt. shaping policies (considering they understand their role and are honestly committed to it).


Yagnesh Revar a reposté

Excellent new example of the color spreading effect.

Ce tweet n’est plus disponible.

Loading...

Something went wrong.


Something went wrong.