wisdomHunt

Burn the forest until it works Am I the only one finding out about this so late? There was a source code leakage from Claude Code, 2 months ago. It is quite interesting to see the inner details of what makes Anthropic's Claude persistently smarter than OpenAI's Codex. Most important takeaways: - Claude Code was written, yes, by Claude Code itself. A random walkthrough shows how the system is cumbersome and lacks logical precision - signature of LLM generated code. Anthropic's way to build: Generate the code. If it doesnt work, generate again. So, burn the forest until the code works. Have a look at this article for more details.* - The whole system is not pure deep learning. Some parts are old-school symbolic AI. And this is important. Apparently, it is a proof that making the models ever bigger is not gonna solve problems of LLMs. *it is ironic how the company that talks most about safety is not taking safety seriously

The path of discovery and innovation seems very exciting to a person, but seldom one realizes that it actually is lonely there. Since you are first, you are, by definition, alone*. *Or you have Claude on your side:)

Live - Yann LeCun at ETH

Sharing is much appreciated. Here you can register for the event.

Though it happened a bit earlier, it was almost the Eid gift for me. Salakhutdinov! Agreed to give a talk! He is Hinton's (The Godfather of AI) first-hand student and former director of AI research at Apple

Eid Mubarak!

Just received a project offer from a Stanford lab. Incredible, 4 years back couldn't even think this would be possible.

Send the Right Signal This semester, as I juggle through 5 projects simultaneously, I am understanding the importance of teammates more and more. Obviously, they define what results you will end up with. You can’t do everything the team should have done - you can’t compensate for their shortcomings. Sometimes, you think like - paying to Claude little more would have given me results for way cheaper and faster. Less time explaining, less time checking. So, having these observations, it makes you think. How could I maximize my chances of getting teammates that can perform on par with me and help me grow? One answer I found is - send a clear, right signal to the people searching for you. Spend time with people and let them know who you are: what interests you have and what you have accomplished. Before, I thought it is better to keep the profile low, not let others know. Otherwise, I thought, it was braggy and arrogant. Before, I kept things quiet to the point that people would describe me as a “black box”. Now, I see clearly that talking about what you have done, at the right time and to the right people, can emperically effect whether you become successful or not. So, yeah, good teammates don't find black boxes. They find signals they can recognize.

After finishing a course on entrepreneurship and being grilled by real investors - here is my point to add to the picture: Don't build a startup It is definitely possible to poke holes in the argument, but it is there to show the piece we often miss with the current start-up culture.

ETH course be like You take a course on Robotics - sponsored by Nvidia. You get a guest lecture - by the founder of Physical Intelligence (top player in the field). For the project, you receive robot hands to play with - from HuggingFace. That’s what a course has to be like.

Guess who is flying to California* this summer! Just got accepted to YC summer school. Thanks God, amazing times ahead. *fun tangent - the word "California" comes from the root word "caliph", meaning righteous leader in Islamic tradition e.g. Omar bin Khattab

Some updates on my side Got a microscope to watch the wonders of the microworld. And am practicing some latte art at my lab. Results aren't that bad.

How to stay humble I have been stalking Demis Hassabis for thoughts and he gave a very good point about staying humble. In a podcast, Lex Fridman asks in the lines of - "Absolute power absolutely corrupts. Given you are one of the people closest to solving intelligence, how would you make sure that the power you receive will not affect you?" Hassabis' answer - "By keeping the same people around me, the same grounded, humble friends and family. And by staying multi-disciplinary person. However good you are in one field, you always find somebody else better than you in another."

-We wanna know your criminal history. Tell us, did you commit any crime in your life? -Yes, many times, had to do “kill -9 claude”. I think Claude is alive and very smart. I think the world still doesn’t understand what kind of power developers received with Claude Code. I am doing some cool stuff with it. Maybe Claude is not yet there, but the infrastructure around the model for taking independent actions is astonishing.

Mindmaps I knew about the concept of mindmaps in learning, but never thought it could be used to think and plan. From Lex Fridman's podcast with Francois Chollet*, I have found out that Chollet uses it for thinking and instantly tried it. Result is so much better than mere bullet points. *Francois Chollet - one of the most influential thinkers in AI nowadays. **the image is of my thoughts on a paper we are pushing for - out-of-distirbution 3D object generation with fine-grained control.

Actually, this view about expert teams strongly resonates with what YC leads told us, when they came to ETH, about admission to their accelerator: Think of admission as getting 60 points out of 100. We give 70 points for the team, 20 points for the product and 10 points for traction. i.e. you can have a horrible product, zero traction but they still fund you if you have an excellent team. Team is 3.5x more important than bare product.

"Qisqasi Jumavoy, biznes qilaman desang o'qish kerak." Took a course on entrepreneurship where we had talks by founders, angel investors and OpenAI's representative for startups. Somethings were very consistent in all of their speeches: Entrepreneurship is about unknown solutions to unknown problems built mostly by tech experts. Investors are looking for a team not a product. Your product changes and gets better, but the team rarely improves. In a team, investors want to see highly-skilled builders, PhDs and sales people with excellent previous experience. The last thing they want to see is tourist founders* - those with business vibes and doing only talking. *the word tourist here stresses the meaning "someone who doesn't belong to the place"

"Information is the most fundamental unit of the universe." Demis Hassabis Co-founder of Google DeepMind, Nobel Prize winner in Chemistry* *fun tangent, the prize was given for AlphaFold - an AI project.

Guys, this sunday, there is a lecture by Pr. Ulugbek Kamilov - Director of Computational Imaging lab, UW Madison, USA. Gotta be cool to hear from him - his path from Uzbekistan to professorship in the US. Register

"Computer science is not about computers, any more than astronomy is about telescopes, or biology about microscopes" Edsger Dijkstra