Founder/CEO AutoInfer. ex-CTO, Niqo Robotics. ex-HighlightCam (YC-S'09), ex-Soliton
Bengaluru, India
Joined June 2009
- Tweets156
- Following754
- Followers270
- Likes643
What they both say aren't necessarily orthogonal. @karpathy's main issue with RL is current credit assignment, which doesn't care about cruciality of actions at states (some actions are more crucial). Secondly he dislikes long chains of supervision flow & very large rollouts.
Octopus, Raven, and Squirrel are my fav examples to show how small a brain it takes to do "intelligent & combinatorial problem solving" at run-time (as opposed to design-time the way we train our LLMs)
@RichardSSutton indeed has a valid point.
@karpathy probably meant that our current methods of RL are terrible. Or did he mean the RL goal or the problem formulation itself? 🤔
I feel @RichardSSutton's argument is very long horizon & asymptotic. He is not necessarily saying in next 5 years, search & learning is most important. But assuming orders of magnitude more compute (say Quantum computers etc), then what should we be working on.
Sumod K Mohan retweeted
In a game of 64 squares, you've opened a world of endless possibilities. Congratulations, @DGukesh, on becoming the 18th World Champion at just 18! Following in Vishy’s footsteps, you're now guiding the next wave of Indian chess prodigies. 🇮🇳♟️🏆
From Bishop's DL book. Think searching for a movie, say with its attributes like genre, lang, actors.. (key). For this key, you could find other movies that fit this char, that is its value. To find similar movies, first find the char (query), then closest key & its value.
Sumod K Mohan retweeted
🌟 Call for Talk Proposals! Do you work on projects showcasing the power of AI and ML/DL applications in creative AI and generative AI? Submit your proposal for the demo track at The Fifth Elephant 2023 Winter edition before Nov 15: buff.ly/3seKk39 #GenAI
Sumod K Mohan retweeted
Chandrayaan-3 Mission:
'India🇮🇳,
I reached my destination
and you too!'
: Chandrayaan-3
Chandrayaan-3 has successfully
soft-landed on the moon 🌖!.
Congratulations, India🇮🇳!
#Chandrayaan_3
#Ch3
Sumod K Mohan retweeted
📣 Calling all Generative AI builders in Pune for a Gen-AI meetup on September 9th!
Join us for a series of insightful talks by speakers from @Replit and Calsoft and a workshop by @llama_index (@ravithejads ). You will also get a chance to interact with the founders of @Peerlist (@designerdada ) and Kombai who will talk about how they are integrating AI into their products and workflows.
This event has limited registrations so make sure to sign up ASAP!
lu.ma/4b0hfgok
Super pumped up for #Chandrayaan3. As a technologist in allied areas (Robotics), mad respect for the folks at @isro. Anyone interested to catch up for watch party in #Indiranagar. If you are curious about what had happened with #Chandrayaan2 m.youtube.com/watch?v=fZ2sNR…
At a big picture level this is very interesting, that total deaths from aviation is much smaller than road transportation. But comparing 176 vs 1.3 million feels like 3-4 orders of mag safer, that is not correct. Aviation is about 30-60 times safer per mile but 3x riskier/trip
One of my friends is doing a workshop for CXO's, Product Folks, VC's etc on how to develop AI products. He has led many AI products and wrote a O'Reilly book on NLP Check out his workshop here.
@GradientAdvisor & @hasgeek present a 2-day workshop in Blore on 21st & 22nd july, 2023, titled "How to Do AI right and achieve in the intended business Outcomes".
hasgeek.com/anuj_gupta/how-t…
Limited tickets available on a first-come, first-served basis
Sumod K Mohan retweeted
I'd strongly support the idea of a Manhattan Project of intense research to make machines more trustworthy and interpretable (regardless of, or in parallel with a moratorium.) The premature super-investment in non-interpretable technologies is the core of our problems.
Absolutely, a Manhattan project on getting AI systems to align with human values would be a great thing.
(As an aside, the letter wasn’t a blanket moratorium and it actually encouraged research to make machines more trustworthy and interpretable; I would not have signed otherwise; I also didn’t write it and didn’t mean to exclude other noble projects).
Had a blast at the hackathon and this is really great guys. Really like the high bar on transparency @dementorSam @NirantK and team.
Having participated in many hackathons as The Bayesian Conspiracy - github.com/thebayesianconspi…
Always wished to organise one with the fun and transparency bar very few adhere to.
Thanks @NirantK and 🍻 to more such things!
Sumod K Mohan retweeted
How do we compare the scale of language learning input for large language models vs. humans? I've been trying to come to grips with recent progress in AI. Let me explain these two illustrations I made to help. 🧵
A very nice interview with Knuth by the folks at Bhavana (the mathematical magazine). bhavana.org.in/the-dawn-of-r…. Knuth answered all questions in 280 words!! Bhavana: what a beautiful name for a mathematical magazine.
Sumod K Mohan retweeted
A great mathematician named Madhava, working in Kerala, India, discovered the power series for sin(x) in the 1300s, hundreds of years before Newton, Leibniz, Gregory, Taylor, et al. In this video, master teacher @garyrubinstein shows how Madhava did it: youtube.com/watch?v=mzigLbCc…
This is so true. I used to hate writing as a fresh grad student, I only wanted to do research. But soon realised that writing was my long term memory, my organizer of ideas and personal dojo for ideas to fight it out. Writing shapes my thinking and writing is huge part of rsch
Sumod K Mohan retweeted
In the LLM-science discussion, I see a common misconception that science is a thing you do and that writing about it is separate and can be automated. I’ve written over 300 scientific papers and can assure you that science writing can’t be separated from science doing. Why? 1/18
