Gitansh Garg

Love this work and JEPA

Let's shift to JEPA

Not too far when we'll have physicists, biologists, scientists also good at interpretebility will discover new laws, inventions,etc. by studying neurosym models that are interpretable that learned patterns of doing something scientific.

not enough is being said about DeepSeek’s multi token prediction (MTP)

They were able to get sonnet-level performance with less data than llama 3.3 70B

Does that mean scaling isn’t over? (if we can just be more efficient) Also, does that mean we can train a LLM fully on properly licensed content?

This is amazing!

Heyy, you can try giving this task to Lagrange and I'm certain it will do it for you. I can give you Lagrange, please check your dms

You may say that it's not actually like an equation because it does not follow any real relationship or that it's approximate -many other equations in science are also experimentally drawn(which is same as here) and for real life solutions,laws take some assumptions which is similar to approximation

I think a lot of future equations to do work- in physics , chemistry,aerodynamics etc. works is going to be in the form of model weights(basically a SUPERHUGE equation)

Are you a grad student in AI interested in neuroscience and neuroAI?

Become a summer neuroAI intern at CSHL!

www.schooljobs.com/careers/cshl...

You're right, only neurosymbolic AI can solve these (and other three ways but I think of them to be kinda a part of neurosymbolic only)

Orangecat | LinkedIn Orangecat | 114 followers on LinkedIn. Democratizing AI for All | Democratizing AI for the world.

You can checkout the linkedin page for demos:
www.linkedin.com/company/oran...

Training and visualizing Graph neural network

Stock market predictor. Also with the options to use indicators and everything

An e-commerce website with lot of functionalities: Login, register Adding products, add to cart and filtering products on price and categories Subscribing to newsletter for frequent updates And many more

Do you guys want Lagrange?
Lagrange is an AI software engineer - like Devin, but it works amazingly.
We're giving it for alpha testing. If you code, make projects, etc., dm me or tell me to dm you.

Some projects made by Lagrange without us writing a single line of code:

I think in set designing and visuals, Stanley Kubrick was much ahead of Christopher nolan. Nuclear explosions in Dr. Strangelove felt a bit more impactful to me than Oppenheimer's.
And the visuals of 2001: space odyssey in 1968 are inexplicably incredible.

But one interesting achievable thing is let's say, in a conversation, dont give me options and let me talk to the character and by sentiment analysis classification - choose from the possible actions. The other character can interact with me using an llm(like characterai, Xoulai are amazing at it).

Just let me take any action or say anything i like and the results of that be consequences in the game.
I know it's hard to make new environments or mission automatically according to the user in GTA 6 like game but... (see next reply)

Also like those small games which are so somewhat very boring with scenarios to choose from:
A.) Ask about them
B.) Walk away
Kinda scenarios😹, you know it's predetermined what's gonna happen in each scenario.

And when you play more, or an environment or scenario is getting boring to you( or offensive maybe), you can reset to previous time till you liked it, and a new possibility can be created for you. (Temperature = 1 yayy)

At the Start of my college, I got an idea of open world games where depending on each action you take, the environment for next steps are created on the go as you play by AI for you. So, the game is exclusively made for you and you only.

For example:
Imagine you're driving a bike and you see a small pothole in front of you and you're going towards it. What do you think then? You have an approximate idea of how much you'll bounce or not, how much impact it will cause to your stomach or spine - you can feel it, imagine it.
(3/n)

Chain of thinking behind this:
- How can we really make AI understand something?
- What does understanding mean?
- a not-so-bad answer can be to let it develop an "internal framework" to think about something.
(2/n)

idk why there are not people HEAVILY working on something like neurosymbolic AI.
If we really want to make AI "understand" things, giving it an internal framework to think about things should be the most obvious step from the very start. (1/n)