This 20 year old algorithm is changing how LLMs think

LLMs are known to perform poorly on algorithmic tasks. Previous attempts to solve this with Chain of Thought and Graph of Thought have been promising.

A new algorithm suggests using Monte Carlo Tree Search, which is a game tree search algorithm that probabilistically finds optimal paths using simulations.

#SystemDesign #MCTS #LLMs

[mberoakoko]

That's Monte carlo reinforcement learning . It will not scale if the action or state space is significantly large.

[dilipkumar-sxne]

0:55
showing that chess position was personal

[akhilsudh]

Person of interest season 4 episode “If then else” shows how a model predicts the protagonists survival odds similarly. The diagram in this short looked so similar to how it happened in that show

[kaushikfunde]

Sir your explaining skill is great . But you can do more core topics to explain is also good . Can make videos/playlist on core algorithms and how do we make one of them from Scratch with full roadmap.

[aksjfh]

Nice chess game, by the way. Was that intentional or accidental?

[iamravishank]

Hi @gkcs
Is there any research paper for this. Can you please share, of any? Thanks

[HyperXD_YT]

AI was supposed to replace Chess engines, but now we have this:(

[tanvir-lab]

Sounds promising, but how long will it take to respond? 🤔

[prasenlonikar]

But mcts is useful in perfect information games right?
What about cfr ?

[abhiraams]

Can post the link to the research paper

[cmelgarejo]

I worry on how much will it take it to create my snake game in python 😢

[MultiPIYUSH]

More will train more better answers can make this better, initial stage of this algorithm can be really really good but it can be breakable right? If the output is wrong and still end party is satisfied

[prnva_]

The compute requirements will be insane