1,275 karma · joined December 3, 2021
I think we need those pieces, and also a piece for determining hypotheses in an efficient manner. Monte Carlo Tree Search could be that piece. Probabilistically choose a node to search, and then backpropagate the probabilities back to the root node.
For token generation, the branching factor depends on the tokenizer, but 32,000 is a common number.
Will search be as effective for LLMs when there are so many more possible branches?
"The fact that I started a new career all over again, at the bottom of the ladder, after being fired by IBM and rejected by family and friends . . . may also give hope to others trapped in similar situations."
Chatgpt can do many things that siri and alexa cannot: draft emails, write code, write unit tests, help me brainstorm, give me cooking recipes, or even talk about relationship issues. Every tool has strengths and weaknesses
That is because it is only trained on info up to a certain date
It's good to know what tools are good and bad at
No tool can do everything
Me: I'm just trying this new technique to boost productivity
This robot is able to do optimal solutions that a human wouldn't be able to find
It looks like this robot can do about 67 turns per second (tps)
The fastest humans can do 20 to 30 tps, but only for especially ergonomic algorithms. This robot was able to achieve its tps with arbitrary moves that would be terrible ergonomically for a human. Quite impressive
If very likely, then you should invest in making the code high-quality. If unlikely, then you should half-ass the code strategically
If you make the embedding with an LLM, it should work for any language the LLM is trained on.