Harsh Bhatt

this is the video of last November, was building with arduino

Experimenting Multi turn RL.

**My vocabulary**: Experience Rollout Trajectories Multi-turn Rewards Momentum Optimizer Loss Add more ...

Asynchronous Reinforcement Learning it is

RL should never stop.

2 Machine, 1 GPU, 16GB EACH.

Distributed training it is.

“By far, the greatest danger of Artificial Intelligence isthat people conclude too early that they understand it.” - Eliezer Yudkowsky

We are in the peak Engineering era; it's like singing. Anyone can write code like anyone can sing a song, but not everyone is Ed Sheeran or Arijit Singh.

.@elonmusk @Tesla It will be cool if Tesla can drive using Reasoning, probably need to have all the CV input that could be input for the reasoning model, and it needs to be finetuned through RL in a simulated environment.

"Once we turn all digital worlds into an environment, solve it with smart RL algorithms, we have digital AGI." - Shunyu Yao

Progress on AI benchmark in the past 5 years.

My life is reinforced by the learning.