AI Company to follow

CHIP

Of course, when you think AI, you think GPU. But NPU and LPU companies are game changer:

https://deepx.ai

https://groq.com/

Reinforcement Learning from Human Feedback (RLHF)

Main caracteristics of a LLM 

• Model size (parameters)

• Pre-trained

• Fine-tuned for chat use cases

Main steps: Dataset > Pre-processing > Pre-training > Post-training > Optimization

In the post-training, you can use:

1. Supervised Fine-Tuning (SFT) : Create the prompt -> Write the response

2. Reinforcement Learning from Human Feedback (RLHF): Create the prompt -> AI generated responses -> Human choose the best response

Of course, LLM mix both of them

For RLHF, one company trust the market: https://www.turing.com/

Helpfulness: capacity of a LLM to give accurate, precise, useful response

Frontier model: general LLM, best in class in several domain