AGI roadmap discussion ---

Community
, ,
1

I have watched interesting video about LLMs and which way things are going.
I was wondering your opinions on this, and how does these different roads align with chatgpt and OpenAI Goals ?

AI Won’t Be AGI, Until It Can At Least Do This (plus 6 key ways LLMs are being upgraded) - YouTube

1 Like
2

Thanks for sharing, here is a summary of the “6 key ways” if you don’t want to watch all 30 minutes:

1. Compositionality: Enhancing LLMs to better piece together reasoning blocks into more complex solutions.
2. Verifiers: Using external systems or verifiers to help LLMs locate and validate the correct reasoning chains or programs.
3. Many-shot Learning: Providing numerous examples to help LLMs better learn how to solve specific tasks.
4. Test-time Fine-tuning: Actively adapting the model on the fly with synthetic examples to improve performance on novel tasks.
5. Joint Training: Embedding specialized knowledge from other neural networks into LLMs.
6. Tacit Data: Incorporating unspoken, experiential knowledge from experts into AI training.
1 Like