6.4K
RA5 things I learned about AI from going to NeurIPS, the biggest AI conference
1. Everyone’s going for human data and reinforcement learning, this is essentially how AI models can get smarter and provide better answers for every situation because they need more data to train on. Essentially If you’re not working in an AI lab, you’re probably working to provide data to labs.
2. There were a lot of respect in the room among frontier researchers for Nano banana pro, anthropic opus 4.5 and xAI
3. Benchmarks dont mean anything in production
4. AI is not going to hit AGI with the current models we are building but that doesn’t really matter
5. The biggest theme being continual learning, how do we build AI that actually improves over time without forgetting everything it knew before, one talk that stuck with me was developing LLMs that can reason by arguing with themselves
Follow for more Tech and AI
#tech #ai
@raycfu
