The way Karl Friston is communicating is already mindblowing, first he is always super nice, polite, humble and active listening (no matter whom sits in front of him) and when he answers it is like he is giving definitions right from the book. With some many details and side notes, that at least myself has to listen 2-3 times to really get everything he wanted to express. It is also, his active inference approach is so beautifully, elegantly simple but at the same time hard grasp when you first learn it, that I am very excited to see what will come out of that for future AI!
I agree. It would be disappointing to realise that there is no AI but LLMs. I'm sure, we'll see quite soon some utterly new architectures beyond transformers.
You know, this is what a talk should look like. At least if it's about a subject of science. Real points made, to reach a rational consensus. I don't care about altmans advertisment talks. I don't gain any understanding about the subject from it, neither do I gain a timeline for the future of the software. So thanks for the great work! I hope WEF will offer a bigger audience and panel next time.
I disagree. Some of us who follow AI closely have serious concerns about LeCun and Friston's conclusions. To me, they are two of the most striking examples of fine intellects gone astray.
@@netscrooge It's your right to feel that way. I personally have a different view of the topic. I think safety and guardrails are important, but I also feel like we are just in the early stages of LLMs. Most of them can't solve first year math student boolean riddles. Yes, they can be used for disinformation and bots - but well, that's nothing new besides perhaps of the scale. And yes, I know, I haven't touched on all the possibilities. But still - I think caution is important, but not fear. Fear is the enemy of the mind.
@@AI-HOMELAB Sorry I wasn't clear. I wasn't focusing on whether or not we should be afraid. I'm talking about their understanding of what's happening inside LLMs. Friston's comment, "deep learning is rubbish," is a shining example.
This is an interesting discussion. For science to progress it is important that we have people working in the current paradigm to push it as far as possible and people working in alternative paradigms to try to find the next breakthrough.
In my opinion, it would be better to have an interview with just Yann Lecun, everything else makes the conversation boring and more distracting...A moderaror that makes better questions would be a major help as well...Excellent ideas by Yann!
To be fair, the time it take a kid to learn to play atari games is predicated on hundreds of thousands of generations of interaction environmental training that builds up a capability to infer world models from attentional primitives in the brain and nervous system. We're basically building this from scratch with modern computational AI.
Try for you imagine a maze you try 1 at a time but imagine how with individual brute force AI every path how the paths different every outcome in an instant with best outcome calculated but people we can do this all
Isn't it weird that in discussions of AI or natural intelligence, we omit such a criteria as whether a species creates means of a total self-destruction. If we put this criteria on intelligence scale, how would we rate humans compared to all other species ?
Provocative, and actually I agree it's ironic that "intelligence" brought us the possibility of self distruction, but if you think through it, it's only a matter of power: knowledge gives power, power means you can even destroy your own planet. I hope we won't.