Let’s stop the “AGI is just around the corner” narrative. I still believe true intelligence will emerge from the collective, and it may well still take quite some time before we get there.
Amazing talk by M. Jordan: youtube.com/live/W0QLq4q...
Let’s move past the hype, build a clear vision for the technology, and partly refocus on real breakthrough research—like what was done in the early days of DeepMind.
Autonomous Agents (now AI Agents), multi-agent systems, economics, game theory, mechanism design, and market dynamics could play a key role in shaping its future.
That said, this is an incredible technology, and we need to figure out where it can be best applied. I fully agree with Michael Jordan that there’s a serious lack of vision in the field.
This is not AGI. It may turn out to be an important building block, but we still need real breakthroughs in architectures that can construct and compress a true world model. Simply throwing more data and compute at the problem isn’t enough—here, too, less can be more.
We can query it using natural language, thanks to massive compute and post-training refinements. At its core, it retrieves and generates responses through pattern matching and data generalization—that’s it.
The way I see the current state of research in LLMs and foundation models is that we’ve figured out a way to build an intelligent database — one that compresses human knowledge (largely from the internet) in an autoregressive manner.
The first textbook on multi-agent reinforcement learning is out - a landmark for the field, the first textbook covering game-theoretic foundations with state-of-the-art deep learning! Congrats to its authors Stefano Albrecht , LukasSchaefer and Filippos Christianos
More details: www.marl-book.com
Fascinating read on streaming RL in deep learning - arxiv.org/pdf/2410.14606
The Andromeda galaxy captured by the Hubble Space telescope
How do people genuinely feel about this? are we not pushing it too far now? I have never associated the terminology described here with being offensive or exclusionary, but perhaps I'm just unaware, curious for your opinions. www.acm.org/diversity-in...
I'm loving this place. It's a technological miracle to only have seen two small outages in the past 2 weeks.. 🙏🙏🙏😅
Ok, last two papers for this week!
A final game-theoretic RLHF method and a different take on RLHF altogether inspired by prospect theory.
1. 🧲 Magnetic Preference Optimization (MPO).
2. Kahneman-Tversky Optimization (KTO).
🧵 1/3.
