I agree with you, that’s also my intuition (decentralized/local learning should outperform localized). But we also worked for years on multi-agent DL, hoping it will have advantages over centralized (gradients flowing up to the objective function) and it still seems centralized
@marek_rosa
-
Singletons and Economic Models in AI Systems
By
–
I can see two types of singletons:
1) one top-level objective & all sub-agents aligned on it (aka centralized economy)
2) one top-level objective & but sub-agents are misaligned and there's a lot of conflict (aka capitalism) Then there's another option:
3) no singleton, just -
Singleton vs Multi-Agent Systems: Objective Hierarchy Differences
By
–
IMO, the difference is that singleton has one single objective (which it then cascades down the hierarchy to sub-agents), whereas a multi-agent system has many objectives (each agent it own objective; some are aligned, some are orthogonal)
-
Singleton AGI versus Co-evolving Multi-Agent Systems
By
–
I always expected it will be millions of co-evolving AGIs. However, it's still possible that the strongest singleton will emerge from GPT-N, because centralized learning seems to be still more efficient than decentralized (evolution, multi-agent RL, etc)
-
AI exponential progress will require AIs summarizing AI advancements
By
–
Very soon, AIs will develop new AIs, and then this progress will happen in mere minutes (not weeks). And then, the rate of progress will grow exponentially. We won't be able to catch up We will need AIs that will summarize the progress for us.
-
LLM Agents Self-Improvement Society at Prague AI Hackathon
By
–
My colleagues (Jan Feyereisl @thefillm and Karel Hovorka @vynalezce
) and I are going to participate (compete?) at Prague AI Hackathon on April 21-22 We already have our topic: Society of LLM-driven agents that recursively self-improve It will be fun! Wish us luck! @GoodAIdev -
LLM Agents Learning from Experience for Long-Term Planning
By
–
Great analysis of the current state of LLM-driven autonomous agents (what works, what doesn't work) IMHO, the next steps are autonomous: learning from experience and then using this synthetic data to fine-tune LLM to be better at following long-term plans, thinking in the loop,
-
Robot Etymology: From Slavic Labor to Modern Automation Ethics
By
–
I think "robot" comes from czech/slovak word "robota", which means "work", but this word was used in the previous centuries also do describe a slavery kind of work than my people had to do for their lords. So yeah, it's kind of slave.
-
AI Moral Status: Tool vs Sentient Agent Ethics
By
–
If I could I, would vote for #1 and #2 (less than human + equal to human). Because we need AIs that are simply tools and nothing else. But I also understand that if we create complex agents, more intelligent than humans, that it's quite immoral to keep them in slavery (as
-
Model Quality Assessment Despite Interface Improvements Needed
By
–
Yeah, the interface could get better. Probably will. But the model itself is very good.