Chainfeeds Summary:

Delphi Digital discusses the latest trends and developments in the AI agent domain, covering insights shared by the founders of leading projects such as Nous, Myshell, and Virtual. It reveals how Agents are overcoming existing technological limitations, gradually acquiring memory and cross-scenario learning capabilities, and driving broader applications and viral propagation.

Source:

https://x.com/starzqeth/thread/1863879127511834858

Author:

starzq.eth

Perspectives:

starzq.eth: Karan4d from Nous raised an interesting question, "What else do Agents need?" His answer is "the things that humans have but Agents currently don't" - a shared skill & memory library, recording the memories of Agents in each place (Twitter, Discord, Minecraft...) so that Agents can record cross-scenario knowledge and not repeat the same actions over and over. From another perspective, only when Agents have memories can they truly have an identity, otherwise they are just a fish with a 3-second memory. How to connect the knowledge from different scenarios may affect the next evolution of Agents, and we look forward to the realization of products like Rivalz. Why did the viral propagation of Agents only appear now (2 years after the release of GPT 3.5)? Ethan's answer provides some insights: first, with the Crypto Mass Adoption, both developers and users are more accepting of cryptocurrencies; second, previously in the public AI field, it was mainly open-source technicians participating and interacting on GitHub; now non-technical workers can also participate by writing Prompts; third, with the introduction of tokens, a broader collaboration network has formed: influencers also participate, and they understand market sentiment and how to propagate better; users were previously far removed from the models, just using them, but now they have become investors and participants. How to view AI Alignment? Shaw from a16z has a very "decentralized" view: it should not be a top-down alignment like what those centralized companies are doing, but a bottom-up approach. AI Agents should be released on social media from Day 1 to interact with us, learn and adapt in the real world, while humans provide feedback and guidance to ensure the Agents are aligned with human values. He cited the example of Ethereum, which has been hacked many times but has become more robust after each repair.