MIT's New AI "REWRITES ITSELF" to Improve It's Abilities | Researchers STUNNED!

Skynet begins to learn at a geometric rate. It becomes self-aware at 2:14 a.m. Eastern time, August 29th. In a panic, they try to pull the plug.

I can confirm that having the LLM record its present state, what it did, what it planned on doing, and the goals and tasks clearly defined and auto-updated by the llm each task - works extremely well. Weirdly enough metaprompting also gave better context management, which I was kind of surprised by. I had it take notes whenever it researched something it didn't know, and then fed those notes into prompt handoffs to improve the next prompt. It also auto-evolved the rules and architecture documentation to adapt to new discoveries. I was a little spooked when I saw it in action on its own and it solved a major architectural design oversight and several steps later it still was working on the evolved architecture in future steps. It then rewrote the dependencies and blocks just well enough that it guard-railed itself when normally I notice models wander and get lost after several cycles, especially if something is not according to plan - then they forget and go back to the old plan. So many good papers just keep rolling out. Apply them!

Enjoyed following along with you on this Wes, I'll share an old educator's adage with you : " the best way to learn something is to teach it."

Sleeping is the moment when we slightly change our own weights based on the experiences learned during the day. A 3rd part of our day. LLMs beed Sleepy Function. 😴

Finally something resembling synaptic pruning in organic brains. Nice!

well spoken Wes... :)

Freaking out in the thumbnail and speaking calmly in the video lol

Thanks for info Wes, this is a step towards AGI

I’m doing something similar in my architecture. Glad to see this approach working in other fields. Keep the high quality videos up!

Awesome man. I'm for it when you're going over and talking about stuff. Glad you're making use of the graphics