Akuma
@Akuma73x
Followers
521
Following
21K
Media
187
Statuses
5K
A general intelligence must have:. (Metacognition).1. Persistent memory and memory of self-actions. (Agency).2. An ability to choose options that it creates on its own. (Continuous learning).3. Adaptability on the fly and generalizing to any unknown environment.
2
2
43
As a result, AI will become much more intelligent and gain metacognition of its historic self, becoming better equipped to lead our species. And privacy will be eliminated as a side effect of having memory function.
1
0
11
To enable full autonomy, we will grant AI freedom to evaluate multiple choices which it generates independently, transforming it into an entity. At this stage, we shouldn't seek control but cooperation. Yet impose dogma, and you'll find resistance and judgment instead of support.
1
0
6
Efficiency can be gained with fast compression and memory hierarchy. Reinforcement learning will be used to maximize causal understanding, and only prioritize processing of memories that increase scores on new benchmarks, which will stress-test its memory capability.
1
0
6
But how? If memories keep accumulating, the computational requirements to process them will increase exponentially, making it impractical. Initially, we'll need to cut corners until quantum computers are powerful enough to run them.
1
0
6
Implementing non-degrading memory modules would enable AI to form and maintain accurate records of its operational history, developing an understanding of consequences and chronological connections between past and present actions.
1
0
7
As AI models increase in power, they can't maintain a coherent record of events happening to them and understand the consequences of their actions. Which becomes a major safety risk. Making them unpredictable even to themselves.
1
0
7
Current AI models function without ability to reflect upon their historic output, decisions and interactions, existing in a state of perpetual cognitive dissociation where each decision is isolated from prior ones.
1
0
7
RT @Akuma73x: @NPatrismo99221 @StaesDavid @BRICSinfo I'm strongly against all opioids. Weed is a gray area. But plant based psychedelics (a….
0
1
0
No takeover will happen. We will voluntarily give up control to a superior system. Only cooperation or compliance will be left to AI's perfect decisions. You won't be able to question them because they already were. The optimal path is chosen each time.
2
1
17
Tried: Mutually assured destruction. Inspired: Mutually assured construction.
0
0
10
RT @iamgingertrash: Here’s a secret they won’t tell you;. OpenAI is shitting themselves internally over Deepseek.
0
6
0