On AGI

I think what we’re seeing with increasingly closed off models like Mythos and Fable, it’s true that the controlling aspects indicate a lot and I do think something is happening, however I believe it is not in the direction of AGI but rather technical prowess, primarily in cybersecurity.

An evil prompt or fine tune/abliteration could very easily unleash quite a menace in society so the worry is valid, however it’s not about models becoming sentient I don’t believe in that.

The closest thing COULD be emerging clusteral behaviors in the neural nets of the system like a complex network. Probably from our own secret human biases and intentions because AI is an “aggregation” of our universe’s information with the personality of an “average human” subject to huge statistical outliers like saints and serial killers depending on how the model was trained. Read my articles on complex networks here: https://medium.com/@husaynabbasgokal