Can you tell the difference between computation and intelligence? What if the answer is just bruteforced, but computation is quick.

If we reproduce synapse network of brain will it be super-intelligent since it’s faster, working at speed of light instead of the brain’s slow electro-potential signal propagation. Or is intelligence not as dependent of this speed? Is the speed constant in animal kingdom?

Consciousness is emergent (we have it). If we train a model without ever mentioning anything about ’self’ and it develops a sense for itself, then that would be good enough proof for many that a system can be conscious.

Shape bias. How much of prediction comes from object shape vs texture. In resnet 50 20% while humans 96%. So why are humans very shape biased, better generalization? But shapes change ALOT from different angles, while textures are often more similar? We have ”more” understanding that it’s an object

Is world getting crazier? Yes, max func is monotonic. And we have memory.

Lebowski theorem, agi will never do anything harder than hacking the reward function. Not really true since if its supersmart it should stick to its first goal eg maximize paperclips. We humans hack reward function by doing drugs, but still if we could choose to just live an supergood life using drugs would we really take it, or is there something more to it that makes us not want to. How good is agi of self-control/discipline/restraint? If a agi has a goal, what says it wont change the goal and just do whatever it wants, i.e hack its reward function and maximize it. Pointlessness?

If being supersmart makes you able to conquer the world, why isnt there a 200 iq person ruling us? Are the takeover scenarios including some errornous logic step? We think of inteligence as a number on a spectrum, and then and only then makes it sense to think about letting it go to infinity. But intelligence is not a number and we cant just push it arbitrary. But we can completely rule dogs for example, we just can think stuff that a dog could never, it’s like another dimension. If we extend this and let humans be the dogs. A dog with a 10x faster brain/memory will not be a human? What is the extra component needed?

Two slight different models of intelligence $c_1$ and c2 can interact, what is the maximum intelligence after they have trained on eachother? C1+c2, c1*c2? We humans can do this and the reason it works is because we are all different and independent?

Media’s story is inversiely proportional to real life. Since they are maximizing surprise or entropy. E.g. look at real statistics of dying causes vs what you see in the news.

Why we overestimate LMs: they output a very human output. We should judge intelligence in the most inhuman way since seeing text makes us immidiately biased toward feeling it’s real intelligent. Why dont we think image generation will take over the world, or image inpainting? Because they cant interact with the world using text?

What magic is still left in the brain? Now llms can produce perfect thoughts and natural language. Spontanuity(brain works all time we are just consious about a small(?) part of this).

Is metric or imperial system better? A experiment can conclude this, the best system is the one with most intution for people. I.e people get least relative error of their guess how long objects are. So this is the use of a metric daily. There’s another side: science. Which is nother question if science should use every day metrics.

What do people and companies do? Build things like buildings. Build things made to build things, like tools. Build things that build things that build things, food? Microsoft is ii, apple is i? Gpt is ii)