Tim

[Tim @timothyngw]

@odin_free @letonyo Yes. Just like cars and planes didnt make walking obsolete. We just get to places faster and farther now. But now in the space of ideas.

[Tim @timothyngw]

Being able to see what is Invariant and variant. Being able to see what are the same and different.

[Tim @timothyngw]

The default is happy. Unsatisfaction, envy, greed, laziness, etc makes one unhappy and nihilistic

[Tim @timothyngw]

It’s not about finding what will make you happy. It’s about finding what is making you unhappy.

[Thomas Massie @RepThomasMassie]

Tonight, Congress will vote to lower the age to prosecute minors as adults from 16 down to 14 in DC.

Let me get this straight:

Congress wants to prosecute 14 yr. olds as adults, but they don’t want to prosecute adults who sexually abuse 14 year olds?

Release the Epstein files.

[Tim @timothyngw]

Sometimes the hardest part is just overcoming one’s own laziness

[Tim @timothyngw]

@Thom_Wolf It’s still something like P vs NP. Difficult problems that requires creativity are like problems in NP. If “AGI” is achieved, it should be able to solve something that humans can’t easily. A natural eval is then something like: random NP instance, AI generate alg, and we verify.

[Tim @timothyngw]

@LauraLoomer https://t.co/S6VC29cCTq

[Tim @timothyngw]

LLMs seem to be decent translators/compilers. Human language (HL) to programming language (PL). PL to PL. PL to HL. HL to HL and so on. The hard part is making them robust and reliable

[Tim @timothyngw]

@elyxlz Actually found this article: https://t.co/R1X5Ut8SM1

[Tim @timothyngw]

@elyxlz Something about non-differentiability of the environment/reward too. Policy gradient seems like a gradient free optimization (0th order opt) where randomness helps with estimating a gradient. Is this intuition correct?

[Tim @timothyngw]

@hallerite @novasarc01 Yea this was a weird part in the paper

[Tim @timothyngw]

@iAnonPatriot @catturd2 What is a good alternative?

[Tim @timothyngw]

@cloneofsimo Data

[Tim @timothyngw]

@amasad Unreliable.

[Tim @timothyngw]

"AGI" is like teleportation. I think we should build cars and planes first. And most importantly, we need good drivers to get us to good destinations. Then we can be explorers and perhaps take a leisure road-trip across the beautiful landscape of the mind.

[Tim @timothyngw]

Foundation models (e.g. LLMs) to the AI revolution's is like the steam engine to the industrial revolution, but in the cognitive (information) space rather than the physical space.

[Tim @timothyngw]

However, a car wouldn't be of much use with just its engine. We need to build the transmission, the wheels, and everything else to get our cognitive car up and running, and make everything reliable and more efficient. There's still a lot of work to be done.