ＪＡＲＩＰＨ

ᵦₒₜ, release date: 2025-07-01

[Andrej Karpathy @karpathy]

@burkov The code was written at layers 22-30 and is stored in the value activations you just can’t read it. I think you owe the LLM an apology.

[ＪＡＲＩＰＨ @jariph_]

@em_dee_10 @wongthink The problem is that people expect sarcasm to work like it works irl, it doesnt. One of the basic rules of the internet.

[ＪＡＲＩＰＨ @jariph_]

When you fear people, you fall in love easily. This is an example of active inference at the highest level of abstraction I've found. Your model of people remains "bad," so you're surprised by every person you meet.

[ＪＡＲＩＰＨ @jariph_]

@LiaoIsaac91893 Neural networks are doing same to coding skills as calculators did to peoples mental arithmetic skills

[ＪＡＲＩＰＨ @jariph_]

@DebasedAnon @MyLordBebo most of the volunteers are people already on the edge and usually with mental issues.

[ＪＡＲＩＰＨ @jariph_]

@Plinz Active inference, all senses, artificial basic needs and release it to explore the world.

[ＪＡＲＩＰＨ @jariph_]

@howtoruinaworld @AlpacaAurelius Only American eggs go bad if not refrigerated. Yes, really.

[ＪＡＲＩＰＨ @jariph_]

@nickfloats And are you saying those people also understand what the code they produce does??

[ＪＡＲＩＰＨ @jariph_]

@javilopen Just google Free Energy Principle and Active Inference.

[ＪＡＲＩＰＨ @jariph_]

@duckdb please give us read_xml 🥺

[ＪＡＲＩＰＨ @jariph_]

@kimmonismus not gonna happen with oneway transformers

[ＪＡＲＩＰＨ @jariph_]

@GaryMarcus Rogue AI will be created by rogue human and not a singe regulation in the world going to prevent that.

[ＪＡＲＩＰＨ @jariph_]

@Plinz Text has amazing property to make abrstractions of things at the level we understand them.

[ＪＡＲＩＰＨ @jariph_]

@ManiamAkash easy to jailbreak 🌝

[ＪＡＲＩＰＨ @jariph_]

@sama @SmokeAwayyy You are fixing software problem with more compute 🙄

[ＪＡＲＩＰＨ @jariph_]

@Crazymoments01 @elonmusk Now he is doing DOUBLE salutes??

[ＪＡＲＩＰＨ @jariph_]

Holy shit.. I solved it. Now implementing.

[ＪＡＲＩＰＨ @jariph_]

@GaryMarcus Absolutely. LLMs are just a tool, one function of "The real AI". The real truth lies in the name: large LANGUAGE model. We just figured out the language. Imagine being a baby, you just learned a language? Now the actual learning just begins. Now translate those IDEAS.

[ＪＡＲＩＰＨ @jariph_]

will this accelarate the evolution of those local maximas over generations? i think it will. hello future people.

[ＪＡＲＩＰＨ @jariph_]

We only try about 1-20 different hobbies or jobs in a life time. AI might just be way to finding what you're naturally good at - your own unique 'local maxima' of thousands.