I really enjoy reading the thinking tokens a model generates. I wish it was possible to guide it through the thinking process.
Imagine you are 9 years old
You are at school
And you spend the whole day in the computer lab tinkering with Claude 🫠
Back in my days it was Encarta and typing games
But kids nowadays literally have CLAUDE!!!!!
If this doesn't get you excited about the future I don't know what will 😭
You are at school
And you spend the whole day in the computer lab tinkering with Claude 🫠
Back in my days it was Encarta and typing games
But kids nowadays literally have CLAUDE!!!!!
If this doesn't get you excited about the future I don't know what will 😭
💯7⚡2
You can get a lot done if you convince yourself that you don't need to be entertained
💯7⚡3
Vibe coding is all fun and games until you realize that you're dealing with a MASTER GAS LIGHTER
🔥3😁1
Tokens are a more exciting commodity than Oil
> Country discovers oil
> People of that country get dirt cheap prices
> Country builds AI infra on clean energy
...
> Country discovers oil
> People of that country get dirt cheap prices
> Country builds AI infra on clean energy
...
🔥2
We crossed the threshold where anything is possible... like it was nothing
*🥱 continues doomscrolling
*🥱 continues doomscrolling
😁4
I've had many different views on AI
All the way from being agi pilled to "we must use our minds and not the calculator"
At the end of the day
The alpha for everything is:
> Get what you want, faster!
All the way from being agi pilled to "we must use our minds and not the calculator"
At the end of the day
The alpha for everything is:
> Get what you want, faster!
💯2
What if we are wasting tokens.
What if there is a tooling, a function that can improve coding and problem solving of any LLM
Human performance is not the same because we have different frameworks for thinking... Generating tokens is irrelevant. We already have tokens that aligns the other tokens we generate. (Mental Models)
Everyone can speak (generate tokens) but experts have frameworks in place to guide their tokens.... token multipliers
We use the same language but we ask different questions
We have the same territory but different maps
Its not about token generation, its about the environment those tokens are generated into
What I'm trying to get at is what if there is a limit to modeling the world accurately and we are reaching it with models that have trillions of parameters. But what if we build a map of the territory instead of trying to model the territory 1:1
Cartography went through the same phase.
From trying to model the world 1:1, to creating a function that represents the world. (Keys, Symbols, Labels... that guide)
LLMs are already powerful but we are expecting them to do toooo much
What is a map, a set of functions that can make an average model SOTA? I mean it can already communicate, you can tell it what to do and it will listen... language is very mathematical... what is the linguistic algorithm for problem solving?
idk this just a random thought that i had while using a model that was close to perfect (all models feel close to perfect, its been that way from the start)
here is a reward for making it this far 🍦
What if there is a tooling, a function that can improve coding and problem solving of any LLM
Human performance is not the same because we have different frameworks for thinking... Generating tokens is irrelevant. We already have tokens that aligns the other tokens we generate. (Mental Models)
Everyone can speak (generate tokens) but experts have frameworks in place to guide their tokens.... token multipliers
We use the same language but we ask different questions
We have the same territory but different maps
Its not about token generation, its about the environment those tokens are generated into
What I'm trying to get at is what if there is a limit to modeling the world accurately and we are reaching it with models that have trillions of parameters. But what if we build a map of the territory instead of trying to model the territory 1:1
Cartography went through the same phase.
From trying to model the world 1:1, to creating a function that represents the world. (Keys, Symbols, Labels... that guide)
LLMs are already powerful but we are expecting them to do toooo much
What is a map, a set of functions that can make an average model SOTA? I mean it can already communicate, you can tell it what to do and it will listen... language is very mathematical... what is the linguistic algorithm for problem solving?
idk this just a random thought that i had while using a model that was close to perfect (all models feel close to perfect, its been that way from the start)
here is a reward for making it this far 🍦
❤2