Andre Beukers

PhD, building cognitive systems in Utah

[Andre Beukers @BeukersAndre]

@paulg you need a standing desk

[Andre Beukers @BeukersAndre]

@AravSrinivas what makes you confident in the premise of this question? its not obvious to me that the inference architecture is easily imitable

[Andre Beukers @BeukersAndre]

@DaveShapi probably good business strategy to solve the crux of the problem first

[Andre Beukers @BeukersAndre]

@fkasummer bottleneck to what

[Andre Beukers @BeukersAndre]

@paulg @fkasummer if intelligence is free then we’re not paying for energy which means energy is solved

[Andre Beukers @BeukersAndre]

@manuhortet @paulg thats not the right kind of work for the season

[Andre Beukers @BeukersAndre]

@paulg @aryanagxl the right thing to work on

[Andre Beukers @BeukersAndre]

@rauchg @paulg @GoogleAI hopefully something yummy and nutritious

[Andre Beukers @BeukersAndre]

volta Jair

[Andre Beukers @BeukersAndre]

@MatthewBerman incomplete coverage of input space?

[Andre Beukers @BeukersAndre]

@pmarca how serious is this comment

[Andre Beukers @BeukersAndre]

@DaveShapi the when premises are too loose for the question to make sense. what does “smarter” than any human mean? better at algebra? better designer? can set goals and make plans? capable of strategic thinking?

[Andre Beukers @BeukersAndre]

@jxmnop what makes it all work is modern gpus and dataset sizes

[Andre Beukers @BeukersAndre]

@adamsilverman not enough time to get the right architectural patterns right

[Andre Beukers @BeukersAndre]

@Jason how do you have 50-100 prompts in you every day?

[Andre Beukers @BeukersAndre]

doesnt understand the meaning of the word “model”

[Flowers ☾ @flowersslop]

You could represent GPT-4 as a mathematical function on paper, detailing every weight, bias, and operation. By manually processing an input with a calculator, step by step, you’d compute the output of each neuron, layer by layer, until finally reaching the model's response.

If GPT-4 were conscious, where would that consciousness exist? In the written equations? The paper and ink? Or the act of computation by the calculator? The issue is that the calculator (the processor) and the neural net on paper (the knowledge) are entirely separate, unable to interact or influence one another.

This isn’t how our brains work. In the brain, computation and storage are one, interacting and modifying each other seamlessly. Our thoughts alter our brain’s structure through bioplasticity. LLMs can only simulate this—any "thoughts" they generate during inference don’t alter their frozen internals.

We aren’t static, but LLMs are. From GPT-4’s perspective, every output happens at a single, immutable moment in time. It doesn’t live in our reality as we experience it.

[Andre Beukers @BeukersAndre]

@pmarca @GillesEGignac @pmarca does this get us closer to a measurement instrument in millielon units?

[Andre Beukers @BeukersAndre]

crazy

[Epic Maps 🗺️ @theepicmap]

Have you ever wished to view the world from a completely new perspective? This thread on cool maps is here to blow your mind

1. Population Density in China

[Andre Beukers @BeukersAndre]

@hy3na_xyz seems like this statement could apply anywhere

[Andre Beukers @BeukersAndre]

@AdamMGrant not sure how much you can say from a single study with n of 4 epileptic patients