It’s funny that you think these things are operating on napkin math. I guess it takes tens of thousands of GPUs and months of compute time to work out some napkin math.

You’re confusing a lot of things and it sounds like when you say ‘AI’ you mean ‘ChatGPT’ or generative LLM/Diffusion models. Because those large models are the only ones that use a large amount of computer resources. That doesn’t represent the entirety of neural network based machine learning (AI)

My phone uses an LLM for spellcheck, it even runs fine tuning (the thing that you’re referring to that requires ‘tens of thousands of GPUs’) on the phone processor. Writing and training text recognition AI from scratch is literally an exercise for Computer Science students to do on their personal computer. AI running object detection models runs on tiny microprocessors inside of doorbells.

My response was that you cannot eliminate AI because the algorithms that are required to build it are already known by millions of experts who can re-create it from scratch.

All of these years of AI research resulted in some napkin math that we were missing all along

As much as you’re trying to be sarcastic. Yes, that is correct.

This is not unusual in science, here’s some other napkin formulas that required years, decades and centuries to discover.

E=MC² – Years of knowledge and research to discover how mass and energy were related

√-1 = i – 1800 years is how long it took for this to be accepted

F = ma – Newton’s force equation

e^iπ + 1 = 0 – Euler’s Identity

E = hf – Plank’s discovery describing how photon energy links to frequency.

etcetc.

What does ‘AI dying’ even look like to you? If you were the dictator of Earth, how would you eliminate the knowledge from the minds of millions of experts across the world?

It’s not just math or if it is, we don’t understand the math. Math is deterministic. These models are not deterministic, an input does not always produce the same output and you can’t feed a response backwards through a model to produce the query. We struggle to make even remotely predictable changes to a model when it does something we don’t like.

Once again you’re confusing topics and also definitions. Determinism, interpretability and explainability are different things.

Neural networks are completely deterministic. A given input will always return the same output.

You’re probably referring to the trick that the LLM chatbots use where they take the output of the the model, which is a list of tokens with a score of which is most likely, and randomly select a token from that list before feeding it back into the model. This is a chatbot trick, it doesn’t happen in the model.

No machine learning model uses randomness during inference. Often people pass in random noise so the output varies, but if you pass in the same random noise then you get the same output.

‘We don’t know how it works’ isn’t exactly true. There’s a lot of work in this field (https://en.wikipedia.org/wiki/Explainable_artificial_intelligence) and there’s nothing fundamentally unknowable about these systems.

The AI industry can die like any other.

Yes, like I said, ‘The AI industry’ that produced generative LLMs and diffusion models is what you’re upset about, you’re upset that capitalists are using AI to fire workers and destroy jobs.

You’re not upset that we’ve discovered the ability to create universal approximation functions or use machines to learn from data.

You’re mad at capitalism, not AI.

Using your logic, anyone and everyone can build nukes, they’re just math and physics and the materials (like GPUs and Power) are easy to come by. We can’t erase the knowledge so let’s all sit back and enjoy the fallout.

This isn’t an argument, this is: https://en.wikipedia.org/wiki/Reductio_ad_absurdum

I said you can’t eliminate the knowledge of AI from society since anyone with a laptop can train one from scratch and the knowledge to do so is available to everyone. The knowledge for making nuclear weapons was also not eliminated despite being far more dangerous and widely condemned.

Also, irrelevant to my point but, in what world are the materials to make nuclear weapons easy to come by?

but AI is a threat even without capitalism. So far they’re making life more expensive in multiple ways for all of us with no real benefit for their existence.

No real benefit? Once again, you seem to be talking about generative AI. That’s a product, created by capitalists, using AI. It isn’t the entirety of the AI field, it isn’t even the field with the most AI workers.

AI is used in science and has facilitated incredible discoveries already.

AlphaFold revolutionized structural biology by predicting the shape of every known protein which has massively accelerated drug discovery. I’m not sure if you’re aware but there are now TWO AIDS vaccines in human trials and the researchers use machine-learning models to mine clinical data in order to spot patterns in immune responses leading promising therapies.

AI is being used to plan, execute and interpret lab work, this is a tedious and laborious task that requires a highly trained person, typically a grad student. This isn’t something that you can scale up 100x or 1000x because you can’t magic 1000 graduate students into existence. Now you can use AI to do the tedious work and a grad student can now run multiple times are labs which are at the heart of almost every scientific discovery.

Diagnostic AI, which is more objectively more accurate than human experts, is used to annotate diagnostic images in order to indicate to a doctor areas to examine. This results in lower error rates and earlier detection than human-only review.

It’s discovered new plasma physics which are key for having working fusion power. (https://www.pnas.org/doi/10.1073/pnas.2505725122)

So, while you may not think these are worth the downsides, it’s disingenuous to say that there has been no benefit.