Egregoros

@jackdaniel @khm @prahou @rl_dane so where do we go from here? Because if people keep ramping up this rhetoric the only option is to abandon trust by default. Now you're a team developing in the open, but you're not openly accepting anything from people you don't know.

I know a many people across a bunch of very high profile projects that are using LLMs in their workflows now and just not disclosing it because they don't want to deal with the drama. The code is passing the smell/review tests and getting merged because there's nothing wrong with it: it does what it's supposed to, it follows the project's existing standards and style guidelines.

Good luck finding out which developers are doing it and banning them from the projects. :)

Unless there's a legal precedent that the generated code is definitely a violation or now must have a viral license attached with it, nothing will change. I have very very strong doubts about both of those happening.

Why?

Because this is *our* legal system. These are our own chains we'd be putting on ourselves. And our governments will see this very differently. If our competitors in the global market will not be enforcing these laws on their citizens, why should we do it to ourselves? They'll be able to build faster than we can because they're not weighed down by a legal albatross that in many respects is just silly and incompatible with modern technology.

@feld @khm @prahou @rl_dane

this is a tired line of argumentation. extensive LLM usage contributes to climate crisis and society crisis.

Laws are for us humans, indeed, and if some activity brings more harm than good, then it should be regulated or outright banned.

This is just a side note, but after people stop creating things on their own, models will rather quickly degrade, that's why llm companies are so desperate to suck in any "organic" data from the internet, even at the cost of scraping the whole internet over and over again.

@jackdaniel @khm @prahou @rl_dane

> this is a tired line of argumentation. extensive LLM usage contributes to climate crisis and society crisis.

this is fearmongering and a lie. LLMs don't *require* a crazy amount of energy or compute resources to operate; this is simply the current status because of very poorly optimized inference engines.

Everyone in the industry knows it, but it's the thing you just don't say out loud because the industry relies on this circle jerk of money flowing between Nvidia, OpenAI, etc etc.

What happens to OpenAI if you could host a thousand concurrent users on a single server hosted in your own company's datacenter?

What good is it for Nvidia to follow through on their research papers pre-2020 discussing how to build these inference engines properly to reduce resource requirements if it's going to destroy your market cap because now nobody needs to order all these monster GPUs?

Right now, the waste is intentional, but the new engines will evaporate all of this.