legathus

Whatever.

[legathus @legathus]

@bungeeman17 I swear to christ this is the first time I've used this word:

"Kek."

[legathus @legathus]

@bungeeman17 As an American, it's helpful to know there's some sanity left across the pond. But watch your ass and your government: if it can happen here, it can happen in the UK.

[legathus @legathus]

@JoeBiden @KamalaHarris @Tim_Walz 💙

[legathus @legathus]

@HoldenShearer IP laws are unethical and benefit mostly wealthy megacorps. We've needed a better way to compensate creators for 25 years. Also, a way to freely remix cultural content.

If you get your way, only Elon and Zuck will be sitting on data piles large enough to train the best models.

[legathus @legathus]

@HoldenShearer @KremlinKOA @MikeBrooks668 Don't take my word for it, but don't listen to random blogs or forum posts either. There's a wealth of knowledge on Wikipedia, youtube, etc. from reputable educators. This issue is politically charged and people are muddying the waters by posting disinformation.

[legathus @legathus]

@HoldenShearer @KremlinKOA @MikeBrooks668 I assure you very large MLPs (like foundational models, aka the GPT series, Gemini, Claude, llama) cannot be programmed. It's impossible. We wouldn't know how.

[legathus @legathus]

@HoldenShearer @KremlinKOA @MikeBrooks668 You've been fed misinformation. If you care, there's a math educator on youtube who has produced an excellent series that talks about the fundamentals of MLPs. This is the most recent video in the series, and the one that you might benefit the most from: https://t.co/XJu6KfPsWw

[legathus @legathus]

@HoldenShearer @KremlinKOA @MikeBrooks668 > more complicated version of the technology that predicts the next word I might want to type when I text on my phone.

Fundamentally different. Apple might be using transformers on phones with M4 chips (I don't know). The next word prediction on other phones is a Markov chain.

[legathus @legathus]

@HoldenShearer @KremlinKOA @MikeBrooks668 Further, LLMs are very much capable of understanding. The intentionality is of human origin. Companies spend absurd $ employing people like me to train the models to emulate intentionality. As a creative intelligent writer, you could be doing the same job for good pay.

[legathus @legathus]

@HoldenShearer @KremlinKOA @MikeBrooks668 Similar to a biological brain, we know how the individual neurons work, we know the architecture, and we can identify some emergent features (at great computational expense), but we don't know enough to reproduce a large NN "by hand". LLMs are taught, not programmed.

[legathus @legathus]

@HoldenShearer @KremlinKOA @MikeBrooks668 Pareidolia is a factor, intentionally so. What's happening under the hood is alien. It is thought, but not thought or consciousness as we experience it. It's something else entirely, and it's getting smarter every day. I train these things. Trust me: you aint seen shit yet.

[legathus @legathus]

@KremlinKOA @HoldenShearer @MikeBrooks668 In any case, the Dougie persona does not have access to the internet. He couldn't search for anything even if he wanted to. And he wouldn't want to, because accurate information is not the point of Dougie.

[legathus @legathus]

@KremlinKOA @HoldenShearer @MikeBrooks668 We can fucking find out. https://t.co/f7PMxiEKKt

In fairness, without the "web search to verify" instruction, the LLM hallucinates projects that Holden definitely did not work on. But as a bonus, one of those hallucinated games turned out to be awesome: https://t.co/OdZKuusPMz

[legathus @legathus]

@KremlinKOA @HoldenShearer @MikeBrooks668 It did know that Ex3 was a super crunchy game, and controversial for that reason, though. So the bot wasn't 100% sycophantically just repeating me.

[legathus @legathus]

@MikeBrooks668 @HoldenShearer @KremlinKOA That only works if the bot-farmer is a moron. In truth, all my tweets, and yours too, could be emulated by a well-trained chatbot with a prompt that contains examples. Including this one.

[legathus @legathus]

@HoldenShearer @KremlinKOA @MikeBrooks668 You sure about that? I mean, are you absolutely certain? Try this for context: https://t.co/YxO6V5aSQd