Live data from Hacker News

What is LLMs and why is it so important?

news.ycombinator.com

1–8 of 8 posts

Re: What is LLMs and why is it so important?

#2
domain/llms.txt is like a robots.txt for telling LLMs how to best interact with the site. The majority of "LLM" discussion on HackerNews has nothing to do with these instruction files though. LLM, or Large Language Model, is what people outside of the tech community are calling "AI". A lot of discussion is happening on HN about LLMs because they are the hot new technology. If they can fulfill their promises, the could revolutionize the world. Currently they are pretty good at producing decent looking text and code, but if you really dig into the outputs you'll find it is often quite sloppy.

Re: What is LLMs and why is it so important?

#3

domain/llms.txt is like a robots.txt for telling LLMs how to best interact with the site. The majority of "LLM" discussion on HackerNews has nothing to do with these instruction files though. LLM, or Large Language Model, is what people outside of the tech community are calling "AI". A lot of discussion is happening on HN about LLMs because they are the hot new technology. If they can fulfill their promises, the coul…

Are they better than modern open source AI models or are models themselves are LLms

Re: What is LLMs and why is it so important?

#4

domain/llms.txt is like a robots.txt for telling LLMs how to best interact with the site. The majority of "LLM" discussion on HackerNews has nothing to do with these instruction files though. LLM, or Large Language Model, is what people outside of the tech community are calling "AI". A lot of discussion is happening on HN about LLMs because they are the hot new technology. If they can fulfill their promises, the coul…

Are they better than modern open source AI models or are models themselves are LLms

The models themselves, like GPT from ChatGPT, Opus from Anthropic, etc. are just the "brands" of LLMs. But there are local models, that run on your own hardware as well like Qwen from Alibaba and Kimi K from Moonshot.

Kind of like how Pepsi and Coke are brands of cola.

Re: What is LLMs and why is it so important?

#7
LLMs are trained on large-scale text and code data, and fundamentally, they predict the probability distribution of the next token given the preceding tokens. There are studies showing that even with this simple learning objective, grammatical and semantic features, as well as relationships between concepts, emerge in the internal representation space in a structured geometric form. However, I understand it's more accurate to say that rather than the model retrieving sentences from a separate 'meaning space,' it converts the prompt into a contextual vector representation and then sequentially generates the next tokens from that state. I could be wrong about this.

Since code is fundamentally built on 'patterns' and mostly follows established conventions, it becomes easier to predict, which makes LLMs very effective for programming. After all, programming is designed to converge toward specific patterns. Frameworks with IoC are a good example of that structure.

Recently, even Linus Torvalds has acknowledged AI as a useful tool, so it's probably good to use it. That said, a lot of people still hate LLMs because they're afraid of losing their jobs

Re: What is LLMs and why is it so important?

#8

Earlier quoted context omitted.

Are they better than modern open source AI models or are models themselves are LLms

The models themselves, like GPT from ChatGPT, Opus from Anthropic, etc. are just the "brands" of LLMs. But there are local models, that run on your own hardware as well like Qwen from Alibaba and Kimi K from Moonshot. Kind of like how Pepsi and Coke are brands of cola.

That clears doubts