3 ms·

> With LLMs it's very similar but rather than a meaningful shift in technological development underpinning it it's something more like a parlor trick with a few

> With LLMs it's very similar but rather than a meaningful shift in technological development underpinning it it's something more like a parlor trick with a few practical applications. What makes you think this, I'm assuming you've tried gpt-4? I've personally gotten a lot of productivity out of it. Assuming it gets a lot better, I don't see how it's not going to minimally become an indispensable tool, and at maximum completely transform the future of humanity.

pydry 3y ago
That would be a safe assumption, yes. Im clearly not a complete moron. It reduced my productivity when I used it. Between having a slightly faster and slightly customized version of google vs. being sent down misleading rabbit holes, the latter won out. I found a few niche use cases where it proved useful but it was all out of proportion to the hype. I've paired with other people who were bullish about LLMs using them and witnessed them falling down rabbit holes. I found the whole experience bizarre - it was like seeing a new "social reality" take precedence over what was happening in front of our eyes. Computer says yes.
- xcv123 3y ago
  LLMs are neural networks. If you understand their strengths and weaknesses they are extremely useful. Use LLM's to explain abstract concepts, discuss ideas, analyze text, as an interactive tutor, not as a datastore of facts to be recited verbatim (like a search engine). That is the worst possible way to use LLMs and will result in hallucinated facts. If that's how you're doing it then you're doing it wrong. Neural networks such as LLMs are fundamentally not made for factual recall. LLMs are designed for solving natural language understanding tasks. I have found GPT-4 very useful for understanding concepts and solving specific problems in programming, science, maths, psychology, relationships, and producing creative writing, by having conversations with it, going back and forth deeper into these topics. But I would never use it as an API reference. Raise it up to the conceptual level and you will be surprised at what it is capable of.

pydry 3y ago

That would be a safe assumption, yes. Im clearly not a complete moron. It reduced my productivity when I used it. Between having a slightly faster and slightly customized version of google vs. being sent down misleading rabbit holes, the latter won out. I found a few niche use cases where it proved useful but it was all out of proportion to the hype. I've paired with other people who were bullish about LLMs using them and witnessed them falling down rabbit holes. I found the whole experience bizarre - it was like seeing a new "social reality" take precedence over what was happening in front of our eyes. Computer says yes.

xcv123 3y ago
LLMs are neural networks. If you understand their strengths and weaknesses they are extremely useful. Use LLM's to explain abstract concepts, discuss ideas, analyze text, as an interactive tutor, not as a datastore of facts to be recited verbatim (like a search engine). That is the worst possible way to use LLMs and will result in hallucinated facts. If that's how you're doing it then you're doing it wrong. Neural networks such as LLMs are fundamentally not made for factual recall. LLMs are designed for solving natural language understanding tasks. I have found GPT-4 very useful for understanding concepts and solving specific problems in programming, science, maths, psychology, relationships, and producing creative writing, by having conversations with it, going back and forth deeper into these topics. But I would never use it as an API reference. Raise it up to the conceptual level and you will be surprised at what it is capable of.