3 ms·

How you use it matters a LOT and also which version you used. There's a reason that GPT-4 is up to 45 times more expensive than GPT-3.5 Turbo (the default). If

How you use it matters a LOT and also which version you used. There's a reason that GPT-4 is up to 45 times more expensive than GPT-3.5 Turbo (the default). If you're serious about such a tool (It's not a bad idea), then you'd use the embeddings[0] to train the AI on your specific use cases or prompt-engineer GPT-4 properly. [0] https://platform.openai.com/docs/guides/embeddings/what-are-embeddings https://platform.openai.com/docs/guides/embeddings/what-are-...

justinclift 3y ago
> then you'd use ... Sounds like a bunch of effort. :/
- vidarh 3y ago
  Why is why we're seeing VC funded startups starting to produce specialised tooling for various use cases so people don't have to do it themselves. ChatGPT is fun, and for some of what I do it's usable as-is, but for most people it'll be a tech demo for something that will be built in and hidden behind layers of stuff to make it suitable for their specific use-case. At least until the base models get a lot more capable.
  - permo-w 3y ago
    this is something I don’t understand. if you want GPT3/4 for your specific use case, then use the API? half the complaints people have about chatGPT are because they’ve never bothered to try the—much more powerful—API
    - justinclift 3y ago
      Oh. It's probably because that's not an obviously better progression. Something along the lines of "why bother looking at their API, when the existing thing doesn't work properly?". You're saying it's better via the API, but the standard (non-api) approach would have to be good enough first to draw people in more I guess.
    - vidarh 3y ago
      A lot of people want a polished product, and go in with expectations it will be something it isn't. I think part of the problem is people are used to equate eloquence with intelligence and knowledge, and when that assumption doesn't hold, they assume it's all smoke and mirrors rather than explore its limits.
usrbinbash 3y ago
The problem can be mitigated by using bigger models or training embeddings, but it cannot be eliminated. At the end of the day, it remains a sequence completion engine. An LLM can only care about the statistical likelyhood of sequences, that the entire universe it exists in. And if a sequence making a wrong statement happens to be more likely than, or close enough for the set "heat" of the algorithm to, a sequence denoting a correct statement, then there is a chance the model will hallucinate. If it were otherwise, we would already have a good solution for prompt injection attacks, and we don't.
- namaria 3y ago
  Large language models do a great job of emulating meaning. That's their biggest strength and greatest weakness. We got a finally good chatbot, at the cost of making people project so much more then a chatbot on it. LLMs can't mean anything. They can only produce surrogate meaning. When you compress vast amounts of text into a model and query the model you can only get back information that is already in the training data set. It's inherently backwards looking, it cannot produce lowered entropy forward in time. You can't get more knowledge out of it, only less then you inputted in the first place.
  - cookieperson 3y ago
    Also, it's important to state that, the BEST case scenario is that you get the knowledge out that went in. The common scenario would be that you have loss. It's also important to remember this thing was trained on Reddit, etc. We all know how much complete trash, bot posts, hot takes, and conflicting views there are on there.
    - namaria 3y ago
      Good point. It wasn't even trained on high grade knowledge, just on very noise bodies of language.
namaria 3y ago
I see we're reaching the 'you're holding it wrong' stage with LLMs.
- lionkor 3y ago
  It's more like "the limitations you point out are fixed by paying more money", which is the OpenAI way.
- viraptor 3y ago
  I don't see anything wrong with pointing it out. If I find a previous legal case with a heading that vaguely sounds like something that could be applicable to me and try to use it as an argument, you'd likely say the same: previous cases could be useful, but you need to know how to use that record properly.
  - namaria 3y ago
    Are you comparing legal proceedings to a chatbot?
    - viraptor 3y ago
      I'm comparing the usage of two resources which can be helpful or completely misleading, depending on how familiar you are with their purpose/limitations and your experience with them.
- rcxdude 3y ago
  It is an issue with the tech, and it's probably not an easy one to fix. Performance depends a lot on how its used, but even when performance is poor the output is often convincing.
  - cookieperson 3y ago
    The first few times I tried chat gpt I tried asking it goofy questions to try to break it. Then I asked it questions across various fields I have information on. It answered each question incorrectly in a way that bothered me. The. I tried asking it about a field I didn't know about, and googled for the answer. Wrong again. I literally never had it give me an answer that mattered. I have a colleague who is gungho chat bot coding. The guy is smart, but ever since he started using it, he's adopted some weird ideas about what code should look like which are based on easily googleable chatgpt lies and he refuses to hear anyone else out. It's kinda sad how confidently wrong chatbots can infect people and stunt their growth...

justinclift 3y ago

> then you'd use ... Sounds like a bunch of effort. :/

vidarh 3y ago
Why is why we're seeing VC funded startups starting to produce specialised tooling for various use cases so people don't have to do it themselves. ChatGPT is fun, and for some of what I do it's usable as-is, but for most people it'll be a tech demo for something that will be built in and hidden behind layers of stuff to make it suitable for their specific use-case. At least until the base models get a lot more capable.
- permo-w 3y ago
  this is something I don’t understand. if you want GPT3/4 for your specific use case, then use the API? half the complaints people have about chatGPT are because they’ve never bothered to try the—much more powerful—API
  - justinclift 3y ago
    Oh. It's probably because that's not an obviously better progression. Something along the lines of "why bother looking at their API, when the existing thing doesn't work properly?". You're saying it's better via the API, but the standard (non-api) approach would have to be good enough first to draw people in more I guess.
  - vidarh 3y ago
    A lot of people want a polished product, and go in with expectations it will be something it isn't. I think part of the problem is people are used to equate eloquence with intelligence and knowledge, and when that assumption doesn't hold, they assume it's all smoke and mirrors rather than explore its limits.

usrbinbash 3y ago

The problem can be mitigated by using bigger models or training embeddings, but it cannot be eliminated. At the end of the day, it remains a sequence completion engine. An LLM can only care about the statistical likelyhood of sequences, that the entire universe it exists in. And if a sequence making a wrong statement happens to be more likely than, or close enough for the set "heat" of the algorithm to, a sequence denoting a correct statement, then there is a chance the model will hallucinate. If it were otherwise, we would already have a good solution for prompt injection attacks, and we don't.

namaria 3y ago
Large language models do a great job of emulating meaning. That's their biggest strength and greatest weakness. We got a finally good chatbot, at the cost of making people project so much more then a chatbot on it. LLMs can't mean anything. They can only produce surrogate meaning. When you compress vast amounts of text into a model and query the model you can only get back information that is already in the training data set. It's inherently backwards looking, it cannot produce lowered entropy forward in time. You can't get more knowledge out of it, only less then you inputted in the first place.
- cookieperson 3y ago
  Also, it's important to state that, the BEST case scenario is that you get the knowledge out that went in. The common scenario would be that you have loss. It's also important to remember this thing was trained on Reddit, etc. We all know how much complete trash, bot posts, hot takes, and conflicting views there are on there.
  - namaria 3y ago
    Good point. It wasn't even trained on high grade knowledge, just on very noise bodies of language.

namaria 3y ago

I see we're reaching the 'you're holding it wrong' stage with LLMs.

lionkor 3y ago
It's more like "the limitations you point out are fixed by paying more money", which is the OpenAI way.
viraptor 3y ago
I don't see anything wrong with pointing it out. If I find a previous legal case with a heading that vaguely sounds like something that could be applicable to me and try to use it as an argument, you'd likely say the same: previous cases could be useful, but you need to know how to use that record properly.
- namaria 3y ago
  Are you comparing legal proceedings to a chatbot?
  - viraptor 3y ago
    I'm comparing the usage of two resources which can be helpful or completely misleading, depending on how familiar you are with their purpose/limitations and your experience with them.
rcxdude 3y ago
It is an issue with the tech, and it's probably not an easy one to fix. Performance depends a lot on how its used, but even when performance is poor the output is often convincing.
- cookieperson 3y ago
  The first few times I tried chat gpt I tried asking it goofy questions to try to break it. Then I asked it questions across various fields I have information on. It answered each question incorrectly in a way that bothered me. The. I tried asking it about a field I didn't know about, and googled for the answer. Wrong again. I literally never had it give me an answer that mattered. I have a colleague who is gungho chat bot coding. The guy is smart, but ever since he started using it, he's adopted some weird ideas about what code should look like which are based on easily googleable chatgpt lies and he refuses to hear anyone else out. It's kinda sad how confidently wrong chatbots can infect people and stunt their growth...