9 ms·

I wish people would focus on these exceptional strengths of the model rather than blabbing about AGI or whatever. Similarly to you, I have been able to find is

I wish people would focus on these exceptional strengths of the model rather than blabbing about AGI or whatever. Similarly to you, I have been able to find issues with logs, formatting, asking it quick query questions in [whatever flavor of query language XYZ service likes to use], etc.. and it's really, really good. The alternative is to muscle through it, using a lot of energy, writing my own parser or something dumb, or to use Google - which basically isn't usable anymore! But you have people who are like "GPT CODED MY ENTIRE WEBSITE" and "GPT TAUGHT ME QUANTUM PHYSICS" and I'm like... uh... big doubt my man...

heliophobicdude 4y ago
Like Simon said, it's a great language calculator! Also, adapting to its quirks is important. Using the responses from both you and it to scaffold the final response is more effective than giving only one shot
sharemywin 4y ago
I think it does speed up learning a new language(programming) or library. It's not perfect but it's a good companion to a tutorial. or next step past a basic tutorial. especially sample code. I created a script to connect to my email address and pull down my emails. connect to HN and pull done articles to put in sqlite. really quick features, not because it's complicated but because it's tedious.
- roflyear 4y ago
  Right but you have to be really careful with that use.
  - spacebanana7 4y ago
    Yeah, it only takes a small hallucination to delete emails instead of copy them
    - 13years 4y ago
      lol, indeed. As long as the hallucination problem remains, I think we are going to see a significant hype bubble crash within a year or so. Yes, it is still useful under proper guidance, but building things on full automation that are reliable doesn't seem to be something that is actually within realms of reality at present. More innovations will be required for that.
      - spacebanana7 4y ago
        Never underestimate a hype bubble ;) The crypto one lasted for many years with far fewer use cases than LLMs, even accounting for hallucinations.
        outworlder 4y ago
        We don't even have to go to crypto. AI has had many boom/bust cycles. The term "AI Winter" dates back from the 80s! Of course, at every cycle we get new tools. The thing is, once tools become mainstream, people stop referring to them as "AI". Take assistants (Amazon Echo, Cortana, Siri, etc). They are doing things that were active areas of "AI" research not long ago. Voice recognition and text to speech were very hard problems. Now people use them without remembering that they were once AI. I predict that GPT will follow the same cycle. It's way too overhyped right now (because it's impressive, just like Dragon Naturally Speaking was). But people will try to everyday scenarios and – outside of niches – they will be disappointed. Cue crash as investments dry up. Hopefully this time we won't have too many high profile casualties, like what happened with Lisp Machines.
        sharemywin 4y ago
        I never upgraded to pro and I spent like $2 on credits so far this month. I could easily see them hitting a $1b/yr which to me isn't niche market or a hype bubble.
        spacebanana7 4y ago
        A large profitable business can still be overpriced or inflated by a bubble. Like Cisco/Intel in the 90s
ravenstine 4y ago
Yes, especially given that nobody can adequately define what AGI even means. But hey, if we did, then we couldn't have so many meandering, unproductive conversations about it on Fridman or Rogan. Recently, I used GPT-4 to read my cousin's writings on his fictional world and generate a chart of the timeline and concepts using Mermaid syntax. I think one of the best things about LLMs at the moment is that it can convert things in an abstract way. Even if it doesn't get it totally right the first time, it can correct itself on instruction, and still saves time over coding something or downloading software.
- tunesmith 4y ago
  Were you able to paste his writings all in one prompt without exhausting token space? Or did you have to do something tricky to get around that?
  - ravenstine 4y ago
    The summary of his world building wasn't super huge, so it was just large enough to fit the ~8000 token limit for the GPT-4 model on OpenAI after I trimmed it a bit. I, too, would like to know how to properly get around these technical limitations.
    - tunesmith 4y ago
      I know it is something along these lines: Install a vector database. Use the API to get vector embeddings for the manuscript, by getting them in chunks. Apparently this is possible the with API even though it's not with normal ChatGPT? Then, think of the query you want to ask. Use the API not to answer the query, but to get the vector embedding for the query. Then, do a search in the vector database to get the vectors that are "near" the vector for the query. Then, finally send the query's vector, and all the "near" vectors to the API. And then you'll get your answer. I don't know how to do any of that yet. So far it seems like milvus might be the easiest vector db to install locally. But vectors for text passages are very large, so I'm not sure why I'd expect the final query of multiple vectors to be small enough for the token limit. And I'm not really sure yet how to send a query to ChatGPT in vector form. (Ideally this could work against an open model that isn't ChatGPT.)
typon 4y ago
That's actually the hidden secret. Millions of people are using it to solve their problems quietly - something Google and other companies whose lunch they're eating should be worried about rather than being worried about grand ideas like "AI Safety". Pretty much all of my non-tech family is using it in ways that I never expected: making meal plans, writing school essays, making assignments for students, writing emails to CRA (Canadian Tax Agency), fixing broken english and translation, etc.
- bolanyo 4y ago
  I'm quite curious about some of these applications because I don't really understand what the cognitive work is that's being offloaded. Making meal plans: what is being done here? making a list of meals to eat each day of the week? isn't this just a question of thinking what one would like to eat? why is it easier to have the meals chosen by someone else? Writing school essays: what is the point of this? Aren't school essays only written in order to learn to write, or to learn about some other topic? Writing emails to CRA: presumably you have to put all the pertinent information in the prompt. Can't you just copy that prompt into an email? (The other couple do seen to make sense to me, fair enough.)
  - darkwater 4y ago
    > Making meal plans: what is being done here? making a list of meals to eat each day of the week? isn't this just a question of thinking what one would like to eat? why is it easier to have the meals chosen by someone else? Do you have children? Meal planning can be a quite tedious and frustrating task if you want to cook at home, eat healthy, eat tasty, vary the dishes and have meals that kids will accept.
    - Bootvis 4y ago
      In addition, for recipes, it’s just a better Google. If I do “Give me a concise recipe for X” it gives me one. No fluff, no ads. Just ingredients and steps. For example. I asked for pasta carbonara, concise and then even more concise. Final result: Quick Carbonara (4 servings) Ingredients: 12 oz pasta 4 eggs 1 cup cheese 8 oz bacon 4 garlic Salt, pepper Parsley (opt) Cook pasta, save water. Mix eggs, cheese. Fry bacon, garlic. Combine, mix, season. Serve. Great if you’re grocery shopping and want to make sure you don’t forget anything.
calenti 4y ago
GPT generated clickbait for me! Ads at 11!
RoyGBivCap 4y ago
I pasted in a code snippet and asked if it was threadsafe (I already knew it wasn't) and it confirmed what I knew, and gave a simpler solution that worked for my particular use case that what I would have done. Anything to do with threads is hard™ and it did it. Though I think the big thing less tech savvy users have yet to realize is: GIGO still applies. If your prompts are bad, the output is bad. And if you didn't know your prompt was bad, it's almost impossible to know the output is too.

heliophobicdude 4y ago

Like Simon said, it's a great language calculator! Also, adapting to its quirks is important. Using the responses from both you and it to scaffold the final response is more effective than giving only one shot

sharemywin 4y ago

I think it does speed up learning a new language(programming) or library. It's not perfect but it's a good companion to a tutorial. or next step past a basic tutorial. especially sample code. I created a script to connect to my email address and pull down my emails. connect to HN and pull done articles to put in sqlite. really quick features, not because it's complicated but because it's tedious.

roflyear 4y ago
Right but you have to be really careful with that use.
- spacebanana7 4y ago
  Yeah, it only takes a small hallucination to delete emails instead of copy them
  - 13years 4y ago
    lol, indeed. As long as the hallucination problem remains, I think we are going to see a significant hype bubble crash within a year or so. Yes, it is still useful under proper guidance, but building things on full automation that are reliable doesn't seem to be something that is actually within realms of reality at present. More innovations will be required for that.
    - spacebanana7 4y ago
      Never underestimate a hype bubble ;) The crypto one lasted for many years with far fewer use cases than LLMs, even accounting for hallucinations.
      - outworlder 4y ago
        We don't even have to go to crypto. AI has had many boom/bust cycles. The term "AI Winter" dates back from the 80s! Of course, at every cycle we get new tools. The thing is, once tools become mainstream, people stop referring to them as "AI". Take assistants (Amazon Echo, Cortana, Siri, etc). They are doing things that were active areas of "AI" research not long ago. Voice recognition and text to speech were very hard problems. Now people use them without remembering that they were once AI. I predict that GPT will follow the same cycle. It's way too overhyped right now (because it's impressive, just like Dragon Naturally Speaking was). But people will try to everyday scenarios and – outside of niches – they will be disappointed. Cue crash as investments dry up. Hopefully this time we won't have too many high profile casualties, like what happened with Lisp Machines.
      - sharemywin 4y ago
        I never upgraded to pro and I spent like $2 on credits so far this month. I could easily see them hitting a $1b/yr which to me isn't niche market or a hype bubble.
        spacebanana7 4y ago
        A large profitable business can still be overpriced or inflated by a bubble. Like Cisco/Intel in the 90s

ravenstine 4y ago

Yes, especially given that nobody can adequately define what AGI even means. But hey, if we did, then we couldn't have so many meandering, unproductive conversations about it on Fridman or Rogan. Recently, I used GPT-4 to read my cousin's writings on his fictional world and generate a chart of the timeline and concepts using Mermaid syntax. I think one of the best things about LLMs at the moment is that it can convert things in an abstract way. Even if it doesn't get it totally right the first time, it can correct itself on instruction, and still saves time over coding something or downloading software.

tunesmith 4y ago
Were you able to paste his writings all in one prompt without exhausting token space? Or did you have to do something tricky to get around that?
- ravenstine 4y ago
  The summary of his world building wasn't super huge, so it was just large enough to fit the ~8000 token limit for the GPT-4 model on OpenAI after I trimmed it a bit. I, too, would like to know how to properly get around these technical limitations.
  - tunesmith 4y ago
    I know it is something along these lines: Install a vector database. Use the API to get vector embeddings for the manuscript, by getting them in chunks. Apparently this is possible the with API even though it's not with normal ChatGPT? Then, think of the query you want to ask. Use the API not to answer the query, but to get the vector embedding for the query. Then, do a search in the vector database to get the vectors that are "near" the vector for the query. Then, finally send the query's vector, and all the "near" vectors to the API. And then you'll get your answer. I don't know how to do any of that yet. So far it seems like milvus might be the easiest vector db to install locally. But vectors for text passages are very large, so I'm not sure why I'd expect the final query of multiple vectors to be small enough for the token limit. And I'm not really sure yet how to send a query to ChatGPT in vector form. (Ideally this could work against an open model that isn't ChatGPT.)

typon 4y ago

That's actually the hidden secret. Millions of people are using it to solve their problems quietly - something Google and other companies whose lunch they're eating should be worried about rather than being worried about grand ideas like "AI Safety". Pretty much all of my non-tech family is using it in ways that I never expected: making meal plans, writing school essays, making assignments for students, writing emails to CRA (Canadian Tax Agency), fixing broken english and translation, etc.

bolanyo 4y ago
I'm quite curious about some of these applications because I don't really understand what the cognitive work is that's being offloaded. Making meal plans: what is being done here? making a list of meals to eat each day of the week? isn't this just a question of thinking what one would like to eat? why is it easier to have the meals chosen by someone else? Writing school essays: what is the point of this? Aren't school essays only written in order to learn to write, or to learn about some other topic? Writing emails to CRA: presumably you have to put all the pertinent information in the prompt. Can't you just copy that prompt into an email? (The other couple do seen to make sense to me, fair enough.)
- darkwater 4y ago
  > Making meal plans: what is being done here? making a list of meals to eat each day of the week? isn't this just a question of thinking what one would like to eat? why is it easier to have the meals chosen by someone else? Do you have children? Meal planning can be a quite tedious and frustrating task if you want to cook at home, eat healthy, eat tasty, vary the dishes and have meals that kids will accept.
  - Bootvis 4y ago
    In addition, for recipes, it’s just a better Google. If I do “Give me a concise recipe for X” it gives me one. No fluff, no ads. Just ingredients and steps. For example. I asked for pasta carbonara, concise and then even more concise. Final result: Quick Carbonara (4 servings) Ingredients: 12 oz pasta 4 eggs 1 cup cheese 8 oz bacon 4 garlic Salt, pepper Parsley (opt) Cook pasta, save water. Mix eggs, cheese. Fry bacon, garlic. Combine, mix, season. Serve. Great if you’re grocery shopping and want to make sure you don’t forget anything.

calenti 4y ago

GPT generated clickbait for me! Ads at 11!

RoyGBivCap 4y ago

I pasted in a code snippet and asked if it was threadsafe (I already knew it wasn't) and it confirmed what I knew, and gave a simpler solution that worked for my particular use case that what I would have done. Anything to do with threads is hard™ and it did it. Though I think the big thing less tech savvy users have yet to realize is: GIGO still applies. If your prompts are bad, the output is bad. And if you didn't know your prompt was bad, it's almost impossible to know the output is too.