3 ms·

I don't understand why they freak out. Every single time I have used ChatGPT (which isn't many), it gave me an absolutely unusable answer completely off the ma

I don't understand why they freak out. Every single time I have used ChatGPT (which isn't many), it gave me an absolutely unusable answer completely off the mark. It's funny to see FAANG CEOs have FOMO over what is essentially smoke and mirrors.

wiihack 4y ago
Could you provide an example of what you're asking for? It appears to work well for me. I would estimate that 80% of the answers are very good, and 10% provide at least a glimpse of the information I was seeking.
- gmiller123456 4y ago
  One of the first things I asked it to do was to "write a program to compute the rise and set times of a star". It got a couple of pieces right, but was way off the mark. I asked it a few other astronomy related questions, and it would try to use a library like Astropy, but make API calls that don't exist. I tried things like "don't use Astropy", and it would just switch to a different library. It seems to do OK on things like the instructions for Fizz-Buzz.
dpkirchner 4y ago
I would guess they are less concerned about where ChatGPT is today than where it will be in 5 years.
- iLoveOncall 4y ago
  Even more convincing when it is wrong?
  - jerjerjer 4y ago
    ChatGPT lacks the ability to prove that output is factually correct (which is an interesting but distinct problem). However, as long as the answer is derivative of supplied information, ChatGPT shines. Even today it can already: - Write coherent short stories based on a prompt - Be a personal diary/assistant - Summarize and extract information from text. There's a lot of applications for that.
  - dpkirchner 4y ago
    It won't be wrong forever, there's a good chance they'll figure out how to evaluate sources and provide more useful results. And besides, the bar is low -- Google search results for some topics (e.g., programming related) are quite poor, filled with garbage and clone sites.
    - iLoveOncall 4y ago
      I think people saying that just don't understand what GPT is. It's not a database of knowledge that sometimes invents stuff. It's a statistical model that ALWAYS invents stuff, which sometimes happens to match reality. It will never be able to evaluate sources, because it doesn't use sources.
    - bagacrap 4y ago
      isn't "evaluating sources" the exact problem of web search that Google has been throwing billions of dollars at for many years, with an equivalently dedicated and motivated horde of SEO specialists on the other side? How is chatgpt going to easily solve that overnight?
amelius 4y ago
It's because the tech is moving very fast.

wiihack 4y ago

Could you provide an example of what you're asking for? It appears to work well for me. I would estimate that 80% of the answers are very good, and 10% provide at least a glimpse of the information I was seeking.

gmiller123456 4y ago
One of the first things I asked it to do was to "write a program to compute the rise and set times of a star". It got a couple of pieces right, but was way off the mark. I asked it a few other astronomy related questions, and it would try to use a library like Astropy, but make API calls that don't exist. I tried things like "don't use Astropy", and it would just switch to a different library. It seems to do OK on things like the instructions for Fizz-Buzz.

dpkirchner 4y ago

I would guess they are less concerned about where ChatGPT is today than where it will be in 5 years.

iLoveOncall 4y ago
Even more convincing when it is wrong?
- jerjerjer 4y ago
  ChatGPT lacks the ability to prove that output is factually correct (which is an interesting but distinct problem). However, as long as the answer is derivative of supplied information, ChatGPT shines. Even today it can already: - Write coherent short stories based on a prompt - Be a personal diary/assistant - Summarize and extract information from text. There's a lot of applications for that.
- dpkirchner 4y ago
  It won't be wrong forever, there's a good chance they'll figure out how to evaluate sources and provide more useful results. And besides, the bar is low -- Google search results for some topics (e.g., programming related) are quite poor, filled with garbage and clone sites.
  - iLoveOncall 4y ago
    I think people saying that just don't understand what GPT is. It's not a database of knowledge that sometimes invents stuff. It's a statistical model that ALWAYS invents stuff, which sometimes happens to match reality. It will never be able to evaluate sources, because it doesn't use sources.
  - bagacrap 4y ago
    isn't "evaluating sources" the exact problem of web search that Google has been throwing billions of dollars at for many years, with an equivalently dedicated and motivated horde of SEO specialists on the other side? How is chatgpt going to easily solve that overnight?

amelius 4y ago

It's because the tech is moving very fast.