6 ms·

There is a new version of GPT-4 Turbo now live in ChatGPT

30 points by Topfi 3y ago

Topfi 3y ago
Sam Altman just posted the following on X: >>there is a new version of GPT-4 Turbo now live in ChatGPT; you should hopefully find it a lot better! please let us know what you think.<< ChatGPT in this case is most likely referring to the implementation accessible via https://chat.openai.com/ https://chat.openai.com/, once again raising questions for myself whether and how this change may impact the model accessible via their API gpt-4-1106-preview. Does "new version" mean what has been accessible via the chat website was less capable than gpt-4-1106-preview and if so, in what manner, or is gpt-4-1106-preview going to receive certain improvements that are currently being tested via the chat website? I'd love to see more clarity on that front.
- samrolken 3y ago
  I'm guessing the reason the model is listed as "preview" is so that they have this leeway to tweak it and optimize it as needed, for example in terms of the balance between performance/cost and capability which seems to be the case here.
retskrad 3y ago
At $10/month, ChatGPT-4 is reasonable. At $20, that's a no no!
- gmuslera 3y ago
  The value it haves for you may exceed or not its price. But "you" is an important part of that equation.
  - gumballindie 3y ago
    What value does a procedural text generator provide?
    - TeMPOraL 3y ago
      Depends on where it sits on a spectrum between, say, a dumb Markov chain, and a human like you and me. GPT-4 is definitely closer to the latter than to the former.
      - throw_pm23 3y ago
        Yes, but what do you use it for?
      - gumballindie 3y ago
        > GPT-4 is definitely closer to the latter than to the former. That’s a seriously dubious statement.
        speedgoose 3y ago
        If you think about how people use and perceive the technology, it’s not that dubious.
        MattRix 3y ago
        Have you used it? A GPT4 chatbot could convince 90% of people it was a human, right now. It could pass the majority of knowledge-based tests in all kinds of fields. 100 million people find it useful enough to use weekly.
        martindbp 3y ago
        No level of capability is enough to convince some people it's not just a stochastic parrot. You're completely right on the Turing test. At this point it's just a matter of how long it could fool us. You have to remember it's not specifically trained to fool us either, it's trained through RLHF to be verbose and "helpful". If you train it right and add in some RAG (vector databases) I think you could fool most people for a pretty long time, some gullible people forever.
        gumballindie 3y ago
        It could convince 90% of the people that are ready and willing to think it was human. Even when it does, it's only a piece of software designed specifically to emulate, therefore it's fake. Regardless of how much people want to claim it is intelligent, it's not. Sure it has load of data at its dispoal to generate well mimicked output, but to some of us it's so obvious and buggy that it's a waste of time.
- hhh 3y ago
  why?
- terhechte 3y ago
  I use it usually 10-20 times per day. 5-10 when I'm not working and like 20-30 when I'm working. It saves me a huge amount of time. It would already make sense at a wage of $20 but I earn more and so it is so much worth it. Even if I earn no wage (say a indie dev working on a game) it would still be useful because I save time which I can invest in other things (or play video games). I'd honestly pay $50-$100 for it as it saves me so much time and gives me so much value.
  - gumballindie 3y ago
    Why are reddit and forums full of people asking how to make chatgpt code work, and are always downvoted?
    - danielvaughn 3y ago
      I think it's because AI is a crutch for junior engineers, and a rocket for senior engineers. It takes some degree of wisdom to know how to use it effectively, and the idea that you're going to lean entirely on it to make your program work is...misguided at best.
  - iot_devs 3y ago
    What do you use it for?
    - martindbp 3y ago
      I went for a walk and had a conversation about Rademacher complexity and VC-dimension of a classifier, then I asked it to quiz my understanding based on what we talked about. It's almost at the level on "Her" but you have a very knowledgeable tutor to ask who won't judge and is always available.
      - apwell23 3y ago
        Thats a pretty neat use of chatgpt. My sister is prepping for USMLE. Wonder if she can do something like this too.
- zarzavat 3y ago
  You have to consider costs. If it costs them $15/month to provide the service (just a made up number) then $10/month would not be reasonable. The Plus users also subsidize the Free users to a certain extent. The other effect at work here is anchoring. OpenAI wants people to be prepared to pay more for AI. Setting a higher price anchors that price in people’s minds and makes it feel normal. This will serve them better in the future since the long-term objective is to provide services that replace employees, who are surely paid more than $20/month.
  - somewhereoutth 3y ago
    But therein lies the crucial question, does the actual business value of the service exceed the costs of running the service + reasonable profit margin? If not, then no amount of anchoring will help.
  - raincole 3y ago
    Nah, you have to consider how much time it saves for you. If your rate is $30/hr and ChatGPT saves you 1 hours per month more than free alternatives then why not? Cost is their business, not yours (unless you're a competitor to them).
- TeMPOraL 3y ago
  Even now that it has image generation via DALL-E 3, image comprehension via GPT-4V, and the Android app can hold a voice conversation with you, with unprecedented voice recognition accuracy?
  - apwell23 3y ago
    what would a ordinary person use all these for beyond curiosities? I pay for but i only use it handful of times after i remind myself to use it .
    - chpatrick 3y ago
      When you have a question, there's no answer on Google and you don't know who else to ask.
      - apwell23 3y ago
        what an example of a question only chatgpt knows ?
        chpatrick 3y ago
        Obscure command-line invocations, some specialist topic that's not discussed much online. Usually I Google something first and if I can't see any straightforward answer in a couple of minutes I'll ask ChatGPT. For example I don't know much about electronics so I asked it how design a circuit for a problem I had and it came up with a useful answer.
        mmasu 3y ago
        for me it’s really worth it, and use it as a starting point rather than spending time sifting through garbage google results. I am using it to study math, to ask about css frameworks, about api best practices, SQL dialects, and so many more things. Then i can drill in and go to source documents if I want to. It took me a while to see the value, however now i feel i have a “generalist” assistant and use it as much as i can.
        3y ago
- unsupp0rted 3y ago
  For me it’s a steal at 5x the price. Maybe 10x.
  - op00to 3y ago
    Not until they stop training on my conversations.
    - blharr 3y ago
      According to them, "We do not train on your data from ChatGPT Enterprise or our API Platform" https://openai.com/enterprise-privacy https://openai.com/enterprise-privacy Though I'm not sure if this covers other analysis of the data, or if it's something only for Enterprise API users.
      - op00to 3y ago
        I use the API now, but I have to maintain my own "Chat" instance. I'd love to be able to just pay $20, but also keep my history. Right now I'm spending less than $10 a month on GPT-4 API costs, so maybe I don't need to upgrade.
- thejackgoode 3y ago
  I cannot get used to American custom of advertising gross and charging net prices. It's 24.20 and would be nice to discuss it as such
smallnix 3y ago
How can I check which version is serving me?
- samstave 3y ago
  [dead]
- murkt 3y ago
  Just ask it!
fassssst 3y ago
The new model combined with the GPTs feature is incredible. I made 10 GPTs yesterday. My favorite is Meme Machine, where you just specify a concept like "return to office, safari animals" and it generates an image with DALL-E and then composites text over it using Python: https://www.reddit.com/r/ChatGPT/comments/17tb7rm/mememachine_goat_gpt/?utm_source=share&utm_medium=web2x&context=3 https://www.reddit.com/r/ChatGPT/comments/17tb7rm/mememachin... I noticed all my GPTs started working better yesterday, must have been the new model rollout.
- fassssst 3y ago
  I just added one paragraph to my Meme Machine GPT, and now it can generate animated gifs with DALL-E. Try it here: https://chat.openai.com/g/g-MP5Cx7F7W-meme-machine https://chat.openai.com/g/g-MP5Cx7F7W-meme-machine
samskot 3y ago
Does anyone know if it has better speed or overall knowledge and reasoning capabilities?
jerpint 3y ago
Not sure if it’s related or not but I noticed the UX has changed on the web interface and when I ask for code, it now defaults to running it through an interpreter with a hidden view by default. This is very annoying to me as I want to mostly read the code, not have it be executed without any context
- logicchains 3y ago
  On the bright side, in some cases at least it includes a test case, runs it, and if it fails then reads the error and tries to fix the code, all in a single run.
- alfonsodev 3y ago
  are you writing Python code? I wonder if this behavior just applies to Python as I recall code interpreter is Python only.
  - jerpint 3y ago
    Yes mostly python
- DonHopkins 3y ago
  Is it smart enough to solve the halting problem and not bother running code with infinite loops?
  - appplication 3y ago
    Infinite loops and the halting problem are separate issues. You can detect infinite loops with basic static analysis.
    - jhardy54 3y ago
      You can detect some infinite loops with basic static analysis, but detecting all infinite loops would mean that you’ve solved the halting problem.

30 points by Topfi 3y ago

Sam Altman just posted the following on X: >>there is a new version of GPT-4 Turbo now live in ChatGPT; you should hopefully find it a lot better! please let us know what you think.<< ChatGPT in this case is most likely referring to the implementation accessible via https://chat.openai.com/ https://chat.openai.com/, once again raising questions for myself whether and how this change may impact the model accessible via their API gpt-4-1106-preview. Does "new version" mean what has been accessible via the chat website was less capable than gpt-4-1106-preview and if so, in what manner, or is gpt-4-1106-preview going to receive certain improvements that are currently being tested via the chat website? I'd love to see more clarity on that front.

samrolken 3y ago
I'm guessing the reason the model is listed as "preview" is so that they have this leeway to tweak it and optimize it as needed, for example in terms of the balance between performance/cost and capability which seems to be the case here.

retskrad 3y ago

At $10/month, ChatGPT-4 is reasonable. At $20, that's a no no!

gmuslera 3y ago
The value it haves for you may exceed or not its price. But "you" is an important part of that equation.
- gumballindie 3y ago
  What value does a procedural text generator provide?
  - TeMPOraL 3y ago
    Depends on where it sits on a spectrum between, say, a dumb Markov chain, and a human like you and me. GPT-4 is definitely closer to the latter than to the former.
    - throw_pm23 3y ago
      Yes, but what do you use it for?
    - gumballindie 3y ago
      > GPT-4 is definitely closer to the latter than to the former. That’s a seriously dubious statement.
      - speedgoose 3y ago
        If you think about how people use and perceive the technology, it’s not that dubious.
      - MattRix 3y ago
        Have you used it? A GPT4 chatbot could convince 90% of people it was a human, right now. It could pass the majority of knowledge-based tests in all kinds of fields. 100 million people find it useful enough to use weekly.
        martindbp 3y ago
        No level of capability is enough to convince some people it's not just a stochastic parrot. You're completely right on the Turing test. At this point it's just a matter of how long it could fool us. You have to remember it's not specifically trained to fool us either, it's trained through RLHF to be verbose and "helpful". If you train it right and add in some RAG (vector databases) I think you could fool most people for a pretty long time, some gullible people forever.
        gumballindie 3y ago
        It could convince 90% of the people that are ready and willing to think it was human. Even when it does, it's only a piece of software designed specifically to emulate, therefore it's fake. Regardless of how much people want to claim it is intelligent, it's not. Sure it has load of data at its dispoal to generate well mimicked output, but to some of us it's so obvious and buggy that it's a waste of time.
hhh 3y ago
why?
terhechte 3y ago
I use it usually 10-20 times per day. 5-10 when I'm not working and like 20-30 when I'm working. It saves me a huge amount of time. It would already make sense at a wage of $20 but I earn more and so it is so much worth it. Even if I earn no wage (say a indie dev working on a game) it would still be useful because I save time which I can invest in other things (or play video games). I'd honestly pay $50-$100 for it as it saves me so much time and gives me so much value.
- gumballindie 3y ago
  Why are reddit and forums full of people asking how to make chatgpt code work, and are always downvoted?
  - danielvaughn 3y ago
    I think it's because AI is a crutch for junior engineers, and a rocket for senior engineers. It takes some degree of wisdom to know how to use it effectively, and the idea that you're going to lean entirely on it to make your program work is...misguided at best.
- iot_devs 3y ago
  What do you use it for?
  - martindbp 3y ago
    I went for a walk and had a conversation about Rademacher complexity and VC-dimension of a classifier, then I asked it to quiz my understanding based on what we talked about. It's almost at the level on "Her" but you have a very knowledgeable tutor to ask who won't judge and is always available.
    - apwell23 3y ago
      Thats a pretty neat use of chatgpt. My sister is prepping for USMLE. Wonder if she can do something like this too.
zarzavat 3y ago
You have to consider costs. If it costs them $15/month to provide the service (just a made up number) then $10/month would not be reasonable. The Plus users also subsidize the Free users to a certain extent. The other effect at work here is anchoring. OpenAI wants people to be prepared to pay more for AI. Setting a higher price anchors that price in people’s minds and makes it feel normal. This will serve them better in the future since the long-term objective is to provide services that replace employees, who are surely paid more than $20/month.
- somewhereoutth 3y ago
  But therein lies the crucial question, does the actual business value of the service exceed the costs of running the service + reasonable profit margin? If not, then no amount of anchoring will help.
- raincole 3y ago
  Nah, you have to consider how much time it saves for you. If your rate is $30/hr and ChatGPT saves you 1 hours per month more than free alternatives then why not? Cost is their business, not yours (unless you're a competitor to them).
TeMPOraL 3y ago
Even now that it has image generation via DALL-E 3, image comprehension via GPT-4V, and the Android app can hold a voice conversation with you, with unprecedented voice recognition accuracy?
- apwell23 3y ago
  what would a ordinary person use all these for beyond curiosities? I pay for but i only use it handful of times after i remind myself to use it .
  - chpatrick 3y ago
    When you have a question, there's no answer on Google and you don't know who else to ask.
    - apwell23 3y ago
      what an example of a question only chatgpt knows ?
      - chpatrick 3y ago
        Obscure command-line invocations, some specialist topic that's not discussed much online. Usually I Google something first and if I can't see any straightforward answer in a couple of minutes I'll ask ChatGPT. For example I don't know much about electronics so I asked it how design a circuit for a problem I had and it came up with a useful answer.
      - mmasu 3y ago
        for me it’s really worth it, and use it as a starting point rather than spending time sifting through garbage google results. I am using it to study math, to ask about css frameworks, about api best practices, SQL dialects, and so many more things. Then i can drill in and go to source documents if I want to. It took me a while to see the value, however now i feel i have a “generalist” assistant and use it as much as i can.
      - 3y ago
unsupp0rted 3y ago
For me it’s a steal at 5x the price. Maybe 10x.
- op00to 3y ago
  Not until they stop training on my conversations.
  - blharr 3y ago
    According to them, "We do not train on your data from ChatGPT Enterprise or our API Platform" https://openai.com/enterprise-privacy https://openai.com/enterprise-privacy Though I'm not sure if this covers other analysis of the data, or if it's something only for Enterprise API users.
    - op00to 3y ago
      I use the API now, but I have to maintain my own "Chat" instance. I'd love to be able to just pay $20, but also keep my history. Right now I'm spending less than $10 a month on GPT-4 API costs, so maybe I don't need to upgrade.
thejackgoode 3y ago
I cannot get used to American custom of advertising gross and charging net prices. It's 24.20 and would be nice to discuss it as such

smallnix 3y ago

How can I check which version is serving me?

samstave 3y ago
[dead]
murkt 3y ago
Just ask it!

fassssst 3y ago

The new model combined with the GPTs feature is incredible. I made 10 GPTs yesterday. My favorite is Meme Machine, where you just specify a concept like "return to office, safari animals" and it generates an image with DALL-E and then composites text over it using Python: https://www.reddit.com/r/ChatGPT/comments/17tb7rm/mememachine_goat_gpt/?utm_source=share&utm_medium=web2x&context=3 https://www.reddit.com/r/ChatGPT/comments/17tb7rm/mememachin... I noticed all my GPTs started working better yesterday, must have been the new model rollout.

fassssst 3y ago
I just added one paragraph to my Meme Machine GPT, and now it can generate animated gifs with DALL-E. Try it here: https://chat.openai.com/g/g-MP5Cx7F7W-meme-machine https://chat.openai.com/g/g-MP5Cx7F7W-meme-machine

samskot 3y ago

Does anyone know if it has better speed or overall knowledge and reasoning capabilities?

jerpint 3y ago

Not sure if it’s related or not but I noticed the UX has changed on the web interface and when I ask for code, it now defaults to running it through an interpreter with a hidden view by default. This is very annoying to me as I want to mostly read the code, not have it be executed without any context

logicchains 3y ago
On the bright side, in some cases at least it includes a test case, runs it, and if it fails then reads the error and tries to fix the code, all in a single run.
alfonsodev 3y ago
are you writing Python code? I wonder if this behavior just applies to Python as I recall code interpreter is Python only.
- jerpint 3y ago
  Yes mostly python
DonHopkins 3y ago
Is it smart enough to solve the halting problem and not bother running code with infinite loops?
- appplication 3y ago
  Infinite loops and the halting problem are separate issues. You can detect infinite loops with basic static analysis.
  - jhardy54 3y ago
    You can detect some infinite loops with basic static analysis, but detecting all infinite loops would mean that you’ve solved the halting problem.