I do admit that there is low probability that it is wrong
By your own admission it can get things wrong, yet you’re arguing it should be trusted at face value.
but simply dismissing it as no evidence at all is intentional dishonesty.
The whole point of citing a source is so that you can confirm the veracity of the how the source came to its conclusion.You have no idea why the LLM gave you the answer it did. You don’t know how credible its input data was. Hopefully those involved in these discussions on both sides are searching for truth. The critical examination of the data and the origin of that data is the bedrock of that. Simply pasting raw LLM output doesn’t allow any of that to occur.
LLM/AI ML can have a place in these discussions as a tool you use for yourself, and then you can search for supporting sources to back up the LLM’s claim. However, that’s work you have to do. Its not my job when you’re the one trying to convince me of your LLM’s conclusion.
Dishonesty is passing off raw LLM output as researched fact. Its also lazy.
I am arguing that it should be given relatively high credence, not “trusted at face value”. Same as with Wikipedia, by the way. As an indication that likely things are true. On Internet forums it is much higher credence than most of the people supply. I am not writing scientific paper here, I am discussing topic with you. Would you rather me stating acts without any sources at all?
For this discussion if you have different opinion, with better argumentation and sources please do so, and I will change my view. This is what discussion on discussion board suppose to be.
And you can absolutely confirm the veracity (or not) of ChatGPT4 itself. You can ask the question yourself. You can collect statistics how likely it gives correct answers to similar questions, or find already published data about this topic. Based on that you can calculate probability that the statement is true. And it is much higher than 50%.
In short, don’t attack the messenger, attack the message.
I suggest asking it about highly specialized technical topics or very specific details. AI either tends to get it wrong, or it’ll tell you it isn’t qualified to give an answer.
I actually do do that. And quite often it does right. It very rare get’s it wrong. More often when it is “wrong” it will give you generic useless answer, not what you are asking, if it does not have information. But just opposite to truth? Not often. You can ask questions of a type “Describe main difference of Cynicism and Stoicism philosophy” or “Explain similarities and differences of EDFA and YDFA” And it give very reasonable answers.
This particular topic, however, is not highly specialized, or at least not specialized more than the questions I supplied above as examples. So I expect similar validity of the answer.
I am arguing that it should be given relatively high credence, not “trusted at face value”. Same as with Wikipedia, by the way.
I know you are, and I disagree. Your example of Wikipedia is a great differentiator.
The reason that Wikipedia is generally a good source is that it too cites its sources. If a Wikipedia entry makes a claim, I can see where that data came from or if its not cited, I know the claim is suspect and not to trust it. ChatGPT has none of that.
I am discussing topic with you. Would you rather me stating acts without any sources at all?
From my perspective not citing any source is exactly what you’re doing. ChatGPT isn’t a trusted or challenge-able source
And you can absolutely confirm the veracity (or not) of ChatGPT4 itself. You can ask the question yourself. You can collect statistics how likely it gives correct answers to similar questions, or find already published data about this topic.
If you want ChatGPT involved, that’s your job. Why is it you can’t use ChatGPT to find the real source which backs its claim?
Based on that you can calculate probability that the statement is true. And it is much higher than 50%.
"much higher that 50% is way way too low a bar to be considered a factual source.
In short, don’t attack the messenger, attack the message.
I can’t attack the message, its not backed by any sources to question it. My only option is to trust it absolutely, which is absurd.
Let me ask you a question differently. Do you think that the fact that ChatGPT stated what it stated has ANY impact on the credence of those statements?
Like what is more likely to be true, if I, none-specialist come up to those statements myself without doing any research OR come up to those statements BECAUSE ChatGPT stated that?
By your own admission it can get things wrong, yet you’re arguing it should be trusted at face value.
The whole point of citing a source is so that you can confirm the veracity of the how the source came to its conclusion.You have no idea why the LLM gave you the answer it did. You don’t know how credible its input data was. Hopefully those involved in these discussions on both sides are searching for truth. The critical examination of the data and the origin of that data is the bedrock of that. Simply pasting raw LLM output doesn’t allow any of that to occur.
LLM/AI ML can have a place in these discussions as a tool you use for yourself, and then you can search for supporting sources to back up the LLM’s claim. However, that’s work you have to do. Its not my job when you’re the one trying to convince me of your LLM’s conclusion.
Dishonesty is passing off raw LLM output as researched fact. Its also lazy.
I am arguing that it should be given relatively high credence, not “trusted at face value”. Same as with Wikipedia, by the way. As an indication that likely things are true. On Internet forums it is much higher credence than most of the people supply. I am not writing scientific paper here, I am discussing topic with you. Would you rather me stating acts without any sources at all?
For this discussion if you have different opinion, with better argumentation and sources please do so, and I will change my view. This is what discussion on discussion board suppose to be.
And you can absolutely confirm the veracity (or not) of ChatGPT4 itself. You can ask the question yourself. You can collect statistics how likely it gives correct answers to similar questions, or find already published data about this topic. Based on that you can calculate probability that the statement is true. And it is much higher than 50%.
In short, don’t attack the messenger, attack the message.
I suggest asking it about highly specialized technical topics or very specific details. AI either tends to get it wrong, or it’ll tell you it isn’t qualified to give an answer.
I actually do do that. And quite often it does right. It very rare get’s it wrong. More often when it is “wrong” it will give you generic useless answer, not what you are asking, if it does not have information. But just opposite to truth? Not often. You can ask questions of a type “Describe main difference of Cynicism and Stoicism philosophy” or “Explain similarities and differences of EDFA and YDFA” And it give very reasonable answers.
This particular topic, however, is not highly specialized, or at least not specialized more than the questions I supplied above as examples. So I expect similar validity of the answer.
I know you are, and I disagree. Your example of Wikipedia is a great differentiator.
The reason that Wikipedia is generally a good source is that it too cites its sources. If a Wikipedia entry makes a claim, I can see where that data came from or if its not cited, I know the claim is suspect and not to trust it. ChatGPT has none of that.
From my perspective not citing any source is exactly what you’re doing. ChatGPT isn’t a trusted or challenge-able source
If you want ChatGPT involved, that’s your job. Why is it you can’t use ChatGPT to find the real source which backs its claim?
"much higher that 50% is way way too low a bar to be considered a factual source.
I can’t attack the message, its not backed by any sources to question it. My only option is to trust it absolutely, which is absurd.
Let me ask you a question differently. Do you think that the fact that ChatGPT stated what it stated has ANY impact on the credence of those statements?
Like what is more likely to be true, if I, none-specialist come up to those statements myself without doing any research OR come up to those statements BECAUSE ChatGPT stated that?