The little bunny!
ProfileGPT: An Example of AI Agents Collaboration Architecture
11–16 of 16 posts
Re: ProfileGPT: An Example of AI Agents Collaboration Architecture
#12> here is increasing concern about the safety of AI. Some countries, like Italy, have taken steps to address these concerns by banning the use of ChatGPT, while other like France and Spain are raising questions about it The ban was because ChatGPT isn't complying with the GDPR, nothing to do with ai safety, as their are no ai safety regulations, at least not yet.
Indeed it's GDPR issues that lead to the ban. I agree with you that the expression" safety of AI" may be too generic. The right to data portability (downloading your own data) was implemented only recently and that's what made this project possible. There are other concerns like the right to be forgotten which implies the ability to delete someone's data. For LLMs this would mean filtering the input data, retraining models which come with high costs.
Re: ProfileGPT: An Example of AI Agents Collaboration Architecture
#13> known for using advanced mathematical models and deep understanding of human behavior to predict future behavior of individuals and large populations. Isn’t psychohistory not able to predict individual behaviors and only effective on large numbers?
Yes, that's right but you can see this as a fiction inside the fiction. This agent "thinks" he is capable of predicting the future of individuals and acts as if he can.
Perhaps we are fiction, within fiction, creating a new fiction?
Re: ProfileGPT: An Example of AI Agents Collaboration Architecture
#14Tip: if blog post has "Conclusion" section at the end, it was very likely generated with GPT.
Re: ProfileGPT: An Example of AI Agents Collaboration Architecture
#15> it is possible to extract various types of information…
According to what definition of “extract information” are these ChatGPT agents extracting information?
Even granting a sense in which the above claim is true (perhaps limited to cases where the yielded text is actually true of the user), you make repeated claims that the agents can “predict,” “analyze,” and “infer.” There is more to prediction, analysis, and inference than the representation in natural language of those things. Just because ChatGPT produces sentences which have those linguistic forms (and even if those sentences are true) does not mean we can rightly climb ChatGPT can actually perform those activities. This is a problem with so much of the commentary around ChatGPT: it can produce the natural language outputs of some cognitive processes, but it does mean it performs those processes per se.
And that is fine! ChatGPT can plausibly be described as performing its own very special and powerful kind of generalized reasoning, I just think we should be careful attributing to it processes just because it can produce the corresponding linguistic forms.
Put another way, words like “inference” have a dual meaning: they refer to a type of utterance with a certain form and they refer to a specific process of reasoning. Production of the former does not prove the presence of the latter.
Along similar lines, why say your “Psychoanalyst” agent was “designed… [with] 20 years of experience in the field of profiling”? In what sense is that even remotely true? Better just to say “it was inspired by an expert with 20 years of experience, and we hope it performs with that level of competence.” But then why “20 years”? Would performance at the level of a 30-year professional be considered failure?
Re: ProfileGPT: An Example of AI Agents Collaboration Architecture
#16> it is possible to extract various types of information… According to what definition of “extract information” are these ChatGPT agents extracting information? Even granting a sense in which the above claim is true (perhaps limited to cases where the yielded text is actually true of the user), you make repeated claims that the agents can “predict,” “analyze,” and “infer.” There is more to prediction, analysis, and i…
> But then why “20 years”? Would performance at the level of a 30-year professional be considered failure?
They're keeping the "30 years of experience" agent for the 2.0 version /s.
This is a good analysis, and I'm glad that there are some people like you giving these things a good deal of thought.