Is there a possibility of a Meeseeks problem, where a GPT gets stumped by a query and then creates a GPT to solve the problem, in an infinite recursion? (https://en.wikipedia.org/wiki/Mr._Meeseeks)
It seems that is a theoretical possibility, which risks the type of recursive breakdown suggested by this paper: The Curse of Recursion: Training on Generated Data Makes Models Forget (https://arxiv.org/abs/2305.17493)
Excellent blog post as ever. You are a cutting edge mind in this field and I am grateful!
On GPTs, it seems a limiting factor to me that the GPT agents one creates can only be accessed by people with paid OpenAI accounts. Is there any way around this currently?
Thanks Allison. I'm struggling with it TBH, the GPT keeps encountering errors whenever it tries to create a doc. Will persist, and thanks for replying!!
Dear Ethan - this is the first time I have read your work and I very much like what you present, with clear examples …
I also agree that students may already be “cheating” through their use of AI! My doctoral research was in Question-Answering, completed a lifetime ago, which uses a form of promptiing, in a much simpler way, for elementary students.
My questions are: 1. Would your feedback be more well-received if the edits were in green (sending a “go forward & improve) message??
2. Is there a way to develop something like “Turn-it-in”, that monitors the use of AI CHAT (in all its forms) so students might begin to actually acknowledge their use of this technology?
NOT sure if I’m signed in or not? Best wishes for your book & I will read more of your work! Gail Brown gailbrown@designedlearning.com.au
Ps I’d be very interested in your answers - but I’m a bit if a novice on these chats? Maybe you could email if you have time??? 👍
Ethan, thanks for a nice, informative, and thought-provoking article. Loved your idea of creating use case-specific GPTs. But, my question is: For most such real-life custom GPTs to be of real use, it’d need to be exposed to either some proprietary (e.g., of a given company) or personal (e.g., one's credit card and/or bank details) data. If so, how secure and safe is such exposure to public LLMs — either from a specific company's POV or a person's safety and privacy POV? Thanks, again, for a great article.
PS: I'm writing this without knowing if OpenAI has such provisions in its T&C's. So, excuse me if it's public knowledge.
Really looking forward to this and keep up the awesome work Ethan. I recently wrote a book along similar lines if it’d help you in any way I’d be happy to chat through the customer interviews, the bumps and ups & downs I hit along the way
I would respectfully like to put in the suggestion box a request for a post that reflects these basic questions
1. What can "AI" do NOW? Not an extrapolation or prediction, which have proven to be wildly overly optimistic in this domain for almost 70 years, but actually happening, now. Not claims about probability or plausibility but factual statements about evidence-based current reality.
2. What can these LLMs and "neural nets" and similar technologies produce that human beings simply cannot produce? The strings of text and images that you've shown could easily be produced by human beings. I'm not talking about speed or automaticity. I'm talking about actually generating something that human beings simply cannot produce in any other way. This would be a requirement for calling any change revolutionary.
3. What can these LLMs and "neural nets" do in the world of atoms, away from a computer screen? How, in any remotely plausible nearish future, will these technologies transform the lived environment in the way that fossil fuels, electrification, and indoor plumbing transformed the lived environment?
Hi, what a great blog-post - thank you so much for your work, Ethan. I'm interested in learning more about the "create your own adventure" GPT you built: could you please share the PDF you used building it?
Hi Ethan, great Blog post with very useful insights: thanks for your work. I am really interessant in the „create your own adventure“ GPT - could you please share the PDF you used to build it?
Has anyone else found that the GPTs are throwing up a lot of technical error messages, such as "You've reached the current usage cap for GPT-4, please try again after 1:23 PM" or "message not found in conversation" and others? Might be growing pains, but this severely impacts usability.
Is there a possibility of a Meeseeks problem, where a GPT gets stumped by a query and then creates a GPT to solve the problem, in an infinite recursion? (https://en.wikipedia.org/wiki/Mr._Meeseeks)
unlike meeseeks if each iteration has variation then it becomes akin to evolution and the problem will be solved by natural selection
It seems that is a theoretical possibility, which risks the type of recursive breakdown suggested by this paper: The Curse of Recursion: Training on Generated Data Makes Models Forget (https://arxiv.org/abs/2305.17493)
Ordered the book- can’t wait to read it. The way you lay the path for anyone to understand & use this technology is amazing.
Excellent blog post as ever. You are a cutting edge mind in this field and I am grateful!
On GPTs, it seems a limiting factor to me that the GPT agents one creates can only be accessed by people with paid OpenAI accounts. Is there any way around this currently?
Congrats on the book!
I am intrigued though... Generative AI is moving so fast, and your book comes out in 6 months. How did you (near-)future-proof it?
Ethan - how do you integrate with Word please?
Not Ethan but looks like this just uses built-in Code Interpreter to handle that.
Thanks Allison. I'm struggling with it TBH, the GPT keeps encountering errors whenever it tries to create a doc. Will persist, and thanks for replying!!
I am having this same problem
Dear Ethan - this is the first time I have read your work and I very much like what you present, with clear examples …
I also agree that students may already be “cheating” through their use of AI! My doctoral research was in Question-Answering, completed a lifetime ago, which uses a form of promptiing, in a much simpler way, for elementary students.
My questions are: 1. Would your feedback be more well-received if the edits were in green (sending a “go forward & improve) message??
2. Is there a way to develop something like “Turn-it-in”, that monitors the use of AI CHAT (in all its forms) so students might begin to actually acknowledge their use of this technology?
NOT sure if I’m signed in or not? Best wishes for your book & I will read more of your work! Gail Brown gailbrown@designedlearning.com.au
Ps I’d be very interested in your answers - but I’m a bit if a novice on these chats? Maybe you could email if you have time??? 👍
What an absolutely brilliant blog! I learnt so much and I love how you write!
I pre-ordered the book (great cover) - Very much enjoy these columns. Thank you.
👏
Ethan, thanks for a nice, informative, and thought-provoking article. Loved your idea of creating use case-specific GPTs. But, my question is: For most such real-life custom GPTs to be of real use, it’d need to be exposed to either some proprietary (e.g., of a given company) or personal (e.g., one's credit card and/or bank details) data. If so, how secure and safe is such exposure to public LLMs — either from a specific company's POV or a person's safety and privacy POV? Thanks, again, for a great article.
PS: I'm writing this without knowing if OpenAI has such provisions in its T&C's. So, excuse me if it's public knowledge.
Really looking forward to this and keep up the awesome work Ethan. I recently wrote a book along similar lines if it’d help you in any way I’d be happy to chat through the customer interviews, the bumps and ups & downs I hit along the way
I would respectfully like to put in the suggestion box a request for a post that reflects these basic questions
1. What can "AI" do NOW? Not an extrapolation or prediction, which have proven to be wildly overly optimistic in this domain for almost 70 years, but actually happening, now. Not claims about probability or plausibility but factual statements about evidence-based current reality.
2. What can these LLMs and "neural nets" and similar technologies produce that human beings simply cannot produce? The strings of text and images that you've shown could easily be produced by human beings. I'm not talking about speed or automaticity. I'm talking about actually generating something that human beings simply cannot produce in any other way. This would be a requirement for calling any change revolutionary.
3. What can these LLMs and "neural nets" do in the world of atoms, away from a computer screen? How, in any remotely plausible nearish future, will these technologies transform the lived environment in the way that fossil fuels, electrification, and indoor plumbing transformed the lived environment?
Hi, what a great blog-post - thank you so much for your work, Ethan. I'm interested in learning more about the "create your own adventure" GPT you built: could you please share the PDF you used building it?
Hi Ethan, great Blog post with very useful insights: thanks for your work. I am really interessant in the „create your own adventure“ GPT - could you please share the PDF you used to build it?
I have to agree with Bing and Bard, I love the cover. Not sure if it was intentional, but I like how the fruit looks half ripe.
Has anyone else found that the GPTs are throwing up a lot of technical error messages, such as "You've reached the current usage cap for GPT-4, please try again after 1:23 PM" or "message not found in conversation" and others? Might be growing pains, but this severely impacts usability.