Cool! I’m thinking something like a cobot model for AI writing tools. (is there a word for that when it’s not a robot?)
Could GPT-3 check if the references in an academic paper are real or made up? Or at least make a statement on the probability that they are real? Or is that kind of truth detection beyond it?
I got an idea for how to train an AI assistant to ask clarifying questions about your code.
For training data take some existing code. Extract all the comment and pass those through a model trained to invert the comment into a question. Then use the inverted comments as your completion for the input code when training the final model.
1 Like
The problem with this is that you need a source of ground truth, which is fundamentally a search problem. I’ve experimented with index search to help add facts to LLM systems but that is a complex ask.
1 Like
Sounds great! Helping to review papers can save many hours of work. I’m looking forward to seeing how far you can go using GPT-3 to approach the enormous amount of information that is available and humanly impossible to review as a whole. Another idea that occurs to me would be to use these abstracts to create explanatory diagrams or drawings of the molecular signaling pathways using DALLE-2, maybe we will have to wait until Dalle-3 for that… It would be cool to try that if you eventually have access to this tool.
1 Like
The video you made on getting GPT3 to critique arguments really got me thinking, and now I’m focused on a new direction for this: a program where the user inputs a short argument and then chooses whether to generate a list of main points, possible objections, strengths of the argument, description of its tone, questions about its meaning, and questions for further discussion and exploration. It seems like a prompt for each of those wouldn’t be hard to write–I played around with tone and discussion questions already, and your video shows how well it summarized main points in a list.
This would be incredibly helpful as a critical thinking tool for anyone reading something and wanting to engage with it: the general public, academics, students. It wouldn’t be creating an end product, so perhaps that would make it more likely to be used to stimulate critical thinking rather than substituting for it.
Since I’m not a software developer (though I’m starting to learn Python), I don’t know how to go about trying this out! Any suggestions?
Programming does take a while, so I’m not sure what advice to give. You can find folks to help on places like Fiverr and Upwork.
As far as formatting, it sounds like you’re imagining an automated Kialo. I recommend you look that platform up as part of constructing and organizing formal arguments.
1 Like
Thanks! I had forgotten about Kialo. Just took a look–that’s a great tool, but I wasn’t quite so debate-focused. I’m thinking of something that looks like an AI summarizer (i.e. Quillbot) but instead of just a summary option it has “list of points,” “discussion questions,” “possible implications,” “argument strengths” and “argument weaknesses” options as well.
1 Like
This might not be the kind of thing you are looking for, but i wonder if this is something you could address. How do you get gpt to error correct. Here’s my interaction with gpt-3.
Write a version of the following chorus for a professor's syllabus. Notice that it has an alternate rhyming scheme. This means the last word in the first line and the last word in the third line should rhyme. Likewise the last word in the fifth and seventh line should also rhyme. Make sure that your versions the same rhyme pattern and meter as the following example:
If you want to call me "baby"
Just go ahead now
And if you like to tell me "maybe"
Just go ahead now
And if you wanna buy me flowers
Just go ahead now
And if you would like to talk for hours
Just go ahead now.
-----
If you have questions for me
Just go ahead now
And if you'd like to talk more daily
Just go ahead now
And if you need more clarity
Just go ahead now
Before our time gets too cavalier
Just go ahead now.
List 10 words that rhyme with clarity.
A: charity, clarity, disparity, filigree, gallery, guarantee, longevity, malady, marauder
Redo the assignment and incorporate one of those words as the rhyme for the next to last line.
If you have questions for me
Just go ahead now
And if you'd like to talk more daily
Just go ahead now
And if you need more clarity
Just go ahead now
Before our time gets too cavalier
Just go ahead now.
You can see that I tried help it. But it did the same thing. I had the temp set at 1, and everything else at the playground defaults.
I think it has difficulty with word rules like that. See this other thread where we tried getting it to find a word without certain letters.
Hey everyone, thanks for sending me ideas and data. I’m a Renaissance Soul, which means I cycle through projects and phases and so now I’m cycling out of research, YouTube, and GPT-3 for a while. Instead, I’ll be working on my books for the next few months.
I have a trilogy I’m working on, which are written but in revisions. I also have a philosophical work that is also written, and is with beta readers. Then there’s the proverbs book I wrote with GPT-3 in the final stages of revision. Lastly, I have a book about artificial cognition that is in early stages (I’ll be incorporating much of what I’ve learned from this recent surge of research).
Many of you have provided fascinating insights and interesting problems, all of which have been invaluable fuel for thought and new ideas. My growing following on YouTube has been equally valuable. I hope to give back to the community in the form of my YouTube videos, GitHub repos, and books. Typically, I go through 3 or 4 cycles per year, so they average about 3 or 4 months. Accordingly, I should be back to research mode sometime in the autumn.
Thank you all once again. Looking forward to the next phase of the journey. Take care.
5 Likes
Thank you so much for high quality content. Best of luck in all your future endeavors and keep us updated with the status of your books.
Will you still be checking in on this forum or are you doing a hard checkout for now?
3 Likes
I have a problem engeniering the correct prompt for Codex to create a game of Hangman in the Javascript programming language.
Thanks for sharing this update–it seems like a great model for a Renaissance person to work in cycles. I so appreciate your ideas, teaching, and desire to collaborate. I learned a lot from the way you approached the argument analysis with GPT-3. Looking forward to learning more from your past videos while you’re away.
Best of luck with the fiction, philosophy, and proverbs. I hope your next projects are enjoyable.
Take care, Anna
1 Like
Dave-
We are following your advise on using embeddings (not finetuning) for our chatbot. So we have a 200 pages document. We want to use Babbage (2048 token max). Do you have any video (I could not find one on your YouTube) on how to divide the file into 100 documents each 2 pages long (<2048 tokens) and create 100 embeddings?
2 Likes
I do have to confess that recommendation was based on intuition and some experience with chatbots and cognitive architecture.
The best way to break up a document I’ve found is to use the wordwrap module in Python. This allows you to rapidly break any document into equally sized chunks, which means you can get the most mileage out of your embeddings.
You can see an example of word wrap being used in my “compress anything” video.
Hope that helps.
2 Likes
how to finetune codex?
You dont, it is not a feature
I want to generate YouTube video ideas with the help of Open AI and YouTube Data v3 to get genuine next video ideas. I’m using the flask framework. Please make a video tutorial on this.
I don’t know if the original message is still valid, but I asked something which can be the actual title of a tutorial.
I’ve seen that others have struggles with the fine tuned models.