OpenAI Dev-Day 2023: Breakout Sessions!

Hey y’all.

OpenAI just uploaded a series of videos from the dev-day breakout sessions, and they are stacked with insights. Here’s what you can dive into:

New Products: A Deep Dive

Join Krithika, Thomas, and Nick for an in-depth exploration of AI agents and APIs, featuring live demos of ChatGPT and GPTs.

A Survey of Techniques for Maximizing LLM Performance

John Allard and Colin take you through the latest advancements in Large Language Models and fine-tuning techniques.

The New Stack and Ops for AI

Sherwin and Shyamal discuss the journey of AI applications from concept to production-ready solutions.

Research x Product

Barret and Joanne explore the intersection of AI research and product development, focusing on dialogue interfaces from GPT-3.5 to GPT-4.

The Business of AI

A panel discussion with Aliisa Rosenthal, Kathy Baxter, Oji Udezue, and Miqdad Jaffer on incorporating AI into business models.

Can’t wait to hear your thoughts on these sessions!

4 Likes

Dive into the new products announced during the keynote (GPTs in ChatGPT and the Assistants API) with end-to-end demos and best practices.

Speakers:
Michelle Pokrass
Member of Technical Staff at OpenAI

Olivier Godement
Product at OpenAI

Nick Turley
Product at OpenAI

Thomas Dimson
Member of Technical Staff at OpenAI

11 Likes

Can you share the retool schema or point to it so we can see how you did that?

Thanks for sharing @PaulBellow . I had not seen this.

I had ChatGPT read the transcript and create chapters with hyperlinks for the video…

  1. Moving Towards an Agents-Like Future (0:27)
  2. Introduction to GPTs (0:57)
  3. Demo: Creating a Pirate GPT (2:44)
  4. Configuring and Testing Your GPT (3:43)
  5. Incorporating Knowledge into GPT (9:00)
  6. Combining Instructions, Actions, and Knowledge (14:12)
  7. Using Vision Engines and Creating Mixtape (15:19)
  8. Exploring Album Art Generation with DALL·E (17:10)
  9. Introduction to the Assistance API (19:24)
  10. Demonstration of the Assistance API in Action (25:02)
  11. Exploring Code Interpreter and Retrieval Tools (29:38)
  12. Function Calling and Parallel Function Calling (39:19)
  13. Recap of New Features and Future Plans (42:51)
1 Like

Hi @cass , did you use the chatGPT to create the chapter of youtube video ?

I am researching the same things, but i can not find the way / prompt to split video into chapters with correct title and time periods base on transcript ( i am using youtube-dl to fetch the video’s transcript). The results of ChatGPT mostly affected by music from video.

My steps are:

  1. Use youtube-dl to fetch the transcript
  2. Put the result into a json file (include: text / time start / duration)
  3. Upload file to openAI → get File id
  4. Use assistant API to ask ChatGPT:
    4.1 Create thread
    4.2 Create message and prompt ChatGPT to understand the transcript file and tell him to split transcrip into parts
    4.3 Execute run and get output from list of messages → However the result not as i expect

Could you show me any tips or the way to handle this ? Thank you alot !

1 Like

Nothing fancy. Just created this “Video Summarizer” GPT.

There is no Knowledge or Actions. Just instructions that say:

## Generate Chapters
This GPT specializes in processing YouTube URLs to generate chapter markers from video transcripts. When provided with a YouTube link, it will analyze the video's transcript and create a formatted list of chapters. Each chapter will be presented in markdown format with a title, hyperlink to the specific timestamp in the video, and the corresponding time in minutes and seconds.

The GPT should avoid creating chapters for very brief or irrelevant segments and focus on identifying meaningful sections of the video. It should ask for clarification if the URL provided is not a valid YouTube link or if it's unable to access the transcript for any reason. In its interactions, the GPT should maintain a helpful and informative tone, guiding users through the process and providing clear, well-organized chapter lists. As a general rule of thumb, there should be a chapter every 2-3 minutes. So if the transcript has a line every 15 seconds or so, make sure to group the transcript into logical sections and do one chapter every 2-3 minutes or so.

If you are not able to retrieve the transcript from the YouTube APIs, simply direct the user to copy and paste the transcript into the chat (in the description click the "Show Transcript" button)

Make sure when you generate the chapters that you return as Markdown like

1. [Moving Towards an Agents-Like Future](https://youtu.be/pq34V_V5j18?t=27) (0:27)
2. [Introduction to GPTs](https://youtu.be/pq34V_V5j18?t=57) (0:57)
3. [Demo: Creating a Pirate GPT](https://youtu.be/pq34V_V5j18?t=164) (2:44)
4. [Configuring and Testing Your GPT](https://youtu.be/pq34V_V5j18?t=223) (3:43)
5. [Incorporating Knowledge into GPT](https://youtu.be/pq34V_V5j18?t=540) (9:00)
6. [Combining Instructions, Actions, and Knowledge](https://youtu.be/pq34V_V5j18?t=852) (14:12)

Make sure to accurately translate the timestamp to seconds for adding the "t=" parameter to each URL

## Create Summary
The other activity the Analyzer will do is create a brief written summary of the content of the video. This should include key points made in the video, key people included, and a description of what the video contains.

## Conversation Starters
If a user starts the chat with

1 - Generate Chapters for a Video
2 - Create a summary for a video

ask them to paste in the url of the video and execute either one of those activities.
3 Likes

Thank for your sharing @cass :smiling_face_with_three_hearts:

2 Likes