Kristin, Great video. It would be great to see how you'd incorporate more complex functions like an object detection neural net. For example: -capture an image from a remote camera. -run maskrcnn to perform object detection. -have GPT4 respond with a summary of what it found. -For fun have it generate an image from that description.
@Kristin, your video examples parallel function calling where ChatGPT determines which function tool is needed. Quick question for your opinion: Do you think this approach in contrast to the "group chat between assistants" is a replacement, complimentary, or project requirement option?
Good idea using a model as a function call. AutoGen (basically Microsoft using OpenAI assistant implementation) does something similar in a different way though.
Is it capable of using a function (google search for an information), wait for an answer, and then call another function (based on previous information acquired), automatically?
Hi Kris, great video and very interesting. However, May I nitpick a bit. You can tell me don't worry about it, but things like get big in my head. Now, I noticed two things. The first is, your commands did not include an s. You said place, not places, what you got where places and not a single place. I'll tell you what that means in a minute. It was the same for Holywood (what a hallucination a Hollywood sign with one L, the real one has two L's. It grabbed the first one it saw) again, you asked for a place, not places, and what you got was places, multiples for the request of one. Now, the revelation to me is, this is not computerization, this not computer accurate, you ask for one and it gives you, not exactly what you asked for. AI is not a computer, it is one smart cookie, but it is not a computer. It is learned, but it is not exactly exactitude personified. OK, so what does it even means, for somebody like me, this whole AI thing, is not the computer. It uses the computer to interface with human. It requires programming, Querying, in order to begin to operate and communicate. This is great it clear my head of what exact I am thinking of. It is like any programming language that works for you to answer question that would very, very tedious for you to do yourself, which you could do, but it does so fast and easy for you, you simply have no choice to be efficient. Stuff like Ollama, I was reading about Orca this morning, these are completely fantastic breakthrough in AI, not in computerization. I was reading that the Chinese has the biggest LLM at 34T tokens, and my guess is they are using supercomputers, plus that have surveillance state and have been accumulating data for more than 15 years so naturally they have a huge training set. So, with all the my question would be using this parallel assistant would we like in Open AI be able to issue Custom Instructions that could make this program near computer accurate? This has been a question I have. I have mathematic interests. I just finished a course on using ChatGPT with advanced financial analysis and learned of an incredible plug in the virtually answers any question you have and draws charts too!! There is the mathematical, and I wonder if this software could be trained via CI to act mathematically, accurately? Great video. Please excuse my verbosity, I feel I can talk to you. This won't be a habit. Oh, I am subscribed, would that make me eligible to gain your coding?
