Комментарии:
If you want Copilot do perform better and complete entire methods, you'd need to just write a comment describing what you want to achieve, then it will often just write the entire multi-line code for you.
Ответитьhaha imagine telling the model if it has an option that has the word item in it, make sure that items are added or removed properly with a qty dictionary.
ОтветитьThis is amazing!!!😇
ОтветитьDo you prefer Mac or Windows for development. I see you are using a Linux setup here but if you had to choose between Mac or Windows which one would you pick?
ОтветитьCRTL + Right arrow........this will take just the next value in copilot.......love the vids my dude........I too no longer code
ОтветитьYou USA guys are really enjoying it but here I'm still waiting for GPT4 API access from 2 weeks
Ответить"teapot" would make sense
ОтветитьDALL-E is so bad compared to MidJourney and even SD.
ОтветитьHello! Thanks a lot for your videos. You have an NVIDIA DGX Station A100. Please make a video to train (FineTune) the LLaMa model on it.
ОтветитьInteresting and endless opportunities ahead.
ОтветитьI have one related question on Embeddings. So every time a user asks the bot a question, the bot searches the knowledgebase and answers. the following use case impossible:
USER: How can I register for starter training ?
BOT: If you meet all the conditions, you can register from the link …
USER: What does it cost ?
BOT: I dont know.
how can i deal with this problem.Since the user specified the course name in the previous question, but not in this question, there will be no match from the knowledgebase. How can I search while keeping the context from the previous question?
You mention the "System" parameter, and you had a problem where ChatGPT responded with "Great!" when you didn't want it to be conversational. As I understand it, the System parameter will be a pre-prompt that can be used to effect the persona that flavors ChatGPT's responses (as well as limiting the kinds of questions you can ask). In other words, I think you can use System to improve the response to get rid of "Great!" and possibly even to return HTML as you mention at one point.
ОтветитьAnother awesome vid, thanks! I was thinking, why even limit to options instead of letting the user type what they want to do next and make it completely open ended? The generative properties make this kind of interactive game possible for the first time.
ОтветитьAccording to Sam Altman, system is for coaching the style of response in gpt-4.
ОтветитьPlease complete the neural network series.
ОтветитьWatch out for power consumption, it's an obstacle for exponential growth of ai. So we will be okay unless they come with efficient hardware to run on (neuromorphic).
ОтветитьThis is a really cool project! I've played a lot of these CYOA HTML games (most of them NSFW lol) so the possibilities do seem crazy
Ответитьi think the entire "button states" is irrelevant oO
Ответитьhave you tried chatgpt as a full on dungeon master ? dnd style
Ответитьi guess it would be a good idea to validate the output of the ai very rigously (i.e. only numbers in Option [0-4]) and if anything doesnt match just regenerate the response
i would also probably try to let the image generation just do a few more attempts if it fails before using the backup
🤙🤙🔥
Ответитьnice
ОтветитьDon't worry about copilot giving you code that's how things are going forward.
ОтветитьHappy you're back sentdex! About the option 1a, 1b, 1c, etc. I think this happened because you did not actually use the input. e.g., you never use your next_inp variable. It's always looping through the same options 😜
ОтветитьCan you create a video of how to build your own gpt from scratch??
ОтветитьMicrosoft Bing has integrated DallE and GPT into the engine
ОтветитьThank you for this amazing content. It was fun to see you using copilot-x.
ОтветитьHello 👋
Ответитьcome on text to image AI, learn to count to 2
ОтветитьThis is incredible Harrison. I can't imagine where we will be in even 6 months with this level of advancement.
ОтветитьletTT THE APPS COMEEE!!!
ОтветитьThis is really great, would love to see a video on the new plugins, but srsly you're the GOAT
Also, when do you think are we going to integrate it with github issues? so for new issue it will: read the repo and issue, create a branch, make changes, run tests, run code, check output is alright, push commit, respond with summery on github issue as a comment, read user comment with further instructions/confirmation, then finally making a PR ftw
(I'm guessing this will use mixed models so not be prohibitly expensive, though it would still probably be cheaper& faster than a development team)
Reminds me of AI rouglite
ОтветитьWith Stanford Alpaca 7b recently released and SD, this can run locally and with some fine tuning probably even on an iphone.... wild times
ОтветитьThe occurrence of the unexpected "Option 1b2" bug after 100 test runs is proof sufficient to me that this tech best be kept the hell away from production environments.
The toy is still neat, the mashup of services is cool, and this is a good live coding vid. Thumbs up!
Why did you not use the role "system"?
ОтветитьThe presenter's passion for the topic really shines through. 🤩
ОтветитьThis feels a little AI dungeon like.
ОтветитьI tried something very similar, also stumbling over the Dall-E filter a lot! Ended up using stable diffusion locally, using CUDA/GPU. That worked like a charm, but now I am struggling to package and deploy the app with the stable diffusion model included. It obviously will bloat the app to over 5gig unless we get Alpacca like models for image generation. But it would still be great for PC games where space is not so much an issue. I just cant get it packaged and deployed correctly. Would be great to hear how others handle that!
ОтветитьThe other method works, but it will crash if the key is missing. Where .get() will not, and .get() lets you specify what to do if the key is missing. .get('list_of_options', ][) will return an empty list if list_of_options does not exist.
Ответитьyou should try using GPT-4
ОтветитьSo to use such a general models like ChatGPT for software applications maybe we need to do fine tuning the model to have a chatbot that serve as a sales man or in the cusomter service in some company. But because ChatGPT is not open source, doesn't that put restraints to develop more software with ChatGPT?
ОтветитьThe new GPT videos are really amazing! Glad to see you're able to manage the family time and also experimentation time!
One note: You mentioned no one has thought along these lines: Visual Chat GPT was a really popular paper that did something along similar lines (I made a mini demo using it)