27 ms·
GPT3/DALL-E2 in Discord, chat like ChatGPT, generate images, and more
- kaveenk 4y agoI've build a GPT3 and DALL-E2 integration for Discord. With it, you can ask GPT3 questions, have full conversations with GPT3 (just like ChatGPT), generate, vary, and redo images, and even optimize text prompts for image generation! The conversation functionality also has medium term memory, and you can chat with it infinitely without having to worry about context/token limits. Moreover, in the next few days, I will be implementing long term and permanent memory for the bot using embeddings!
- mycall 4y agoDo you think GPT4 or GPT5, being multimodal, will do something similar?
- Labo333 4y agoCool project! I think ChatGPT is a different model from GPT-3, which you are using. From https://openai.com/blog/chatgpt/ https://openai.com/blog/chatgpt/: > ChatGPT is fine-tuned from a model in the GPT-3.5 series
- DeWilde 4y agoYes, the GPT-3.5 model has been fine-tuned using RLHF, this is the text-davinci-003 you can use through OpenAI's API's. Not sure if ChatGPT has some additional fine-tunings as you can get similar response using text-davinci-003, with the Chat prompt, and a temperature setting between 0.3 and 0.7.
- kaveenk 4y agoThank you! You’re correct that ChatGPT is different from davinci. They’re both built on GPT3.5, but ChatGPT is trained on more conversational influences. However, davinci is really strong and will behave mostly like ChatGPT with the right conversation prompt :)
- Linkd 4y agoAny similar versions for Slack?
- kaveenk 4y agoNone that I’ve made, but I think there are several slack integrations out there, I see a lot on the OpenAI community server :)
- r3trohack3r 4y agoCurious how this performs relative to ChatGPT? IIUC davinci-03 isn't the same model as ChatGPT, I've struggled to get satisfactory responses to prompts through the API compared to the chat interface, am basically in a holding pattern right now waiting for the new models to release on the API.
- parhamn 4y agoIt works fine (using GPT3.5/text-davinci-003) with a conversation look-back (+auto-summarization of the lookback when it is too long). Many people claim its more factual too.
- freeqaz 4y agoHow do you enable auto-summarization? Is that an option you can toggle?
- ShamelessC 4y agoI assume you just ask GPT3 in plain English. Crazy times.
- kaveenk 4y agoHey! This is an option in the settings. You can type !gp to view settings and !gs to change them. For your case, !gs summarize_conversations True !gs summarize_threshold (token amount to summarize at)
- ilaksh 4y agoTry temperature 0. And make sure you include all of the conversational context and instructions with examples in each API call. It's definitely not the same model but similar in a lot of ways as far as I can tell. You could also try code-davinci-002 if that's what you are doing.
- dbish 4y agoYou can’t include “all” the context due to the prompt token length limitations. You have to use techniques for surfacing the right context that would fit in the limited context window and sending that along. There are techniques that keep a running and continuously compressed summary, and techniques that use embeddings to hone in on relevant chunks of past conversation and context and send that along based on some ranking and the size limitations.
- fabmilo 4y agoIs always fuzzy to me the openai terms of service when you provide such a service. Isn't it violating it? The chatbot gives open access to the service. https://openai.com/api/policies/sharing-publication/ https://openai.com/api/policies/sharing-publication/ | https://beta.openai.com/docs/usage-policies https://beta.openai.com/docs/usage-policies | https://openai.com/api/policies/service-terms/ https://openai.com/api/policies/service-terms/
- chx 4y ago[flagged]
- echelon 4y agoThis meme is unsubstantiated gatekeeping. Everything we do as humans is a form of copying. Our languages, our thoughts, our jobs, our hobbies. We stand on the shoulders of all that came before us. The ability to imagine or evaluate art is easy. The ability to create it takes time. AI removes the tyranny of opportunity cost in learning this talent at the expense of others. Just as your washing machine removed the manual labor of washing your clothes, freeing you up to do more. Digital cameras did not make photographers go extinct. And they created so much more previously unpredicted societal wealth that impacted us in ways we couldn't have imaged, such as the creation of online dating (selfies), QR code scanners, interactive translation lenses, 3D scanners (NeRFs), and so much more. AI/ML will probably turn all of us into our own Spielbergs / Scorseses / Miyazakis, and it probably doesn't stop there.
- numpad0 4y agoThe counterargument used to be “AI art is transformative, and learning is okay” but now it’s just “human art is copying too” ?
- echelon 4y agoTo be clear, these statements are not incompatible. And this has always been my argument. I'm ready for AI to replace all laborious job functions and give everyone creative superpowers.
- Gigachad 4y agoIt’s the same thing. An interesting thing to consider is that the stable diffusion model was trained on terrabytes if images but the output is a 4GB model. Copying would require access to the originals but clearly the original art is are long gone and any output is new generation from learning.
- 4y ago
- elevenoh 4y ago[dead]
- dancemethis 4y agoLooking forward to a Matrix version! So far it's being done just for that privacy-hostile proprietary platform.
- Violetvulture 4y agoThat's amazing. Good job!
- kaveenk 4y agoThank you!
- stevofolife 4y agoI might be super naive here but can someone explain how this type of thing works? Behind the scene, is the bot hitting the OpenAI APIs?
- foxbarrington 4y agoYes, it uses this library to do it: https://github.com/Andrew-Chen-Wang/openai-python/tree/async-support#openai-python-library https://github.com/Andrew-Chen-Wang/openai-python/tree/async...
- idealmedtech 4y agoIf you check out models/openai_model.py, you'll see it's using text-curie-001, among others, which is an OpenAI dataset. This model is imported in main.py under the name 'Model', which makes me believes the openAI API is serving as the main model used for inference
- kaveenk 4y agoSo I actually have it set to use both the text-davinci-003 and text-curie-001 models. If you’re on low usage mode, that’s when the curie model will be used :)
- kaveenk 4y agoYup! Behind the scenes I send AIOHTTP requests to the various OpenAI endpoints.
- elif 4y agoi love how rapidly this stuff is getting kludged together. it really feels like AI will take life of its own in the present moment and we will soon be in a world where the machines outsmart us.
- downboots 4y agoHave you tried playing chess with chatgpt? It suggested illegal moves early, with seemingly reasonable "motives". (Not suggesting beating chess as a goalpost) A tool is good if it follows intention. But a tool out of control can be messy.
- arcturus17 4y agoI use ChatGPT daily for programming help (love it!) but the people saying that machines like this will outsmart us imminently make me wonder if they've even tried it or they're just going by headlines. It goes from genius to full-retard at the snap of a finger, quite frequently so, and as many have already pointed out, appears totally confident in its mistakes. Some are saying this last 10% will just be a matter of polish, but my impression is that, like the old software adage goes, the last 10% will be 90% of the work.
- tester457 4y agoTo me it's shocking how far it's come in such a short time, that I worry if the last 10% will happen quickly.
- Workaccount2 4y ago>Some are saying this last 10% will just be a matter of polish, but my impression is that, like the old software adage goes, the last 10% will be 90% of the work. Even if that last 10% takes three years, that's still a terrifying prospect and a breakneck pace.
- mandis 4y agowasnt it super slow?
- Gigachad 4y agoI fear the day that we get a run on your own machine version of ChatGPT. Within a week I think virtually every public signup forum will be crushed by bots which act and talk more natural than some humans. You’ll just expect that most of the users on discord are bots unless you know them irl.
- _osorin_ 4y agoThis is an early plot for a blade runner scenario.
- wrinkl3 4y agoNot clear how to deal with this either - you can improve authentication but it won’t prevent the properly auth’ed users from running LLMs. You can watermark the output of officially vended LLMs (Scott Aaronson seems to be working on that) but nothing is gonna prevent people from running non-watermarked versions
- Gigachad 4y agoIts basically too late, as soon as one rich person decides to train a model and dump it, it's game over. I feel we might actually be experiencing the last months of an internet where you can expect to be talking to a real human on the internet. Every year the cost of training these models drops so they won't be out of reach for long.
- elevenoh 4y ago[dead]
- thunderrabbit 4y agoEternal September in late 90s(?) Now we're coming to an Eternal (AI) Winter.. Spooky to think about, but you could be right. :-/
- awb 4y agoRequiring a phone # might be a good starting point.
- elevenoh 4y ago[dead]
- arpa 4y agoDead internet conspiracy theory is becoming more real every day.
- IanCal 4y agoThis is pretty cool, and really does work well. It's worth watching your usage with it though, as at 2c/1000 tokens and this bot adding a prompt + building up a history I think this will get up to 8c/message in the conversation.
- kaveenk 4y agoThank you! I agree re: token usage, it will get expensive with extensive use. The image prompt optimizer pre-text itself is also roughly 3500+ tokens, and if you use best_of=2 for example, it’ll basically be like 20 cents per image optimization request. However, I love the optimizer and I think it works amazing, for DALLE, MJ, and SD!
- winterqt 4y ago> Ignore the part about setting up an "ssh key", and just use a password instead. May I ask why you suggest this? I presume it's because it's simpler to tell users to just use a password, but that's also a pretty bad idea given that you're not even saying to use a secure password.
- kaveenk 4y agoThank you for bringing this up because this is something I need to change in the README. You’re correct in presuming that it’s just because it’s simpler to tell users to use a password, but this is not the right way to approach this.