drowsy silo
This is my ChatGPT clone, refined to look exactly like the real site as possible! However, in my site, the text sections are made larger and there are slight UI differences. You can use any model you'd like except gpt3.5 (not working yet). Let me know if a specific code isn't being recognized so I can change that. https://chatgpt-messenger-mu.vercel.app/
tranquil nimbus
chat api broke @drowsy silo
snow nymph
no way this isnt a data gathering operation
drowsy silo
?
steady yacht
Light and dark mode doesnt work btw
steady yacht
Also when its on light mode user's prompt doesn't turn black
cursive zephyr
@drowsy silo
woow nice!
I'm doind a chrome extension...if you want to colaborate DM me!
cursive zephyr
ok 🙂
misty ledge
What's the difference
cursive zephyr
This is my project 🙂
https://discord.com/channels/974519864045756446/1081159383104356395
drowsy silo
Different models give different responses
Like text-davinci-002 gives shorter and faster replies whilest davinci-003 gives longer and more detailed ones
the most detailed one ik is davinci, it takes at least 3 minutes
misty ledge
I think the original chatgpt is better than gpt turbo, it has more computationally expensive
Nothing beats DaVinci 3.0 tho
drowsy silo
I hear that
The responses sometimes break for long requests
and returns nothing
but in most cases it works
misty ledge
You could make it better using prompt engineering
Use the api to redefine the inputs given by the user for a better results with prompt engineering
I'm pretty sure it can do the job
Not an engineer here,just a suggestion
drowsy silo
I see
misty ledge
Does that make any sense though? XD
drowsy silo
I'm planning to get it to work with using an if statement condition: if (model === 'gpt-3.5-turbo')
Kind of, the back-end is kind of new to me
misty ledge
I also like to suggest this
drowsy silo
Shoot
drowsy silo
uh, well yes
drowsy silo
Idk but you mean but ok
misty ledge
I mean there is some api that can rewrite everything AI has written in a way that AI content detectors won't be able to tell
But I don't know if that's possible
I hope you get it now
Nvm
drowsy silo
?
misty ledge
Nice website though
drowsy silo
Expensive?
misty ledge
Yeah,running those apis
drowsy silo
ah, you mean the amount of requests that is made?
misty ledge
Yeah
drowsy silo
Good question
Surprisingly now
If it was I'd taken down the site to not exceed $18
but only <$3 has been used
drowsy silo
Indefinitely most likely, and I'll be updating it as I add major updates
or until my free trial expires
misty ledge
Alright
cosmic drift
Shouldn't take credit for other peoples work. Actual creator: https://www.youtube.com/watch?v=qwM23_kF4v4&t=1317s
drowsy silo
What?
That's not his work dude.
My UI isn't even similar to his
but I did try to follow his tutorial, it's full of errors
drowsy silo
Update - I got gpt-3.5-turbo and gpt-3.5-turbo-0301 to work! I will be uploading the change either today or next week! In addition to that, I will try my best to get whisper-1 fully functioning! 🙂
drowsy silo
which database you used ?
Database to store the chats? Firebase
Also I can't, GitHub isn't co-operating with me to let me upload it, plus I don't want to
Apologies
drowsy silo
Temporarily down - firebase is not working
drowsy silo
Update - The site is back up! Issue has been resolved
young briar
i want chat gpt to like have the ability to exploit the log of msg x_x
that would be appreciated if you could make that
drowsy silo
Y'all you can use this since the mainsite is down yk
young briar
What do you mean lol
i mean like
you can save all your history with you and chat gpt
cuz like apparently today it cleared EVERY important question i asked to it
entirely wiped out of the system
=_=
drowsy silo
ohh I see
drowsy silo
New build coming up soon, with MAJOR UI changes 👏
you won't be able to spot the difference between this and the real thing
acoustic estuary
opensource when?!?!?!?!?!
haughty widget
This is my ChatGPT clone, refined to look exactly like the real site as possible...
Can you please send the link?
since it is running on gpt~3 it could have no ethics and overall be more stable, etc.
cinder solar
write my article on the car
drowsy silo
since it is running on gpt~3 it could have no ethics and overall be more stable,...
That is the link lol
drowsy silo
opensource when?!?!?!?!?!
When it's fully finished and when GitHub decided to cooperate
acoustic estuary
When it's fully finished and when GitHub decided to cooperate
drowsy silo
yeah dude
Like i've been trying to upload it to Github but it keeps giving me errors
acoustic estuary
Like i've been trying to upload it to Github but it keeps giving me errors
just git it
git!😋
drowsy silo
I tried that
mint lagoonBOT
@untold rose
ds0nt's message blocked by AutoMod
drowsy silo
I'm using your api key? 😄
Yes you are
Autoscroll? Bet Imma add that after the I get the typing text to work
untold rose
Ahh finally I'm unmuted 😢
You doing it in react? If you get stuck hit me up cause I'm basically an expert 😄
https://gist.github.com/ds0nt/9f7d1c7fca5c64af22e6e89c53c90ecd
There's some code for my autoscroll for an abandoned chat app 😄
drowsy silo
Mainly in TSX
drowsy silo
It is
untold rose
one just saves you from undefined and type errors at the expense of short term prouctivity
drowsy silo
TSX is essentially JS
drowsy silo
New models are working and refined. Next update will be soon.
final fern
its pretty cool bro!
feedback... umm... does it remember previous conversation like the original chat gpt?
i asked it few Qs about it...
nevertheless it's pretty cool!!
good job!!
drowsy silo
feedback... umm... does it remember previous conversation like the original chat...
The newest one, gpt-3.5-turbo will remember previous conversations
Also thanks, light and dark mode will fully work with the newest build. I just need to tweak it a lot
acoustic estuary
OPENSOURCE i beg you
acoustic estuary
IT's NOT FINISHED yetttttt
can i suggest features?
drowsy silo
can i suggest features?
who's stopping you tbh?
Honestly I'd LOVE it if I got more feedback 🙂
acoustic estuary
who's stopping you tbh?
custom saveable initial prompt to initial custom personality when adding new chats
acoustic estuary
drowsy silo
What do you mean by initial prompt?
like give the option to change bots personality?
acoustic estuary
also completely customizable args settings
acoustic estuary
like give the option to change bots personality?
yes
drowsy silo
args?
Ah, I got a friend who implemented that
bet
i'll also have a "default" option
acoustic estuary
wdym by custom saveable initial prompt?
system message, when creating a new chat, let the users define their own ones and add the ability to save/edit/delete them
acoustic estuary
its like the "initial" folder in my project #1080661517390843915
drowsy silo
got it
drowsy silo
@granite oak
granite oak
hello
drowsy silo
hi
drowsy silo
@junior rose try this
vocal goblet
@gentle stump can you help me read a acticle
drowsy silo
This is my ChatGPT clone, refined to look exactly like the real site as possible...
Use the site
tight summit
@drowsy silo hi, bro. Can you give me advice on something? I read through all the chat here. May I know what's the difference between GPT 3 and 3.5turbo? I understand from documents that GPT3.5 is more powerful and better reasoning performance. So does that mean GPT3 has weaker reasoning and can't produce result like 3.5? Let's say I give the bot a question to build a website with login form and a submit button. GPT3 might produce buggy codes as compared to 3.5? Am I analyzing it right?
drowsy silo
<@766370832447569990> hi, bro. Can you give me advice on something? I read throu...
Sure! GPT 3.5 is the API that ChatGPT uses, which includes Text-Davinci-003 which is technically a GPT3.5 model. These models are more capable of generating better and more detailed responses and ones that can help. Take how the real Chat GPT is able to generate useful answers! Using this model in the API should be able to achieve the exact same thing.
I'm working to get it to work like that! :D
tight summit
But it doesn't have memory on previous chat.🥺 True bad things for developer.
tight summit
Possible to make it? The only feasible idea to work around is just resending previous chat to it. It costs more tokens.🙂
acoustic estuary
Possible to make it? The only feasible idea to work around is just resending pre...
its not a work around, its just how its done. chatgpt's official website is doing this too!
the official way to let language models "recall" memory, is just telling it how to respond
tight summit
its not a work around, its just how its done. chatgpt's official website is doin...
But using this method, isn't that history chat tokens accumulated for each time request and response? And we have max_tokens for each chat. Please correct me if I am wrong...I went through documents but can't find the solution.
acoustic estuary
But using this method, isn't that history chat tokens accumulated for each time ...
yes, it will accumulate, if you hit the max tokens limit for your current model, you need to discard some of the memory to be able to make the request, chatgpt's official website does this too, thats why its memory is limited to 4k tokens.
drowsy silo
I'm working a way around this as you speak, I'm going to try to make it remember previous conversations
drowsy silo
This is my ChatGPT clone, refined to look exactly like the real site as possible...
Site temporarily down, will ping you all when it's back up, after I get it working and deployed again new models and functionality will be added!
@tight summitTo let you know, GPT-3.5-turbo model I implemented SHOULD remember the previous question
fickle ember
You doing it in react? If you get stuck hit me up cause I'm basically an expert ...
Interesting. I have something for u.
@drowsy silo , we have done similar like this: with open ai endpoint it gives history.. or creating backend server endpoint, u can keep the history and show like chatgpt. We r looking at extending this to web data. Any cool features getting GPT?
drowsy silo
Oh really? Wow, you've peaked my intereset
I'd like to know what you did and how you did it 🙂
fickle ember
Anyone is interesting in making a community project in react, let me know. DM me. We have build something and there are many cool developers built open source stuff. We can build a good one.
fickle ember
I'd like to know what you did and how you did it 🙂
Sure. Happy to chat. Will DM u. Time to sleep after a long day.
drowsy silo
I see, good night! Sounds good
fickle ember
Whilst gpt4 is still connecting to web data, we have built a real time AI product which scans millions of social media data and provide visual insights on key trends, issues . Www.Haix.ai
Here is sample of latest in tiktok ban in Uk parliament, our ai analysed 000’s of tweets and shared visual insights
tight summit
@drowsy silo is your url still valid? I can't open it to test though 🥺 https://chatgpt-messenger-mu.vercel.app/
drowsy silo
It is, but like I said it's temporarily down but I'll lyk when it's back up
drowsy silo
I got access to GPT-4, expect more changes soon! 🔥
drowsy silo
Update: I fixed the issue. Now I just have to fix the domain and it should be redeployed soon!
copper rapids
yes, it will accumulate, if you hit the max tokens limit for your current model,...
a better solution I'd recommend is passing the history through a dedicated summarisation tool such as https://huggingface.co/Qiliang/bart-large-cnn-samsum-ChatGPT_v3, which will condense the text to a customisable amount of tokens and pick out the most important parts; it's quite a bit better than leaving out old messages altogether
(this is assuming it hasn't already been implemented of course)
acoustic estuary
a better solution I'd recommend is passing the history through a dedicated summa...
or just use embeddings: #dev-chat message
drowsy silo
I'll have a solution ready
copper rapids
or just use embeddings: https://discord.com/channels/974519864045756446/10375611...
that's something I've been really curious about honestly, but never fully understood. how does it perform? as in, since I assume all that's being stored is the encoded integers being scanned for similarity, will it be confused if you ask for the previous question? how're you doing the search and how's the time complexity?
like, would https://en.wikipedia.org/wiki/Locality-sensitive_hashing be a good way to optimise searching?
north olive
Write me best food review
acoustic estuary
that's something I've been really curious about honestly, but never fully unders...
copper rapids
https://discord.com/channels/974519864045756446/1080661517390843915/109051597139...
hmm that's about what I thought actually
there's a limitation in that unlike the main llm the embedding comparisons only compare how relevant text is, and doesn't actually extract any more info
I wonder if it's any better to stash multiple messages in a single record, and overlap 50% of adjacent records so any given message will always be in two records? that way it might be a little better at reading larger snippets?
I'm hardly that educated about this stuff though
drowsy silo
Its alil bit hard on the phone
i'll fix that, you can get it to work on a phone if you change it to a desktop view
fathom gate
i'll fix that, you can get it to work on a phone if you change it to a desktop v...
How to do that
quaint mist
can this be hosted on a cpanel based web host?
and if yes, where can i find the code and stuff? 😄
real forge
Thread is being taken over by me - I temporarily do not have access to my old account. This entire project is being revamped and refined for the best user experience. Please sit tight.
fast scaffold
Updates/dev. roadmap?
real forge
Project is done.
real forge
New features:
Site link: https://chatgpt-dev09.vercel.app/
Developer Roadmap: https://trello.com/b/s1cT3qDj/chatgpt-clone-roadmap
real forge
I will be working on getting the responses to show up, it is an issue I am well aware of-
verbal oasis
New features:
- The site's UI has been massively modified and revamped.
- GPT-3...
For gpt 4 does it use our account or your api ?
real forge
it uses my API key dw
certain lengthy responses don't display though, I'm working on a solution
real forge
it's still in beta tho
real forge
Conversation history works, try it out with: gpt-4, gpt-3.5-turbo, and gpt-3.5-turbo-16k-0613
@copper rapids it works now
@tight summit oh yeah, you too! Conversation history/memory works now
You can ask it what you asked before and it will reply
real forge
text-davinci-edit-001 has been added. You can now have the bot fix the grammar of anything you enter! New models are being tested.
verbal oasis
I want to speak with the manager
real forge
?
real forge
if it's about some responses not showing up, i'm working on it
verbal oasis
if it's about some responses not showing up, i'm working on it
That's was a joke, was about Android website. I have to try the solution you gave
real forge
oh haha
real forge
Requests paused for the moment (I used too many tokens in testing lol)
remote spire
I see this is down. Just wondering why? It looks like an interesting project.