Love you !codecels, 
so I'm gonna show you how to get GPT-4 access for free. A few major caveats:
Scale will definitely shut this down at some point, so use it while it's available.
This might not actually be GPT-4. There's not really a way of knowing. I'm about 98% sure it is, but they may swap it out for 3.5 Turbo during outages (?)
You need some form of API interpreter for the JSON it spits out. Here, I'm using TavernAI, which is designed to be a Character.AI-like "chat" interface, with the ability to import and design "personalities" of characters. Great for coomers. Here's some pre-made characters, if you're interested (Some NSFW). Just download the image and import it.
Every message you send will pass through OpenAI's API, Scale, and If you don't change the API key, it will also pass through the 4chan guy who hosts the github's Spellbook deployment. There is exactly ZERO expectation of privacy. Don't be retarded and type illegal shit or personal info.
As of right now, the OpenAI API is having an outage. These are pretty frequent.
(This is why I'm writing this thread rn instead of fucking around with GPT-4)
Now, how to actually set this up, using TavernAI, for lazy retards:
Have Node.js and git installed and know how to use them. This is /h/slackernews, I won't explain this part.
Make a temporary email. Just google 'temp email'. Turn on a VPN for the entire session as well, if you're really paranoid.
Head over to https://spellbook.scale.com/ and make an account with the temp email.
Create an "App", name and desc. don't matter.
Make a variant, and select GPT-4 in the dropdown.
- If you're wanting to use the API with Tavern to emulate a chatbot, you should add the following to the 'User' section:
Complete the next response in this fictional roleplay chat.
{{ input }}
Set the 'Tempurature' to somewhere between 0.6 and 0.9. 0.75 works fine for me.
Set the maximum tokens to 512 for chatbot length responses. (You can increase this but it requires tweaking the TavernAI frontend.)
Save the variant. Go to the variant and hit "Deploy". You'll see a "URL" and an "API Key". Copy these down or come back here in a minute.
Open a terminal in a new folder, and run
git clone https://github.com/nai-degen/TavernAIScaleNow run
cd .\TavernAIScale\then.\Start.bat(or.\start.shfor lincux)TavernAI should launch automatically, but if it doesnt, go to
http://127.0.0.1:8000/in your browser.In Tavern, go to 'Settings' on the right. Switch API to 'Scale'. Copy the API Key from the Spellbook page that you saw earlier into the API Key field. Same thing with the URL. Press 'Connect' to verify it's working. If it fails, either the API is down or you pasted the wrong Key / URL. Make sure you're using the URL from the URL field here:
- Now, use one of the default anime characters
, download a coomer character from here, or make your own.
- The API has many other uses, obviously, but the chatbot is the simplest way to get this up and running. Try fucking around with the "Main Prompt" and "NSFW Prompt" in the settings for some interesting results, or to tweak your desired output. Try pressing "advanced edit" on a character (or making your own) and messing around with personas and scenarios. It's pretty damn cool.
That's it. Have fun until this shit dies in like 3-4 days. Please try not to advertise this or make it known outside of rdrama and /g/. We don't want Scale to shut this down earlier than they already will.
Jump in the discussion.
No email address required.
This seems quite promising
Jump in the discussion.
No email address required.
Turn this shit down a bit I think
And also don't copy paste an entire book into it
Jump in the discussion.
No email address required.
Sorry, i meant that this warning suggests that it's indeed GPT-4 since GPT-4 famously supports 8192 tokens rather than 4,096
Jump in the discussion.
No email address required.
Ah shit, good point. I think it might swap out during high load (according to 4channers who are likely wrong) but seems like it's real.
Jump in the discussion.
No email address required.
More options
Context
More options
Context
More options
Context
More options
Context