Reducing token and character costs [Discussion]

[orchid bloom]

If you use camelTextToWriteToOpenAIDoesItSave on token cost and work in the same way?

[true hill]

Not sure, but you can check out the tokenizer to see visually: https://beta.openai.com/tokenizer it lets you see how many tokens are made for ur input text

[orchid bloom]

It seems it does not... [edit]

[true hill]

Very very interesting!!!

I'm curious to see if this affects understanding of the model, however

Have you tested the two versions with standardized prompts with temp = 0 to see?

[orchid bloom]

https://discord.com/channels/974519864045756446/1058566991025287270

[orchid bloom]

Temp = 0

[orchid bloom]

Temp = 1

I would like 1% of the value of all of your savings. Thank you x

[swift vortex]

I suppose this would work for fine tuning as well?

[orchid bloom]

And Dall-E

Whitespace is for chumps

[true hill]

Oh sorry I meant like, comparing the answers to various prompts of both the camelcase version and non camelcase version, with temp = 0

Ill try it out later tonight too when I get some time 🙂

[orchid bloom]

Similar outputs

[true hill]

Very cool

looks like its camelcase time >:)

[barren thunder]

am I reading your pictures wrong?

it looks to me like the camelcase was worse, using more tokens

camelcase = 81 tokens
normal = 66 tokens

[true hill]

oh wait

I also misinterpreted it

@fiery topaz It's the other way around

Sorry, wrong fish tag

@orchid bloom

[fiery topaz]

Got me scared for a sec

[true hill]

hahaha

[fiery basalt]

camelcase uses more tokens lol

[orchid bloom]

Jesus christ. Here is why you don't post revolutionary comments at 3am.. Morning all..