“Shit, maybe we should have waited an extra month or two before releasing our broken product…”
And I also mainly use ChatGPT for code, which would make the converrsation a bit interesting without any visual information presented lol.
There’s not enough info though to say whether or not this might complement something else, or be it’s own separate standalone method of interaction.
Also, at first, I thought the same thing. But then I thought about it more deeply, and remembered that if you could combine an interface like this with GPT’s memory capabilities, it could become quite powerful over time. It would also allow the model to understand, memorize, and interpret far more than just what you type.
If it’s audio-native, and has genuine capabilities to assess intonation, this would behuge. This would allow a GPT to interpret sarcasm, and potentially provide it if prompted for it. These kinds of things can be extremely difficult to detect in text without proper context.
My one eval submitted for that sweet GPT-4 early access was sarcasm detection. Remember me!
I use ChatGPT almost always in any lonesome drive. Stopped listening to the radio. I’m having a hard time understanding what difference there is from what I currently use, and what could justify an announcement.
So, I hear you. It’s not just the textual side. It adds the tonal dimension of communication.
But, truthfully, hmmm… I don’t know.
I think we all know that ChatGPT and any LLM by extension is completely emotionally retarded, and that we don’t use these models to emotionally understand us
We’ll find out soon enough though! OAI has gotten very comfortable with just casually rolling stuff out without telling anybody, so the fact they even want to do an announcement clearly means that it’s something worthwhile.
Lord knows we’ll be the ones answering the questions
I will definitely attend the first official LLM - Human wedding
“You have been talking to other guys haven’t you!”
"As a large language model, it is my … "
“YOU BITCH”
Ooohhh… I could definitely see a Las Vegas “get married by ChatGPT” altar
My lord I would resent an API that tries to measure the tone of a human.
“Uhmmm… excuse me, FRANK. Based on our current ChatGPT monitor you have been expressing SOME DEGREE of snobbishness while talking to yourself. You will see this in your ChatGPT generated report at the end of the week”
“Our Breitbart-focused sponsored media outlet by ChatGPT has indicated that this snobbishness results in flamboyant liberal thought processes and WE DO NOT SUPPORT THAT HERE”
“Cambridge Analytica? What’s that? My personal ChatGPT assistant is sensing some sort of snarkiness in your tone. I am blocking you for a week.”
Dropping gpt-4-turbo price to something similar to gpt-3.5-turbo would be magic to me… gpt-4-turbo is 20/30 times more expensive than gpt-3.5-turbo, this is just unusable for a lot of projects… and last gpt-3.5 versions got lazy/crap, the last good version is gpt-3.5-turbo-0613 but will be deprecated on next month
model:
description: *model_description
type: string
instructions:
description: &assistant_instructions_param_description |
The system instructions that the assistant uses. The maximum length is 256,000 characters.
type: string
maxLength: 256000
null
Considering new assistant maximum instruction length was checked in 3 days ago, no, not a month old. 256k char = 64k-256k+ tokens, depending on world language.