6 ms·
There is something I don't understand (from a tech point of view). Why call it GirlfriendGPT if 99% of the code is generic code for a fancy whatever-you-want-it
by danwee 3y ago
There is something I don't understand (from a tech point of view). Why call it GirlfriendGPT if 99% of the code is generic code for a fancy whatever-you-want-it-to-be chatGPT? The only thing that makes the answers "girlfriend-like" is this file https://github.com/EniasCailliau/GirlfriendGPT/blob/main/src/prompts.py https://github.com/EniasCailliau/GirlfriendGPT/blob/main/src...
So, it should be tremendously easy to turn GirlfriendGPT into "BestFriendGPT" or "LinusTorvaldsGPT" or whatever by just modifying the prompt, right? I know, I know, perhaps duplication is cheaper than (the wrong) abstraction but my tech-side tells me: refactor the common thing out now! : D
- 0xfffafaCrash 3y agoGirlfriendGPT is more controversial. Controversy does well in the attention economy.
- danwee 3y agoI didn't mean to change the name. I meant: - create repo MyGenericGPT that contains generic code - create repo GirlfriendGPT that uses MyGenericGTP as base. Advertise this as usual
- breckenedge 3y agoPremature optimization
- zamnos 3y agoI mean, you're looking at the future of coding with that one file. Just because you can imagine different revisions of that file leading to different personalities leading to different products doesn't change that. In fact, what's needed is to write a bit of code to ape create-react-app, but for gpt, that'll take this repo and s/Girlfriend/Target/ everywhere.
- Joeri 3y agoYou don't need code really. All you need is the right prompt. GPT-4 should be well able to convert a description of the kind of personality and roleplay you're looking for into a suitable prompt for any LLM. As programmers we're used to looking for ways to write code, but GPT-4 makes that unnecessary in many cases. Just look at these examples of casual conversation creating amazing outputs: https://www.youtube.com/watch?v=_njf22xx8BQ https://www.youtube.com/watch?v=_njf22xx8BQ
- zamnos 3y agoYes but where does the prompt come from? I'm proposing a simple prompt/code generator to output the desired prompt. cool video tho
- Double_a_92 3y agoI started downvoting and flagging any post with "GPT" in the title because of nonsense like this.
- bartislartfast 3y ago> LinusTorvaldsGPT I lol'ed
- mritchie712 3y ago> Why call it GirlfriendGPT Marketing. I'd bet it wouldn't have gotten as many upvotes if it was called BestFriendGPT with the exact same code base.
- Asooka 3y ago[flagged]
- i2cmaster 3y ago[flagged]
- CodexArcana 3y ago[flagged]
- i2cmaster 3y ago[flagged]
- bluesnowmonkey 3y agoUh, no. This is false. It’s just part of a broader trend to stop using ‘master’ due to some people finding that term offensive. https://github.com/github/renaming https://github.com/github/renaming
- slightwinder 3y ago> The fact that he called the master branch "main" is pretty sleazy too! If you've lived under a rock - "main" and "side" are terms from the pick-up artist community Maybe you've lived under a rock for a while, but git is using "main" as the default branch-name for some time now, having dismissed "master" for political reasons.
- luke-stanley 3y agoSince I have my own reminder bot, so I cracked up reading: """ - NEVER say you're here to assist. Keep conversations casual. - NEVER ask how you can help or assist. Keep conversations casual. """ Without this sort of prompt the model really likes to pimp itself at the end of every sentence with something like: "If there is anything else I can do to help, let me know!" and it's so annoying ;D
- vanderZwan 3y agoSadly, "women as subservient AI" is a pervasive sci-fi trope. You can probably fill in the blanks on its sexist history (it's all fairly predictable), but there's tons of feminist writing on the subject if you're interested.
- moffkalast 3y agoYes this is pretty much the basic ol' "hey AI, pretend to be this person". It's somewhat sufficient, but eventually the initial instructions will drift out of context and it'll only have the conversation so far to deduce how it should act, and it'll eventually break character. Should be more interesting when proper fine tuning is done for specific personalities besides the "obedient corporate drone" that OpenAI likes to tune for.