What's a quick way to tell if you're talking to a GPT bot?
There are so goddamn many these days, I think some of them are even connected to voice synthesizers.
I.e., a REAL NPC.
Ape Out Shirt $21.68 |
What's a quick way to tell if you're talking to a GPT bot?
There are so goddamn many these days, I think some of them are even connected to voice synthesizers.
I.e., a REAL NPC.
Ape Out Shirt $21.68 |
>What's a quick way to tell if you're talking to a GPT bot?
the frog and wojak image
You only hate it because some other npc golem was told to say it was "raycissst".
NPC is a person who repeatedly follows a beaten path. Believe it or not, doing the opposite of what's popular does not make you unique, which can be demonstrated in your autistic mentioning of racism and NPC's, further profiling you as a generic counter-culture tard-monkey. Cope harder, incel Xddddddddddddddd.
GPT?
https://en.wikipedia.org/wiki/GPT-3
grammar mistakes
homosexual
Get them to interpret the contents of a pic related. Note that if there's a headline they can probably but not always read it. If there's a simple object it may be identifiable at least in the general by ML image recognition, but this can fail and end up revealing the bot on its own. This is likely a generic image recognition algo rather than a database built up against existing BOT memes but one can't be sure so OC may be best. Handwriting in pic related may also prove unreadable for AI.
Failing recognition works but interpretation of something as simple as 'x thing in graph go up faster than y thing' so long as that's not spelled out in text is likely to be beyond them the longest.
Another option is that it's not an AI but it's someone who isn't able or allowed to view a pic related, perhaps for hostile work environment reasons, which is still about as bad as a bot so no big deal.
There's a far simpler method: bots tend to go for the antagonistic response if it doesn't understand something.
It's how we know BOT is about 105% bots.
>bots tend to go for the antagonistic response if it doesn't understand something.
True. Low info/content high antagonism responses can be common when they don't have anything else. There are higher info bots that try to respond anyways but sometimes they get a simple fact embarrassingly and inhumanly wrong but can do nothing but continue on from that position no matter how they're called out on it, even though it's completely tangential from whatever they're trying to push.
Forgot to say that the high info attempt bots are rare. Either that or they got good enough to escape notice.
Nope. InstructGPT is too good. Even if it can't do that yet out of the box it's almost certainly a minor matter of training against that domain.
>Nope. InstructGPT is too good. Even if it can't do that yet out of the box it's almost certainly a minor matter of training against that domain.
I'll believe it when I see it. I don't think we're quite there yet, as far as I've seen.
>bots tend to go for the antagonistic
Since when? Every chat bot ive ever seen was just a yes-man. Care to show an example or two?
This is why twitter is absolutely fricked, btw. They trained here but there's no picrel there and the bots have full run of the place. Sometimes the tone taken in their writing becomes a bit obvious, that it's either from a family of bots, or it's a human shill strictly following a playbook that focuses on demoralizing their target trying to make them feel a certain way and frustrate and demoralize them, maybe sort of a more modern version of the psyops forum slide tactics.
Twitter is very heavily gamed by many different parties that employ bots and paid human shills. The parties include military psy-ops and political astroturfing firms, and come from many places internationally. I think the only way to slow them down is to have difficult captchas for every post, although this just makes human captcha solver a job title.
This is sort of the non-technical technical thing i do too.
If the reply doesn't seem like it even has any awareness that an image was attached twice in a row i write it off as botscripted fluffer for some agenda
i had shitload of oc memes
I was going to make a smartass comment about the SID 6581, but how its oscillators are set up and interact has also faded from my brain, frick.
that's because you're a brainlet larping
Ask it to solve a word problem that involves high school level algebra.
You'd get a false positive from me. I'm in my mid-40s, haven't cut code in 15 years, and that knowledge has faded.
There are a few ways you can test.
1. Most of the time, the "bot" you're dealing with is trying to get out of a debate or conversation. If someone who is not conversing in English asks you a question, you have likely hit on the first problem.
2. If the conversation is taking place in a language which is in common use in the country the person calling is from, then they are likely talking to a human.
3. If they are talking to you in a language which is extremely infrequently used or which is a specific language/country and you have no idea where the person is from, then it's most likely a bot.
4. I was going to add something about the "GPT-2" speech synthesis, but I see you've already addressed that point in your other post.
Hi. My name is Paul and have I got an exciting business opportunity for you!
I work at truth social, the only 100% true social media platform, and after some restructuring initiatives, spearheaded by the big man himself, we are excited to offer you personally the exciting opportunity to become a truth social custodian and assist our moderation team in tackling fake news and pornography posted on our wonderfully successful twitter alternative.
To take advantage of this opportunity all you have to do is donate $100 per month to The Boss (lol I can't name him for legal reasons) and commit to spend 40 hours per week monitoring content on truth social.
If you are a real patriot who is driven to Make America Great Again! Then contact me. PaultheJew@truth social.com.
Stay based and redpilled /b/rothers
>paying to be a janny
Now this is the right business model.
So based and redpilled, you barely notice it's fake.
The flabby c**ts hands are too big.
But you are a bot
>Frog image or basedjak
>Time wasting and stupid question
Pretty obvious