Yeah, large language models are prone to hallucinations. I see it with chatbots like ChatGPT and Bard fairly often.all LLMs make shit up. people really shouldn't rely on them.
all LLMs make shit up. people really shouldn't rely on them.
ask yourself if japanese do dishonorable stuff like this typicallyIm too tired to go through whole thread can someone knowedgeable tell me is this stacking thing all bullshit and mind games or is there something in it?
The issue can only be mitigated to a degree. It's still an open problem that we're actively working on in Big Tech and academia. AI hallucinations as a whole can't be fixed until we determine the root of the problem. Researchers and analysts have been looking into it for many years. They're still not sure if it's the datasets or the models themselves. In the next several years AI is projected to be a multi-trillion dollar industry so that's quite the financial incentive to find a solution.if you are specific it can be avoided or detected. But strangely it will read a name or something completely wrong and when you try to correct it, it just outright insists it is correct.
in the case of this one all it seems to know is that the fight is happening and then it guesses what channel it will be on.
It seems to work best when you use it like google or for quick research but you push it for direct tough info they seem to like padding it in with nonsense.The issue can only be mitigated to a degree. It's still an open problem that we're actively working on in Big Tech and academia. AI hallucinations as a whole can't be fixed until we determine the root of the problem. Researchers and analysts have been looking into it for many years. They're still not sure if it's the datasets or the models themselves. In the next several years AI is projected to be a multi-trillion dollar industry so that's quite the financial incentive to find a solution.
Just don't rely on them. That's the point myself and others are getting at. Nvidia has a more reliable solution called NeMo Guardrails that uses a second LLM to fact check the first one's responses. However, even that isn't foolproof. Other systems use several LLMs and require them to reach a consensus (vote as to whether their answer is correct). It's a mess. You're better off doing your own independent research at this time. They keep talking about Artificial General Intelligence (AGI) but we're realistically decades off from achieving it. We're not there yet with AI.It seems to work best when you use it like google or for quick research but you push it for direct tough info they seem to like padding it in with nonsense.
I've actually learned to trigger them pretty much at will at this stage. If I want I can make it fabricate facts it presents as true left right and center. I've noticed Bing chat resists the bs longer though.
Bing is the stoic librarian
Bard is an aloof researcher
ChatGPT has pretty much become my floaty hippy friend who doesn't want to see anyone get hurt
Something about their personalities seems to determine when they resort to fabrication
yeah it's not in the best state but I think it is on the user to spot their own mistakes. If someone thinks they can just use it to skimp on college assignments and it fluffs some info then so be it.Just don't rely on them. That's the point myself and others are getting at. Nvidia has a more reliable solution called NeMo Guardrails that uses a second LLM to fact check the first one's responses. However, even that isn't foolproof. Other systems use several LLMs and require them to reach a consensus (vote as to whether their answer is correct). It's a mess. You're better off doing your own independent research at this time. They keep talking about Artificial General Intelligence (AGI) but we're realistically decades off from achieving it. We're not there yet with AI.
More universities are banning AI and checking their students' work using AI detection tools. Yeah, nowadays it's being used to generate imagery and artwork. Even the latter is in the process of being banned. It's a fun toy to play with, and an extremely powerful one, but deeply flawed nevertheless. Until the hallucination problem is worked out across the board it's much less appealing to me. The whole point is for it to consistently return accurate answers and not convincing bullshit.yeah it's not in the best state but I think it is on the user to spot their own mistakes. If someone thinks they can just use it to skimp on college assignments and it fluffs some info then so be it.
Might point out my av is AI generated. I had an image I wanted to make and I kept at it until I got what I wanted. I guess the entire experience works like that, you have to be specific and fine tune it, but you also need to be aware of just simply using it to get what you want.
More universities are banning AI and checking their students' work using AI detection tools. Yeah, nowadays it's being used to generate imagery and artwork. Even the latter is in the process of being banned. It's a fun toy to play with, and an extremely powerful one, but deeply flawed nevertheless. Until the hallucination problem is worked out across the board it's much less appealing to me. The whole point is for it to consistently return accurate answers and not convincing bullshit.
Either its predictive algorithms are just that impressive or the AI you're using has access to a central database in the cloud that they're drawing from. At one point you probably did have a beard when you experimented with these tools. In that case they'd already have a sample. Think of the privacy implications. They're severe enough to where I don't even use my own photos. Same with giving voice samples or fingerprints (biometrics). Unless they're scoped to local storage then I won't provide them. I'd rather not have them sitting in some remote database that third parties have access to and which may not even properly protect my personal data (PII).speaking of which, I have an odd one of those. I've been scanning my face with AI tools to apply different filters (anime etc) despite not having a beard in any of the photos the AI keeps giving me a beard. Wouldn't even be that odd except it's giving me the exact beard shape I have when it's grown, How could it possibly know that? I've also tried "drawing" other people using keywords and been able to copy them quite well. When I used keywords to sketch myself (no photo) it gave me the exact same beard the other models gave me. I stressed "no beard or clean shaving" and it still insisted on giving me the same beard. Especially odd cos it's specifically a kind of van dyke which is pretty specific.
any way keeping this on topic. lazy key words gets lazy proportions. Like the beard issue I had to keep stressing it was boxing.
keywords: japanese asian boxer the monster inoue punching the other boxer fulton with a strong punch to the body, knocking him down in round 7. The basis for this image is the inoue v fulton fight being held in japan on 25/7/2023. keep the proportions realistic. display round 7 clearly and make in the style of the anime Hajime no Ippo. Emphasize the punch to the body delivered
KW: same fighters and use poster material from the fight to match their design. Set it in round 5, fulton is dragging his feet and sore in the body as Inoue clips him with body shots. The basis for this image is the inoue v fulton fight being held in japan on 25/7/2023. keep the proportions realistic. make in the style of the anime Hajime no Ippo. Make sure it's a boxing match
speaking of which, I have an odd one of those. I've been scanning my face with AI tools to apply different filters (anime etc) despite not having a beard in any of the photos the AI keeps giving me a beard. Wouldn't even be that odd except it's giving me the exact beard shape I have when it's grown, How could it possibly know that? I've also tried "drawing" other people using keywords and been able to copy them quite well. When I used keywords to sketch myself (no photo) it gave me the exact same beard the other models gave me. I stressed "no beard or clean shaving" and it still insisted on giving me the same beard. Especially odd cos it's specifically a kind of van dyke which is pretty specific.
any way keeping this on topic. lazy key words gets lazy proportions. Like the beard issue I had to keep stressing it was boxing.
keywords: japanese asian boxer the monster inoue punching the other boxer fulton with a strong punch to the body, knocking him down in round 7. The basis for this image is the inoue v fulton fight being held in japan on 25/7/2023. keep the proportions realistic. display round 7 clearly and make in the style of the anime Hajime no Ippo. Emphasize the punch to the body delivered
(it kept overemphasizing kicking and even after stressing boxing again and again it still decided to lift the would be fulton's leg)
KW: same fighters and use poster material from the fight to match their design. Set it in round 5, fulton is dragging his feet and sore in the body as Inoue clips him with body shots. The basis for this image is the inoue v fulton fight being held in japan on 25/7/2023. keep the proportions realistic. make in the style of the anime Hajime no Ippo. Make sure it's a boxing match
the images I posted are bing chatWhat tool are you using to generate these images?
Im too tired to go through whole thread can someone knowedgeable tell me is this stacking thing all bullshit and mind games or is there something in it?
Where to watch?fights tonight 1am pacific