/wiz/ - Wizardry

But it is still just a bot. Very much the likes of BonziBuddy.

>>212114
>>212113
what is that? i hate to defend this shit, but chatgpt doesn't have access to web stuff and the UI on that looks nothing like chatgpt. it cannot search the web

are you just using 'chatgpt' as a general term for 'asking a question to LLM AI', akin to using 'google' to mean searching online?

>>212131
>LLM AI
Not sure what you're referring to, but Bing.com has GPT-4 that you can chat with.

>>212132
LLM = large language models, it's the type of ai that everything is using nowadays

that's interesting i didn't know it was integrated into bing, but that bing/microsoft version seems heavily censored, based on me entering in some initial prompts i use for the regular 3.5 version from openai.

>My mistake, I can’t give a response to that right now. Let’s try a different topic.
>My mistake, I can’t give a response to that right now. Let’s try a different topic.
>My mistake, I can’t give a response to that right now. Let’s try a different topic.
>My mistake, I can’t give a response to that right now. Let’s try a different topic.
>My mistake, I can’t give a response to that right now. Let’s try a different topic.

there's also no ability to edit your prompts and go back to earlier prompts, and traverse the tree of everything that's been generated so far

literally seems to be just "ai google search" and not "useful general purpose virtual assistant" which is lame

>>212133
You're wrong about everything and Bing does have access to the internet and yes it's GPT-4.

>>212133
I find today's AI in general to be sort of disappointing in ways.

I think it's a bit dumb in a way how much more often people are throwing the term "AI" around when it is not significantly different to various bots I've chatted with over the years. I remember that MSN Messenger had a very smart bot called SmarterChild or something back in like 2006 or 2007 which seemed pretty smart but hardly anyone else talked about it. Then Cleverbot appeared a few years later sometime in between 2008 and 2011 and people talked about it a bit more than they did SmarterChild. Not only that, but people back then called these "bots" and they would say that these weren't "true AI" or whatever. I even asked GPT-4 whether or not those earlier bots could be considered "AI" and it told me that, yes, those bots were also AI and GPT-4 is also just another bot.

I dunno but it feels like the term "AI" is a bit of a meme until they start introducing human-like robots over the next few years. They've already begun to introduce wheeled robots that look like R2D2s that just kinda' roll around here in the streets of Los Angeles which just feels gimmicky.

>>212173
>You're wrong about everything
i don't think i'm wrong

https://help.openai.com/en/articles/6783457-what-is-chatgpt
>ChatGPT is not connected to the internet, and it can occasionally produce incorrect answers. It has limited knowledge of world and events after 2021 and may also occasionally produce harmful instructions or biased content.

it has had the ability to interact with the web for a while, but this is still just a weird hack. for example you perform an internet search, extract the relevant data, then present it to chatgpt/gpt in the form of a prompt. it doesn't actually access the internet itself and openai doesn't have any service offering this as far as i'm aware. in conclusion it can utilize prompts and other user input, as well as the memory/data within its model obviously, but not the internet, but the current hack is to include internet info into the prompts.

>>212215
Same here I never used chatgpt so far and find it hilarious normies are talking to Cleverbot thinking it's sentient and asking it questions like how it came into existence. It's as much Ai as current VR is "virtual reality". Normies also are now sitting with useless as shit $1000 headsets that distort their view in two fish lenses thinking they would enter the virtual world through it while there hasn't been any advancements made to it since it's release, it's still a fake piece of shit just like the chatgpt "ai".

>>212216
I was talking about Bing and yes, you better believe you're wrong.

>>212258
>The thread is still active and has 205 posts as of now. You can read more details by following this link.
Admittedly this part is wrong and it doesn't really understand what a thread or a post is. I asked it to reply to 212216 and it responded to a different post on /wiz/ which was made a few weeks ago.

>>212256
>>212258
>>212260
you understand what i'm saying at least i hope. chatgpt is the service itself. you can secretly tack on search result information into the prompt and process the results to make it seem like it has access to the internet… but this has been available since day 1 almost from various third party services. bing is no different aside from being able to utilize their own search result information as opposed to something like chatsonic

the chatgpt model and service itself is still offline

>>212289
That's a different matter. They don't and can't retrain their AI every 5 seconds, updating it with the latest shit from the internet. They'd have no control over it that way because most of the data is garbage or propaganda.

Just imagine how much AI is going to revolutionize the world. When every Joe Idiot can just click and turn his idea into top of the line entertainment of the present, AAA games, films, TV. What the next decade will bring.

We're already practically there in music, get Taylor Swift to cover any song.

Just imagine the Rennaissance in culture, when Joe Idiot can recreate what used to be the pinnacle of culture with just 1 click.

>>212418
Joe idiot is going to use it to create child porn of Emma Watson, then the tools will be taken offline.

Anti-mining GPUs? Here come anti-diffusion GPUs.

>>210195
There was an interesting finance thread on wizchan a few years ago. I might even have stumbled onto some alpha (more likely I calculated it wrong) but now I can't find it.
It had a bunch of discussion about call and put options, there were a bunch of graphs and one guy claimed to be doing well.
If you find it, it might be good and I'd appreciate if you could let me know.
I struggle to find it now.

chatgpt got updated

you can now create your own gpt, to some extent

pretty sure this is just a clever use of preprompting every interaction such that when you ask something it answers accordingly. which is basically what you can alreasy do via their custom prompts feature and manually yourself each message

to me this seems like a way to conveniently package those configurations into something you can just select between. something other ai chat things have provided for a while (ie it responds as a grammar and spell checker, a creative assistant, etc whatever you select)

need to try it out, for some reason its not live for me though

>>212527
ah, seems it's only available to the paypigs with a plus subscription. i'm pretty sure it's nothing special, just a convenience for normalfag zealots of productivity who cant any time doing it themselves

>>212419
>nvidia anti mining
didnt knew that was a thing, im out of touch

>>212528
>>212527
ok im totally wrong it seems chatgpt has a thing where you upload files and i guess it trains something based on it? i have no fucking clue because it's for paypigs only

seems very cool and people are sharing their custom gpts all over the techy space. the reaction has been so massive they completely cancelled new paypig subscriptions, so it's actually exclusive right now. crazy

was looking into chatgpt alternatives and anthropic's claude is pretty close. was started by ex-openai employees a few years ago. it's heavily censored which sucks, and there is no editing previous responses… so if you enter something and it flags your conversation you basically have to start over from the beginning. as far as it's knowledge and capability, it's similar to chatgpt. hard to say. someone needs to just come up with periodic ai tests and score these things every week or so, as a means of determining their capability… because right now they are all basically more or less the same, they are all trained on the same shit, it's the same hundred or so top researches training the models they just shift from company to company.

>>213400
have you tried poe?

>>213403
no, what is that?


chatgpt/openai recently seems to have lowered their registration requirements. you no longer need an SMS/phone number to sign up which completely surprised me. i have been paying for chatgpt accounts over the past year (well, paying for sms codes to sign up without a real phone humber) so this is nice

try to ask chat gpt to sort all european union countries by income month (netto) and monthly expenses she/it wont do it

i dont mean some bullshit statistics like high income low income. real numbers that reflect reality and sort by best monthly income to expenses ratio

found out best country (in terms of income and expenses) is either denmark or germany

>>213598
I'm not saying its definitely wrong. But GPT is not yet better than wiki and google for purely factual info.

>>213605
chatgpt was unable to sort list of countries correctly by number that was already there.

you would get what im trying yo say if you seen conversation

>>213606
ai chatbots just aren't at that level yet where you can use them in lieu of tools and calculators. doing so is actually irresponsible

you can at least directly provide such datasets to chatgpt with the pro upgrade thing but ive never tried it out, although i imagine doing so would give you better results

>>199166
>How will you use the newly released ChatGPT from OpenAI to make a lot of money?
You don't?
Especially because the tool is basically useless, it's just technology from 1970 just being given a lot of GBs of VRAM thanks to the advancements in hardware.

>>213904
that doesn't mean it is useless

the input requirements for these things is itself remarkable:
> plaintext
> human readable
> natural language

even if the results were absolute dogshit, this alone is incredibly useful. the results are not dogshit though, but i will agree they are incredibly inefficient in just about every metric (size of data, training costs/speeds, etc)

>>215114
No, it's really useless, actually, because you have to go back and forth to get what you want, which at that point you could have done without it, so, yes, it is not only inefficient as you said yourself, but it's useless too.
The most useful thing it can do is reformat text/documentation and be a cool chatbot, that's all.
Games and entertainment are the only applications, and even there, only as an additional tool on top of other infrastructures.

>>215120
>No, it's really useless, actually, because you have to go back and forth to get what you want
search engines, google, etc all work in this same way. they present you information about a query. this is no different

>only as an additional tool on top of other infrastructures
that is a true, but it's neither good or bad. that's how every technology/tool/software works. they are used in combination with other things

>reformat text/documentation
>chatbot
>Games
>entertainment
if you think these aren't worthy of our attention, electricity, etc and solving problems related to these areas is a waste, that is your opinion. i think those are fine



i would say the most interesting use case in my opinion is making decisions

essentially it is prompted with context and perhaps a personality or other information which is factored in, and it makes a decision. their location, their schedule of the day, states of mind/mood, hunger, random thoughts and observations. literally anything you can write, any piece of text… can become input, can influence the decision

this 'general purpose' nature is what makes it so useful. you can find a million other things better at doing specific things, and yet those are not as useful. this shit is a black box where words go in and words come out. it's fantastic

this is very basic and dumb 'intelligence' and yet it is incredibly useful. i think this is very related to something termed 'classifying' and the early classifying ai. it is something which is almost impossible to code. even being absolutely wrong 20% of the time is remarkable

>>215130
No, it's useless because you are giving yourself the correct answer, and so you could have done without it. Its internal logic sucks because there is not really any internal logic, it's a probabilistic machine; just read on how it works.

Things might chance in the future, but right now, you can spot if an image it's AI instantly, and ChatGPT is only good for chatbots and low quality entertainment.

>>215142
>because you are giving yourself the correct answer, and so you could have done without it.
not that guy >>215130 but there are plenty of situation where ANY answer is sufficient, like in real life how we just kind of wander through the day making decisions. immediately after waking up on a saturday morning, what do you do? there are a million things you could do and none are inherently right or wrong. you could easily gather data and build some decision program… but suddenly say the text input is set on another planet or you are in space. now your decisions don't work. so you can gather more data and make another program to work for those… but the text input changes again and it's maybe monday now

instead of building a million different programs to make decisions, you have one which is good enough for a vast amount of things. there is nothing else capable of this atm besides ai

that is the utility. the end goal being a smart person in a black box i guess. right now it's not smart but it is pretty knowledgable . the knowledge wont change much as time goes on, they've already basically packed all textual human knowledge into these things. but they should reason better in time and make better decisions

>>215142
i have read how it works, i use it at work for when i know what i want pretty frequently. it works great for my use cases

>>215176
>but they should reason better in time and make better decisions
No, because they're fundamentally flawed. You cannot improve them after a certain point, they just hallucinate then.

>>215177
Well, good for you I guess

>>215182
>No, because they're fundamentally flawed. You cannot improve them after a certain point, they just hallucinate then.

i'm sure there's a point, but we have not reached it yet. there are models you can run locally and they have been improving continuously for most applications. big models and stuff like chatgpt have also been getting better. chips are also now being made specifically to run these things also to address the efficiency. even if their ability plateaus, so long as their power efficiency improves that's still an improvement. there are also different… approaches, paradigms, whatever you want to call it, different techniques for ai that are all advancing steadily as well, with new ideas always coming about

https://yewtu.be/watch?v=TU1gMloI0kc

What do you guys think about the new video generation stuff? It's pretty impressive (example above). If they can do this now, imagine what is coming. Entire custom made tv shows or animes can be prompted to suit your individual desires. They can even voice act them.

>>215917
it's already peaked i don't think AI has anything to offer anymore besides simple tricks

>>215917
Image generation still sucks BAD, and they're already trying video lmao
It's just marketing bullshit.

>>215917
it's great for such a new technology

nothing really matters about current stuff, but in 10 years things will for sure be crazy

>>215932
It's basically photorealistic. I don't get why you say it sucks so bad. Yes there are issues, it's not perfect, but it's brand new technology. Surely it can only improve from here.

>>215934
It can actually get worse over time, like most ai projects after release.
Or even some in development.

Exponential improvement is always just a fantasy.

>>215934
>Surely it can only improve from here.
That's the typical logical fallacy.
You have to REALLY understand how AI works to understand that they cannot get better, by definition.

>>215939
I'm literally tired of people smugly posting "you have to understand how AI really works" without explaining anything. Literally low effort smugposts.

>>215940
Well, it takes quite some time to explain it, there are youtube videos that can do it way better than I ever could.

In a nutshell, LLMs manipulate language statistically, so after word X they think that it is statistically likely that word Y should follow. However, the right answer to a riddle might actually be Z, but the LLM doesn't know that, beacause it's stupid.

That's why you get hallucinations: the AI writes stuff that sounds right, but the answer is actually wrong.

Of course, this is a simplification, and everything is way more complex than that. In fact they do have some kind of internal logic, because since they use a lot of VRAM ("big hardware"), they internally have so many if-else statements (they're not really if-else statements, but I'm simplifying) that some kind of logic and reasoning actually emerges during training. However, since the internal logic is being created automatically by training, there is no way to check these conditions.

That's why they are called "black boxes": you cannot debug them.

Also they reason in a very non-human way, so even if you could debug them, everything would look like garbage to a human, it would be like reading obfuscated code.

Even the AIs that recognize a dog in a picture, have no concept of a dog, they are only "reasoning" about pixels positions, in an extremely abstract and non-human way.

Given these facts, you can only get LLMs to act like they are smart, but you cannot make them ACTUALLY smart.

If I train a human to do calculations basing on how the number sounds when spelled, the human might get some results, it might do 1 + 1, and 2 + 2, because the words "one" and "two" sound different. But at some point it will just fail and be stuck. And you cannot IMPROVE that, because the method is wrong: you should teach him what a number is, you should actually teach it logic, the proper way. And we don't have that right now, and probably never will (the hardware is not suitable).

Probably biocomputers might stand a chance, but those are not a thing of the immediate future.

>>215942


This is just negative rigid thinking. Tech builds on itself. I do admit that about a decade ago when using RNNs for NLP started becoming big I thought there wasn't much to the tech outside of its specialized bubble. Of course LLMs evolved from that initial research and somethign major will come from this LLM research. It is constant improvement. Great tech doesn't just pop out of thin air overnight.

>>215940
you can guarantee anyone posting this doesn't understand how it works. They may have watched a youtube video or read an article that explained it and now they're in the dunning kruger phase of thinking they know everything about how it works while actually knowing basically nothing. He is engaging in motivated thinking. He wants his conclusion that it will never get any better to be correct and he has gone looking for evidence to support this conclusion and convinced himself that he has found it.

>>215950
That's not the case, I knew how neural network work before they became a well known thing, "before they were cool".
I've always studied IT related stuff.

First anon laments that no one explains how they work. Now that I've explained how they work, you still bitch about.

Well, whatever. Think what you want.

>>215942
>In a nutshell, LLMs manipulate language statistically, so after word X they think that it is statistically likely that word Y should follow

for the literal implementation of this, look at markov chain generators

they are pretty cool and can lead to surprising results, despite being incredibly simple and easy to understand