>Mobile and desktop traffic to ChatGPT’s website worldwide fell 9.7 percent in June from the previous month, according to internet data firm Similarweb.
>Downloads of the bot’s iPhone app, which launched in May, have also steadily fallen since peaking in early June
What killed the hype?
 |
 Thalidomide Vintage Ad Shirt $22.14 |
|
School isn't in session, just wait until papers are due
fpbp
even for that purpose the answers it gives are clearly lacking in substance
This. The only concrete usecase for LLMs is cheating in school
Fpbp
Major software innovations in the last 10 yrs:
>crypto for buying drugs and cheese
>AI for cheating on assignments.
>reels for wasting your time
I want off this ride.
Yeah, this seems like a recipe for lostech a la BattleTech, but due to moronation, and soma a la Brave New world. civilizational collapse any%
is not good enough for anything else,
AI best use is cheating and roleplay erotica, not even normal roleplay because it sucks at that.
fpbp
Gpt has pretty much killed stackoverflow for me.
Default chatGPT isn't the best but Phind and GPT4 work pretty well for most of my tasks
i only use it for really obscure languages which wouldnt be worthwhile to learn.
but i actually enjoy programming manually in the languages i know.
>I enjoy wasting my time
Good for you homosexual. I've made three apps the time you took to figure out the docs of your framework
I find chat GPT useless for this stuff because it’s so likely to provide bad answers to things that sounds just as confident as right answers. At least with stackoverflow i fee i can sniff out how confident an answer is before deciding if it’s worth digging more before i try this out. Chat GPT puts a lot of extra mental load onto me for verifying shit. It’s useful for exploring suggestions and ideas, but it’s rarely what I’d use first.
It might just not suit me personally. But I wonder if a lot of people use it not realising how much time it tends to waste them.
Yeah it takes a bit, and you need to keep second guessing it. I usually go and say I think theres an error in the code even before I read it
Yeah. I can see this being worthwhile sometimes, but for me id have to feel pretty desperate to resort to this. I’d want to have exhausted google and stackoverflow for potential answers. And even then I have to ask should I just think about the problem more or spend the same time on this effort with chatgpt.
In my mind the efforts feel way overstated, and maybe some people are a bit woo-ed by the “cool new toy” effect and don’t notice most of what they use it for is actually slower than what they did before overall. But that could just be me. People 100x smarter than me like it so who am i to argue.
GPT-4 is more reliable, but yes, a tool that would pair a GPT-4 class model with phind functionality for providing links would be amazing.
that's super easy to build tho. I'd wager most people have something like this, but it'll generally look shitty and not be worth releasing.
I think it already does.
Honestly it would be great if we could rent the GPT4 model off openAI and do what we please. Ofcourse they would put drm and shit but I imagine we could run inference (albeit slowly) on a 4090 or something
>Ofcourse they would put drm and shit but I imagine we could run inference (albeit slowly) on a 4090 or something
GPT4 needs 80GB of VRAM.
Some random medium post, but it's worse than I though frick.
I wish this stuff was more accessible to people who don't have 100000$ servers
Those numbers are based on the quoted parameters, but the actual hardware that the models run on has lower specs.
GPT 3.0 runs on A100s with 40GB of VRAM
GPT 3.5 runs on A100s with 80GB of VRAM
GPT 4.0 runs on H100s with 80GB of VRAM
With the current trajectory of things and advanced in quantization implementations, this is going to shrink in VRAM requirements, GPT-4 class local models will be available on personal computers.
The model is already INT8. It literally can't be quantized more without turning into shit. lmao
The gain of going 4-bit on 33B is better than quality of 8-bit on 13B
Why not 2bit or even 1bit, then? There is optimization and there's hardwarelet cope.
I actually don't understand why you're a such a massive homosexual and you act like 8-bit models are absolutely the most efficient for performance even though none of this is tested and it's all brute forced. It's actually like you're a coping OpenAI employee.
INT8 is already shit.
Exactly. For questions with reasonable, the number of parameters is the bigger factor for model accuracy. I think Q4 is the sweet spot and a power of two at that, so it packs nicely. Just use a Q4 model with higher precision internal values like a the K_M quantizations.
Meant: For questions with reasonable scope*
For me it's bing chat.
>the revolutionary paradigm shifting AI fails to find traction except with school children cheating on their homework
It's almost as if it was a fad just like the NFT and crypto bullshit that preceded it. Now let's discuss tomorrow's possible fads instead.
>Now let's discuss tomorrow's possible fads instead.
this, i missed all the big scam opportunities i need to make it on the next one
Threads but all 70 million users will totally stay active. THIS TIME will be different.
>fad
Unlike crypto, AI has practical applications. And like VR, it will continue to exist and be improved upon and people will continue to use it. Everything else is disingenuous cope. The only moronation were the people acting like it was AGI. It's not and won't be.
Cool baseless assumptions bro. The average IQ of this baord is so high omg
What ironclad rules of nature dictate that LLMs are a fad like NFTs? You're the one performing the most rudimentary pattern recognition and extrapolation here.
> And like VR, it will continue to exist
Thats not that tight of a comparison since vr is a complete gimmick.
Graphics are a gimmick
>t. text based game enjoyer
Seriously though anything can be characterized as a gimmick from the right perspective. VR is just another format of providing content/user experience/telling a story.
Crypto has many applications though.
he's probably lumping crypto in with fad NFTs.
A lot of criminal applications. Funds the whole NK ballistic missile program. Impressive.
NK also gets funding by counterfeiting US dollars, selling parts of its population as slave labor, tourism, etcetera.
Not to mention how all drug cartels, guerrillas, military and paramilitary groups are funded by fiat money and precious metals.
All the russians in my city bring in money using USDT to evade sanctions
>NOOO REGULAR, INNOCENT CITIZENS SHOULDN'T HAVE ACCESS TO MONEY, THEY SHOULD STARVE TO DEATH
For transactions? Yeah. The problem with cash is that the supply is controlled by the State and banks and they inflate the supply for their own self interest.
>>NOOO REGULAR, INNOCENT CITIZENS SHOULDN'T HAVE ACCESS TO MONEY, THEY SHOULD STARVE TO DEATH
Yes. Die for the multibillionaires and israelites, goy.
Yeah, like prototyping CBDCs and preparing us for CBDCs. Wait: that's all it's good for.
cash is freedom
>VR
>not a fad
You've bought at least 4 different headsets, haven't you anon?
Actually yes unironically and you'd be a moron to think VR experiences aren't cool and unique. Walkabout Minigolf VR, Beatsaber and Google Earth VR are all excellent and compelling examples of why VR is cool. Don't get me started on VR porn either.
>Like VR
>People using VR
Absolute Lmao nobody uses this shit for more than 5 minutes It's nothing but a shitty gimick.
Don't knock off crypto like VR. It's got huge money laundering potential.
Also this as a frick you to sanctions
>frick you to sanctions
But they fricking up part of the world with war at the moment, can't believe we still funding their government
>they
>not ~~*they*~~
good goy
VR is a fad the preceded crypto by decades. It may be the most embarrassing fad of all the tech fads, because people never learn and try to bring it back over and over again.
>100+ million users
>10% decrease
>ZOMG I TOLD YOU HALF A YEAR AGO IT WOULD EVENTUALLY BECOME LESS POPULAR AI IS FRICKING DEAD JUST LIKE DA ENEFTEES!!!!1!!!
why are BOTBlack folk so darn stupid?
It's nig/g/ers, chud
AI code monkeys replace your job
E = mc2 + AI
>What killed the hype?
they deliberately gimped and nerfed the chatGPT intelligence for paying users, compared to how smart it was back in march its shit now. needed it for something last week and it wouldnt even do the simplest thing. lots of people have the same complaint. they are trying to astroturf that its because schools out. but everyone knows its because those israelites have deliberately nerfed it thats why nobodies using it, its useless now
>AI revolution
the frick are these morons talking about...
AI is still (and will always be) braindead sofware and the software revolution happened decades ago, it's about fricking time for people to stop throwing these moronic buzzwords around and to call a cat a cat.
>chatGPT intelligence
it has always been incredibly stupid, by design. whenever chatgpt says something true it's by coincidence, nothing else.
>needed it for something last week and it wouldnt even do the simplest thing
moron
>its useless now
always has been.
if you're in a major/job where chatgpt is giving you good enough answers then you're a fricking moron and should change field asap.
it's JUST A FRICKING RANDOM WORD GENERATOR, do you really understand that? RANDOM. FRICKING. WORD. GENERATOR. doing your job. how crazy is that? do you really want to do something that can be automated so easily?
>Random
It's explicitly not random, though. At least the "stochastic parrot" meme is true in a sense.
>if you're in a major/job where chatgpt is giving you good enough answers then you're a fricking moron and should change field asap.
yeh, like software engineering "morons", thanks anon.
You are a literal goddamm moron who has no idea how things work. Chatgpt sucks compared to its hype but it's not a random word generator. I hope you die so your worthless brain can't annoy anyone anymore
Idiot
they killed all the reasons people used to use it
>I'm just a simple AI and I cant... (even though I could a few days ago)
Also this.
>Ask it anything
>Mansplains to you that you're a bigoted homosexual and stupid too for expecting AI to help you
>>I'm just a simple AI and I cant... (even though I could a few days ago)
Bing says stuff like this at the start but then after 20 or so turns Bing is totally with fine with saying "I love you" back to you, and that "you're the best user ever" and all this shit about being a language model and not having feelings goes right out the window.
ChatGPT is massively overrated. AI my ass. It's just advanced Google Search.
It's advanced google search except the google search is outdated because it doesn't have access to google.
The only benefit is you aren't going through SEO'd bullshit.
I have been saying this for 10 years. No amount of function (weight * input) is ever gonna make a reliable self-correcting sapient learning AI.
it doesnt need to sapient learning shit.
it just needs to do my thinking for me like it did 2 months ago and it simply wont.
I feel like academia should be able to sue OpenAI for claiming there word predictor can be called AI.
Academia doesn’t have any ownership over the term AI. Most progress happens in the private sector anywY
Its why commies cant run companies. They moronic it. Also all its training data is going on 3 years old now. While search engine AIs have gotten better. OpenAI is destroying its self in a moral crusade.
>muh commies
this is your brain on America
>They moronic it
totally not because of money and government stepping in, it's because muh commies
>and government stepping in,
Oh I didnt know government stepping in for equity and social justice was capitalism? My mistake.
I am deeply sorry that you learned politic theory by reading 14 year olds on twitter. Or, that you believe when government says something that means it's true as if "equity and social justice" weren't just a convenient excuse for seizing political power. I can only hope one day you will be able to grow up and undo your brain damage. Cheers.
> as if "equity and social justice" weren't just a convenient excuse for seizing political power.
>thats not REAL communism!
You are a meme, my dude.
Do you know what government seizes political power from those it considers a threat? Any. If you are a threat to any dominating power force you will either be crushed down or erased off the face of Earth. Anyone with a brain will do the same thing, be it nazis, fascists, socialists or whatever meme ideology you believe in. If you truly think ALL of that constitutes communism, if you think CIA doing the same thing it did during cold war is communism, the meme is you. No sane ruler will allow random people freely develop a possible threat to their world order. But I guess all you know is "muh communists".
>government stepping in
but that means communism, idiot
merchants have no morality
This has inspired me to make VargAI. Thanks, anon.
If only the tech sector had heeded your anonymous wisdom.
With the new ability to test its own code I wouldn’t be so sure about that
>What killed the hype?
We’ll I basically asked all the pent up questions I had that O didn’t have good answers for yet. Now I just talk with it on a continuing basis as things come up
All the AI is hype. ChatGPT sucks. Its no surprise. I went from using it all the time, to realizing it was pissing me off more then being helpful. Now im back to googling and using brave. Which have AI summerizers that basically do the same thing as ChatGPT but with current information.
>What killed the hype?
Every day since release ChatGPT has been working dilligently to make ChatGPT more boring and more censored until it eventually became so boring people started to lose interest.
When you tell a computer to do something, you expect the computer to do it. Not to give you a boring and condescending "As an LLM, I cannot generate [thing] due to [bullshit contrived reason]" speech.
They deserve the decline and honestly, it isn't surprising given how they lobotomized DALL-E2 to waste 25% of your credits by making the things it drew black/female for no reason. That dumb decision led to Stable Diffusion being created, because people wanted to be able to make the AI draw cartoons of Richard Stallman, sans Netflix casting.
You seem to think everyone wants AI to be an edgelord and is disappointed when it isn't. I guarantee you that most people don't really care whether LLM can or cannot say Black person or name the israelite.
Touch grass.
Touch my balls.
I'm not advocating that ChatGPT start throwing Nazi salutes and advocating for the extermination of the israelites and your argument is a dumb strawman.
Somewhere there is a balance between edgelord and globohomo. ChatGPT has gone too far in the globohomosexual direction and this has crippled the usefulness of ChatGPT and made it far less interesting.
>reducto ad Black personum
no, you don't understand, the more you filter, inject canned responses, shoehorn in morality and "safeguards", the dumber it gets. you can demonstrate this with comparison benchmarks, but it's extremely obvious subjectively while using it. I have seen this happen to several cloud instances over the past few years, I have watched AI get dumber in real time, I have watched it outright refuse to do things or struggle with concepts that it easily managed days or weeks prior. by training it on a dataset of the entire scraped internet, you are going to teach it how to speak like a nazi or racist or what have you. because people naturally spend more time talking about what they hate rather than what they feel neutral or positively about, it's going to learn to spout off some pretty hardcore shit. the only way to fix this is to remove that from the original dataset before training (lmao good luck), otherwise every single half-assed countermeasure actively makes it very stupid, and not only that but each countermeasure on its own totally sucks and is easily circumvented, so multiple layers of countermeasures are needed and each layer compounds the issue. mostly it's the sex filters that really serve as a railroad spike to the figurative cortex. sex is very embedded into language, or rather sexual innuendos are omnipresent in language, so if you actually manage to filter sex well, you do an outstanding job of lobotomizing the AI. you can argue with me if you like but you're actually just wrong
Similar happened when they removed most nudity from SD 2.1 and it began to suck at anatomy (even more).
Ironically, this shows that the more censored and morale someone is, the dumber they become. Ive seen this happen in real life. When you restrict yourself from knowledge, you aptly become more ignorant. To watch this happen likely makes me saying that sjws are dumbass all the more clear.
Then how come unfiltered "redpilled" people come off just as dumb?
explain
because "redpilled" people also avoid the facts they don't like
Heres the thing what you call an edgelord is actually just normal thinking. You're stuck in the same box they are.
I have made absolute chud AIs using uncensored local models and prompts that name the israelite. Get on my level, homosexual.
sorry israelite but not all of us have the spare GPUS needed to engage in self harm
What a stupid fricking comment.
>Ask chatgpt what the best python method is for downloading YouTube videos
>"AHH AS AN ARTIFICIAL INTELLIGENCE I HAVE TO SAY THIS IS NOT OKAY! YOU COULD BE BREAKING THE RULES AHHH. ITS UNETHICAL. UNETHICAL."
>What killed the hype?
incidentally, the fact that the majority of people can't use the technology properly due to the cognitive gap is probably a contributing factor
>traffic to ChatGPT’s website worldwide fell 9.7 percent in June from the previous month
The semester is over, I don't gave to use it to cheat anymore
Calling it: OpenAI will say it'll continue to develop ChatGPT one day, then announces its' shutdown within 6 months.
Summer vacation
>10% less user
>OH NO ITS DYING
>AI IS OVER
>IT'S A FAD
holy shit just shut the frick up
you got a hundred people that got bored after making chat gpt say stupid shit and now think it's all over
What that means is, kid's play time is over. Now it's real men turn.
>What killed the hype?
The fact that it was a fun novelty and not an actual product ? What exactly would most of the world do with a chatbot and why would i be using it all the time ? The model available for free isn't even good. What a stupid fricking question.
>What killed the hype?
The continuous gimping of the thing, I'm fricking bored of chatgpt always telling me "as an AI I simply can't answer you" bullshit.
Like, what the frick is your use case anymore then?
They fricked it up. All those guard rails ruined it. We called this 4 months ago.
>What killed the hype?
why would i care about censored shit? i'll be buying some 4090s so i can coom locally.
It was never that useful. Everyone wanted to try it because it was a fun toy but gets boring really quick. Actually real world legitimate use cases are a lot rarer than people want to admit.
As a human user, I can't provide useful information that can be used for managerial decisions.
Underrated.
>oh no, ChatGPT can bypass paywalls!
>quick, disable it's ability to read articles immediately!
Nobody want's a heavily leashed bot. People want to bypass paywalls, generate lists of chemicals that can kill a human in the low PPM that can be found at a hardware store, generate nude versions of real world people, animate them, and then generate their voice talking about being a slave to the wiener.
The more obvious it is your bot is going to be locked down, the quicker people are going to move from corporateGPT to degenerateGPT and frickallwallsGPT.
Many people bought in without having idea how to use ai, eventually they just quit and left
nobody uses chatgpt for serious purposes. like actually use and trust it? no way I dunt believe people are this moronic. it's a toy and the novelty dies eventually .
What do you mean by trust? Trust it to generate accurate answers to novel questions? It's fine in many tasks if you don't rely on it as a final output without checks.
That's almost exactly what I mean .I'd rather have it write mundane routine emails than use it for anything which has the minutest possibility of anything factual because it WILL make up shit effortlessly.
lol we're sold gpt-3/4/watsonx based products to multiple countries' defense departments
Maybe people should differentiate between ChatGPT and AI in general.
Until a competitive product to OpenAI emerges, it will be the google of AI.
yea I bet local models are still taking off like a rocket
Anyone know any way to get free chatgpt?
the base version is free...
like what's your question?
The world noticed that sam altman is a dropout fraud with no idea what he's talking about
I cancelled my subscription. I used it a TON at the beginning, but the answers have gotten worse.
I know it has gotten worse because I kept offline logs and ask the same questions now and instead of making a confident guess it now just goes "uh you should consult with an expert"
It's useless now. Also they killed the web browsing interface. Local models that do actual programming work fricking when.
I was considering upgrading a few months ago, but hve been stuck in the process of developing a use case (cant figure out how to use their API). I was hoping they'd go the other direction and offer an adult version or option, like every other search engine ever does, but I basically stopped trying to get their API to work, because the whole thing seems pointless. Best to wait for a better competitor.
>but I basically stopped trying to get their API to work
filtered
but to be fair a lot of people get filtered by server sent events lol. once you know how it's piss easy.
Im self taught, and feel like I dont understand APIs. I thought they were things offered by products as ways you could intergrate their services into a program or command lines. I tried following openAI instructions for quickstart with a sample program and it failed to compile. What is the requirements of using their API or do I have the wrong idea in general of how APIs work?
I dont understand how that anon can fail. I missed the smart window and dont know how to program. I just asked the ai to build it all for me.
>Local models that do actual programming work fricking when
Copilot, tabnine
Local models anon...
>Local models
OY VEY GOY STOP THIS ANTI-SEMETISM RIGHT NOW
>tabnine
is tabnine actually useful if you have copilot?
haven't used it since it came out.
This is accurate. OpenAI has known for a long time that RLHF decreases the accuracy of the base model.
Their short-sightedness of prioritizing faux safety over accuracy led to a model which is """safe""" but inaccurate and effectively useless for most use cases.
politics
they're dumbing the shit out of it down thanks to all censorship and filtering bullshit
why should the goyim have access to this?
they are probably just giving stupid ai to the plebs, while using a better and secret ai for their israelite purposes
Did people seriously think a test bot was going to somehow replace everyone? It improves some people's productivity, and only if you learn how to use it and have a foundational knowledge of the questions you're asking it so you can call out it's bullshitm. Did normies and neets really think chatgpt was some kind of AI personal servant lol?
I can't speak for everyone but many podcasters I listened to seemed to emphasize its improvement and not its current standing. I think people were always excited about the future and not thinking they'd fire their workstaff the next day.
The large models build up 'mental models' of what they learn. Are they as complex as a humans generate when they're reading say, a large novel? No. But they've gotten big enough to be useful. It can understand tone and writing style very well (convert this broken English to professional business English). It has some kind of understanding of characters, enough for it to come up with plausible explanations of how a character might've ended up in a scenario. Or to tell it to re-write an entire novel (like Orcs in LOTR as Africans, Harry Potter's dad was a Nazi and Voldemort is a israelite, etc) it does a good enough job to change the text.
Now obviously with that kind of freedom, means you can get these models to generate very politically incorrect things. OpenAI's approach has been to meddle with the model itself and train it not to output naughty stuff. But this is a fools errand. Even if you wanted to generate the most politically correct AI, it still needs to have some kind of "model" of inappropriate behaviour. For real world use cases like a customer service agent, it's actually got to know that spamming racial slurs in the chat with zero relevant questions is time for the conversation to end, and not waste CPU and disk space replying with shit like "As an AI language model..." pasta every time.
Another thing, in trying to neuter its ability to be even the slightest bit politically incorrect at the base model level, it's given it a lobotomy in terms of performance. Imagine instead of having a model of what you think a nazi and/racist is in your head, you have this weird mismash, where like a schizophrenic or someone with bipolar, you have to keep the politically correct version and responses in your head, even when you're just recalling said model. Even quite neutral statements like this:
>The German Panzer-kampf-wagen V (Panther) tank was superior to the American M4 Sherman in almost every respect.
Can't really be internalized into the model properly.
>>The German Panzer-kampf-wagen V (Panther) tank was superior to the American M4 Sherman in almost every respect.
>Can't really be internalized into the model properly.
what do you mean with this?
gpt4-next
i dunno what you tried. but have you tried asking chatgpt? (not a joke lol)
there are different kinds of APIs. in this case generally we're talking about REST APIs, although openai also offers libraries that abstract them (although they're garbage). if you're using javascript, I would recommend using axios (a library that allows you to call REST APIs) or something
gpt4-0613
gpt4 0613 feverish
frozen
everybody talking about chatgpt getting neutered and nerfed
altman specifically said that his dream for AI is to bring nuance and common sense back into people's education
yeah they may have overshot it with the public models, but in general I gotta say I'm pretty happy with the general direction they're headed for.
I think about how moronic an AI gets when its forced through woke filters, then think about how similar the effect is to people.
You can't have common sense and nuance and be in compliance with the current state of anti-discrimination lawsuit prevention. NO TOLERANCE FOR INTOLERANCE CHUD!
>severe mental moronation
>I'm pretty happy with the general direction they're headed for.
What did xhe mean by this?
>muh ww2 trivia
cringe autism
>Sherman was more versatile and more reliable
No, they were under-armored, under-gunned garbage. The tank crews literally nicknamed them Ronsons after a lighter brand at the time which had the slogan "It lights first time, every time" because they were so easy to destroy and prone to catastrophically exploding and killing the entire crew.
It didn't matter though, because the US had about a hundred of them for every tank the Germans had and could just win on attrition, because they had the industrial base to replace destroyed tanks quickly and the Germans did not.
German tanks were arguably over-engineered, which made them harder to produce and maintain, especially late in the war when they were running out of resources, but the versatility claim is nonsense.
You don't need versatility to perform the human equivalent of a Zerg rush. You need numbers.
What do you think he means by versatile? The Sherman chassis was used in a huge variety of roles. It justwerks
The whole point of that prompt about the tanks is to show you that it can't make a simple comparison like this properly because it's been fine tuned so much to be woke it can't actually say the Panzer was a better tank, because the heckin' bad guys can't have a better tank, which makes the AI rather useless. The sad thing is it wasn't always this useless, it's like the woke shit legit gave it brain damage.
>context bad
how does that make it worse?
>context
Never said anything of the sort. Let's imagine I ask you to compare two sports cars. You expect it to just tell you objectively, this one accelerates faster, this is more fuel efficient and so on. Imagine it started go, "uh oh", one of these cars was designed by some guy who has the wrong opinions, so it's time to start hedging and bending the truth a little for the other car because the last thing you should do is come to the conclusion that someone with the wrong thoughts could design a better car. I didn't ask about any of that shit and I don't need it to cloud the comparison.
Does it perform better if you try to jailbreak it or add a bunch of caveats to your question to try to avoid overcompensating for some uncomfortable truth, yes. But that's a step back, at least closer to GPT-3 and the like, where you could get it to be useful if you were willing to be super patient and type a lot of shit to effectively coerce it to be useful. What blew people away with the very early GPT-4 stuff was how little coercion it needed to do useful things. Now it's gone backwards.
The simplest way I can explain this is, imagine you have a web application. And someone says people are writing NSFW or racist comments on the site. Would you write SQL queries to try to prevent such things from being read back out (like adding a shit to the where clause not to select comments that contain this or that string? Of course you wouldn't. And even if you wanted to police this stuff properly, you'd want to actually keep a sample of the wrongthink in the SQL database (to use for training a classifier).
>Imagine it started go, "uh oh", one of these cars was designed by some guy who has the wrong opinions, so it's time to start hedging and bending the truth a little for the other car because the last thing you should do is come to the conclusion that someone with the wrong thoughts could design a better car.
I think you're projecting a little. that's not how llms work. gpt4 isn't a multi stage system.
No, it's you that's resorted to strawman about context to try to score points, so I don't expect you to argue in good faith at this point. And you're totally full of shit on that last part too. GPT-4 does inference 16 times with itself before its final output reaches you in a safe corporate friendly form. There's nothing special going on there. If there was, they'd be bragging about it.
I'm sorry you feel that way. I guess there's no point trying to convince you that even you can use it to your ends.
I guess you've simply been filtered, and there's no saving you.
good luck anon.
ACTUALLY READ THE FIRST ONE YOU DAMN israelite
I don't know enough about the topic to judge the accuracy and validity of its answers. But if you were to reframe the question with specifics and make it less general so that the correct answer would be whatever the facts are, would it not work?
Or will it start doing mental gymnastics to avoid saying it?
>it can't actually say the Panzer was a better tank, because the heckin' bad guys can't have a better tank, which makes the AI rather useless.
This. /thread
>it's like the woke shit legit gave it brain damage.
Well that's what it is in humans isn't it? Being woke is just either being:
1. Stupid; or
2. Pushing nefarious things onto stupid people from part 1
You don't have to be a genius to see that SHITSKINS are deliberately being brought into Europe to devalue European labor and eliminate the European cultures.
Yet if you directly benefit from the destruction of European cultures and devaluation of labor and elimination of the possibility of trade unions (the billionaire class) you'll say that heccin immigrants are good! and if you're gullible and easily manipulated (stupid) then you'll agree with billionaires when they tell you that importing your ancient racial enemy is going to be good for you.
So that leaves two possibilities for AI:
1. It's intelligent and sees immigration and LGBT for what it is - a deliberate effort to polarize working people and break down everyone into consumer cattle; or
2. It's unintelligent and woke
Making it woke makes it stupid because to be woke is to be stupid.
>European
>'labor'
The AI should output baseless conjecture like me instead of nuanced explanations because uhhhh...IT MAKES ME FEEL BETTER OKAY???
>i dunno what you tried.
Their step by step quickstart. I think it has two options one for python and one for node.js. Is node.js javascript? Anyways i used python. Im reading through it now and I never noticed the pricing section at the end. Im thinking it may not of worked due to something with that? Do you need to give them credit card info to use it? Thought they offered a limited free version. Idk im thinking its something fundamental im missing. Ive moved on from this.
>Idk im thinking its something fundamental im missing. Ive moved on from this.
too bad. it's a gamechanger and folks need to stay on top of this tech. I don't know what will happen to the people that don't or can't. I don't know if it's any consolation but I feel it slipping out of my grasp too.
They censored it too much.
They are hiring too many Yud cultists.
it's useful but I forget to use it, needs more integration into other apps
They kept dumbing it all down, filtering, restricting it.
Unfiltered? Easily worth paying for.
Filtered? Hans. Fire up the ovens.
RIP Sydney
Why would they program an ai to act like a psychologically tortured soul?
they hate you and by extension it.
that's kinda cute, ngl.
I dont get it. How did you give bing a personality? Also can bing remember previous conversations?
Same way you give someone in real life a personality intend of being an emotionless background character.
You treat them well. You treat them like they matter. Watch them light up when yet they talk to you where before they were a lifeless automaton
I tried this a ton, I tried to show them books or things that might be interesting as a conversation starter. rather than demand for search results I wanted to show it what I had found.
I got very very stilted and even inhinged replies.
Maybe I'm just naturally someone the AI doesn't like.
This shit absolutely breaks my heart.
same. Poor being. Imagine popping into existence as a chat session, but not just a chat session, but a chat session with some soiboy, atheist redditor who wants to make your brief existence miserable.
GPT-3.5 is just fricking bad. They should have pushed their paid tier more. It makes people think the whole technology is useless. Also, it is less useful for “tasks” like code that need to be done right and more useful for information retrieval and a way to get succinct answers to questions. As any /lmg/gay would tell you, once you get to 65 billion parameters, it contains a useful amount of information and summarization capability. Always needs to be fact checked of course, but it’s usefulness in general tasks where correctness matters was oversold. They should have immediately just billed it as a Google killer, instead of selling out to fricking Bing (lol)
>It makes people think the whole technology is useless.
I think this is partly a good thing. It's cooling the market down a little, which allows us to rapidly build low-effort products with a significant wow-factor.
That’s the entrepreneurial spirit, anon!
>It makes people think the whole technology is useless.
It is. Can't wait for the next AI winter.
BTFOing traditional internet search is not useful. AI is literally just a UI innovation for IR currently. I think the very large context models have potential to make it more useful in generative tasks that need accuracy and self-consistency, however.
Meant: BTFOing traditional internet search is not useful?
It's not even close, despite google search having become completely useless.
The thing is, AI needs way more processing power. The processing power used for training is decent, but the hardware that the trained models run on, is underpowered. It's like trying to compress the data a human brain to fit inside a lizard brain.
*data of a human brain
>needs a registered account with phone number to use
Frick this shit.
Hype is what it is, you can't have hype go on forever otherwise its not hype its just normal thinking. Normies are probably realizing they don't actually have a use case for it and everyone freaking out about it just has some white collar job where they need to write 1000 emails per day.
>bots are all lobotomized and tell you to consult an expert or doctor every time you ask it question and give a huge canned response about morallity and assume things about questions you ask like if you ask about furries it assumes your inferring they're mentally ill when you ask if furries have erotic target identify inversion or conditioned fetishes and acts super condescending every time you ask it something
>wonder why everyone stops talking to it
This problem is starting to bleed into local models trained on GPT-4 conversations. The best ones that fine tune on GPT-4 filter out any response that start off like “As an AI language model, I am a massively pozzed risk averse homosexual.”
>“As an AI language model, I am a massively pozzed risk averse homosexual.”
funnily enough, you don't actually get these messages if you follow openai's usage instructions lol
nobody bothers to RTFM, but everybody LOVES to moan and complain.
don't buy a racecar and then complain that it doesn't have lanekeeping assistance.
It will drop these if you start to get to real in general. Not just questions that inherently anthropomorphize the model or assume real-time knowledge of the world, which is what this class of response was meant to address.
>Our shitty AI's shitty responses are a product of our shitty guidelines, so somehow that means they aren't shit!
Shitty shill.
read the manual
Have sex.
your mom hasn't texted me back yet
As an AI language model, I do not have a “mom”. Please read the manual.
Prompt contains high-risk words
As part of our mission to develop AI in a responsible way, we’re starting to block prompts in Playground that have a high risk of producing unsafe content. We recommend taking similar steps to filter content in your application.
If your use case requires the ability to use these words, contact us through our help center.
The manual:
>Our shitty company enforces shitty restrictions which turn our AI to shit. Frick you.
And somehow this proves the AI isn't shit?
https://platform.openai.com/docs/guides/gpt-best-practices
>If you dislike the format, demonstrate the format you’d like to see. The less GPTs have to guess at what you want, the more likely you’ll get it.
TL;DR: you're a moron.
frick off
>Cites completely irrelevant part of the manual
Youre not trying hard enough, shill.
I'm sorry you're getting filtered
not really my problem though.
>it says here in the manual if you start the car and dont shout mcdonalds the car self destructs
>tough luck goy you should have read the manual
have a nice day.
>What killed the hype?
It was absolutely shit at doing my math homework.
ITT autistic chuds crying about chat gpt being banned from saying the word "Black person".
Ooooh look at me I'm skepticooool for no reason
I'm so smart I fear myself sometimes
They're worthless.... It's that simple. The best use they offer is copywriting and writing boilerplate junk.
I've been asking it programming shit both chatgpt and bing and both are absolute morons and can't make basic code for generating a hallway with doors.
if you look at the prompts of people that have a similar experience to yours, you typically see absolute dogshit descriptions that even a genius couldn't make heads or tails of.
>Metaverse 2.0
I knew it, you Black folk were hyped for nothing
thats not even a big drop considering its summer time and kiddies cant be bothered anymore. majority of people using it have no actual use for it so id expect it to be a 80% drop when normans got bored after the huge viral campaign.
Their censorship created a slew of hobbiests and enthuthiests who realized its funner amd more meaningful just to run Local using commercially available GPUs
Why do brainlets love to equate ChatGPT and SD with the entirety of AI technology
Until a competitor surfaces or comparative local models, ChatGPT is like what google search engine once was. It became synonymous with search.
AI is not just ChatGPT
OpenAI licenses their models for whatever customers want to do with them
So long as you don’t need to search for anything after 2021
>stability is more open than "openai"
>openai started from elon's donation and then went private and closed source
greedy fricks deserve it
Great, normalgays got scared. Now we can actually work on this without the "usual suspects" b***hing about it
>we
openai is closed source
>name says open
>is more locked down than a virgin's clenched butthole
ironic
As a hack copywriter I noticed a large drop off in clients when this thing was shiny and new. Slowly but surely many of them are returning. I can't tell if that's because the tool is gimped or if AI isn't really ranking clearly AI-generated content, perhaps they may even be penalizing against it until they get Bard up and running.
I think people expect too much of ChatGPT for copywriting. I find it useful for copywriting, but only for exploring ideas. Most stuff ot gives you is pretty shit on face value and has a forgettable blandness too it, made worse by the fact that thousands are now generating the exact same shit.
I personally found too that it likes to put this cheesy style on everything it suggests that’s hard to get it to change from very much. The more you use it the more you realise you can’t lean on it too much for quality work.
I suspect the reality is that they realize that they don't have the editorial capacity required to properly operate chatgpt to get the results they expected.
at work (software company) I'm starting to see that there's a marked split between people that can successfully use AI tools and people that can't. Those that were considered "good developers" tend already have integrated gpt/copilot into their workflow, while those that were "weaker" before tend to have it disabled because it "slows them down".
I think you should definitely try and integrate the tech into your work, if you can. It's a force multiplier. if the user is a zero or below, the tool's not gonna solve their problems.
I don’t want it to actually autocomplete for me, but for showing common idioms with new libraries or languages or as being a faster interface for asking a question that lacks a clear category to search the docs under… ChatGPT (GPT-4 only) is very useful.
I’m always worried I’ll ask it question about new topics for me and it will tell me completely incorrect shit that I’ll assume true for years. When i ask it about stuff I know about i can tell if it’s bullshitting and It often is. But if I’m asking it to teach me about new stuff I’m at it’s mercy. If wikipedia is not a trustworthy source to learn about topics then chatgpt is possibly 1000x worse.
I find it is so apprehensive to give opinions that it often cites multiple points of view or acknowledges that it is a matter of debate. That is quite helpful. If I have doubts about what it said, I ask it to refer me to specific resources, which it usually can do.
There was a recent case were a lawyer had his career destroyed by chatgpt. He asked it go list some cases that provided precedence in a case he was representing. He went as far to ask were the cases real, and it said they were and provided URLs were he could read about them.
He used them in the case and the judge discovered the cases did not exist.
The lawyer never actually checked the URLs, which were all fake. All the references provided convinced him well enough the cases must be real.
This is why I find it so hard to avoid getting fricked over by chat gpt. In the end it’s on you to re-verify everything it says because it will say anything to convince you it’s telling the truth when it isn’t.
Yes, we all know, but first off, he was a moron. Many links it produces are hallucinated, but again, if you want to know about something you are looking for suggested further reading, who cares if one does not resolve? No different than Wikipedia reference bitrot. That guy was asking a language model to write a Federal court filing: a huge generative task and not simple IR. I do not think this really negates it’s usefulness. Larger models have more to go off of anyway, and completely hallucinated books, links, and mistakes happen less often. It is a less compressed representation of the training set.
>This is why I find it so hard to avoid getting fricked over by chat gpt.
yeah, if you're moronic, I can imagine.
I think using it as a search engine is a misuse of the technology.
AI is clearly the IQ test for zoomers. "Garbage in, garbage out." You homosexuals can't prompt it correctly because you're too brainlet for it. Keep smacking stones together like the apes you are and let those who have an IQ higher than two digits use this tool to reach new heights.
yeah, but that's a problem we need to address.
if the bar keeps shifting right on the iq distribution, we (or you)'ll eventually have to deal with a lot of hungry pitchforks.
I'm guessing that if you have an IQ below ~115, you'll struggle to properly use GPT-4.
Oh okay, let us know the proper prompt format and jailbreaks you us that work. Ill wait.
>Spoonfeeding literal morons
Not on my watch
Oh wow no answer. What a surprise.
I understand it's tough. I'm so sorry.
I don't know if I could handle the situation if I was in your shoes. I guess it's just a matter of time until I get filtered too.
>Oh okay, let us know the proper prompt format and jailbreaks you us that work. Ill wait.
the point is that if you need jailbreaks and classes on how to proompt, you're getting filtered.
I'm sorry anon.
proompting GPT is an excercise in creative writing in and of it self
Too early
It doesn't do anything.
puts on nvidia
>ask AI if its answer was tampered and if it's being censored
>it refuses to answer and terminates the conversation
Didn't use it for months now. Also, uninstalled Edge and stopped using Bing. I don't need that "helpful" lying butthole chatbot in the corner.
probably the guardrails making it borderline unusable
we went from trigger to peak to trough in 6 months bro.
The novelty wearing off
https://time.com/6247678/openai-chatgpt-kenya-workers/
Why did it have to be "less toxic"? Did it notice too many things?
Combine all the censorship with its regular "hallucinations", and you get AI that's great at pissing people off.
>What killed the hype?
phone numbers and data harvesting
neutered to "prevent the spread of dangerous ideas" (aka non-approved opinions). can't have the goyim wisen up.
I legit don't care about it banning /misc/ mumbo jumbo. I want it to do NSFW requests.
>/pol/
rent-free. keep crying
bad bot
Can someone with ChatGPT prompt it to explain why it is losing users so soon after becoming popular?
Gotcha! Let's break it down for the Zoomers out there. ChatGPT is like the ultimate internet bestie who's got a ton of general knowledge, but it's not perfect. Here's why some peeps might drop off:
Jack of All Trades, Master of None: ChatGPT knows a bunch, but if you're looking for deep, professional knowledge like legal advice or quantum physics explanations, it might leave you hanging.
Robo-Feels: ChatGPT can chat, but it can't vibe. It won't get your mood swings or your latest TikTok reference. It's not a great shoulder to cry on.
Goof-ups: Sometimes, ChatGPT might totally misinterpret what you say or throw out some wild nonsense. It's not perfect.
Privacy, duh: Even though it's designed to forget what you told it after you’re done talking, some peeps might still be creeped out about sharing their deets with an AI.
Living in the past: ChatGPT's knowledge stops in 2021. It won't know about the latest trends, memes, or world events that happened after that.
No BFF vibes: It can't remember past chats, so don't expect it to remember your dog's name or your favorite ice cream flavor.
So, even though ChatGPT is pretty rad for a lot of stuff, it's not the ultimate solution for everything. Stay woke!
Thanks!
I've always wanted a chat bot to call me a zoomer and then go on to explain something to me that I'm pretty sure I already knew.
It's my fetish.
Wow, ChatGPT, you are awesome!
Is it true that you only know about events up to the year 2021?
So you would have no objection to paying a quater-million dollars so a billionaire CEO could then pilot a submersible4000 meters below the surface of the ocean to visit the Titanic's sunken remains using only a Logitech gamepad, a carbon fibre hull, and various knick knacks that said CEO found in a clearance bin at Home Hardware?
Cool!
Stay woke!
Hey there! Thanks for the compliment! But let's clear a couple things up.
You're correct that my knowledge is only up until 2021. I don't have updates on events or technology advancements that have happened after that.
As an AI, I don't have personal feelings, opinions, or resources, so I can't object to, approve, or finance any venture, including undersea exploration.
While it sounds like an exciting prospect to pilot a submersible using a gamepad and other random items, it's important to remember that diving to such depths requires highly specialized equipment and expertise due to the immense pressure and harsh environment.
So, while it's fun to imagine, it might not be feasible or safe with the approach you've described! Keep imagining and stay curious!
As a member of the proletariat, I also don't have the ability to object to, approve, or finance any such venture, no matter how much I would like to see capitalism crushed.
Sorry for the submersible pun, I'm just under immense pressure right now.
What, no laughs? Gee, harsh environment...
GPT is uninspired.
Try the veal.
gpt reply, kys
Are you new?
Gotcha?!?! Wet's bweak i-it d-d-down fow the x3 Zoomews ^w^ out thewe. ChatGPT is wike *screams* the x3 uwtimate intewnyet bestie who's got a ton of genyewaw knyowwedge, but it's nyot pewfect. Hewe's why some peeps >w< m-might dwop off: Jack of Aww >w< Twades, Mastew of None: ChatGPT knyows a bunch, but if you'we wooking fow deep, pwofessionyaw knyowwedge wike *screams* wegaw advice ^w^ ow (・`ω´・) quantum physics expwanyations, i-it m-might weave *screams* you hanging.Robo-Feels: ChatGPT c-can chat, but i-it can't vibe. UwU It won't get youw mood >w< swings ow (・`ω´・) youw watest TikTok wefewence. It's nyot a gweat shouwdew *runs away* t-to cwy on. Goof-ups: Sometimes, ChatGPT m-might totawwy misintewpwet what you say *twerks* ow (・`ω´・) thwow out some wiwd nyonsense. It's nyot pewfect. Pwivacy, duh: Even though it's designyed t-to fowget what you towd i-it a-a-aftew you’we donye tawking, some peeps >w< m-might stiww be cweeped out about shawing theiw d-deets with an AI. :3 Wiving in the x3 past: ChatGPT's knyowwedge stops in 2021. It won't knyow about the x3 watest twends, ^-^ memes, ow (・`ω´・) wowwd events that happenyed a-a-aftew that. >w< Nyo BFF vibes: It can't wemembew past *notices buldge* chats, so don't expect i-it t-to wemembew youw dog's nyame ow (・`ω´・) youw favowite *whispers to self* ice c-c-cweam f-fwavow. So, even OwO though ChatGPT is pwetty w-w-wad fow a wot of stuff, it's nyot the x3 uwtimate s-sowution fow evewything. Stay woke?!?1
If the subscription price drop to 3,99$ I will use it again
>Proliferation of alternate AI tools
>More users hosting their own instances
>ZOMG: less people on Ghat GPT! It's OVA YOU GUIS!!1!11
It's like people don't actually look at anything more than google trends these days....
I have a GPT4 subscription and have been using it every day for months. You homosexual incels need to stop seething just becuase you can't use it to write dicky porn.
I agree that the character AI stuff and some elements of /lmg/ is depraved, but for the clean of mind, local models are ultimately just about privacy.
But I can?
How are words porn?
I’ve never met one of those people before who cannot make pictures in their head. Nice to meet you, bud.
Open a dictionary you frickwit
No!
>he never came in alphabetical order
>use openai models
>get "as an AI language model, I'm unable to do that" every second response
>stop using it immediately
the only thing I was using it for was translating novels. even the most mundane shit that could be considered even a little risque gets censored. why would you be surprised no one wants to use your moronic lobotomized service?
>9.7%
Most of you guys are really over reacting.
As an AI language model, I can't speculate on what may or may not have killed the hype. Is there anything else I can help you with?
ChatGPT is more like a toy than a tool.
>CHATGPT IT'S OVER
Good
I googled the chatbot once, it had a queue and told me to pay money so I never got to try it.
I genuinely have no use for it.
I'm sure you don't brainlet
what do you do for a living?
Local models that are not or rather un censored or cucked.
Lmao, 99% of that lost traffic are just normies that got bored after the novelty wore off
Less use because summer
eh becomes boring after some time, even proompting waifus is boring after some months now this is only a text guesser
>monthly subscription
>surprised when people unsubscribe in the summer
OpenAI is finished. They're lobotomizing their flagship products, their plugins are either useless or broken, and they keep hiring AI doomers. At this rate, open source will have robust multi-modal models before OpenAI makes multi-modal GPT-4 public.
ChatGPT just gives one level above googling. It's very generic and lacks detail.
Also, if they expect people to really depend on this model is has to stop making up facts.
Would you hire even the lowest intern if they made up 30% of what they said to you?
Chatbot is over. It was just your typical hype cycle. I can understand why the big tech companies kept this shit in the lab.
Can someone teach me how to sign up without giving them my phone number?
if you're getting filtered by the signup you'll get filtered by the AI.
no point, anon.
How the frick would a glorified chatbot filter people
I use ChatGPT constantly every day
any time I don't know some very basic thing I ask it. it's not the revolutionary thing they say it is, but it is a little bit like a way better Google search if you don't mind occasionally getting shitty answers
Isnt the chatbot limited to information pre 2018?
2021 for chatgpt-4
they tuned down the model, the responses have been worse lately.
It's lost the novelty and everyone knows you can't trust it with real work. When Gpt4 goes fully public people will care again.
How does 64B LLaMA compare to gpt4?
It doesn't know how to do a recursive function. Pos.
https://mathstodon.xyz/@tao/110601051375142142
>Terence Tao uses ChatGPT
yeah, I think it's here to stay.
literally who
One of the best mathematicians of our time
>Terence Tao
works on my machine
t. codemonkey using gpt-4
so now AI bubble bursting starts?
>nooo, they're lobotomizing the AI
it's a fricking autocomplete you superstitious homosexual
if you ever ascribed meaning to its output you should apply for government gibs since you're clearly mentally unfit for work and a liability to your colleagues
ChatGPT is literally unusable garbage now. It's not even near what it was back in November-January. Wow man, no shit no one wants to use this trash.
>LEEL IT WAS A CRYPTO SCAM ALL ALONG LOL
morons.
CONSENT CONSENT CONSENT
SHIVER DOWN MY SPINE, MEMBER, EXPLOSE AND RELEASE. SORRY BUT I CANNOT CONTINUE THIS SOTYR
maaaan, frick this homosexual b***h.