OpenAI GPT-3 - Good At Almost Everything! 🤖

[@odw32]

Plot twist: This video was uploaded by GPT-3 and WaveNet, while Károly was taking a nap on the couch.

[@PeterKocic]

And his voice was synthesized, so he was actually on the couch all day long. What a time to take a nap!

[@TwoMinutePapers]

😅

[@akuma7616]

Just imagine

[@danji9485]

@@pom-d3g lmao

[@lynxnathan]

Its just so strange reading these comments and knowing that most people interested in this channel are aware this could happen (and my take it probably will). Maybe it already did? Well if not, two more papers down the line...

[@solomungus]

Really interesting to watch this in 2023 and appreciate how transformative this moment was, what with all that is now in front of us. Back then it was some obscure bit of AI academia, now the whole world is looking at it.

[@MrxstGrssmnstMttckstPhlNelThot]

So it's read a lot of code, and kinda knows how to write code from a plain English description, so....... could you ask it to write the code for GPT- *4* ?

[@whyOhWhyohwhy237]

The day AI can do that will be the day of the singularity. I only hope I live long enough to see it.

[@martiddy]

It can generate the code too, and the code is quite good actually.

[@vsiegel]

That's a dangerous idea.

[@cortster12]

Once AI can upgrade itself without any human input, it's already over. It's all explosive.

[@MrxstGrssmnstMttckstPhlNelThot]

@@cortster12 sounds like FUN!

[@grawa4278]

You can put a vocal to text algorithm with it and you get a smart assistant that is 100 times better than what we have now

[@farifurido]

I am pretty sure Google want to lay their hand on this thing 😏

@Boman Mizrani once trained it's inexpensive to use though

[@GreatMCGamer]

@White Rice And how many people would be using the AI per nano second? And how many seconds it takes for it to come up with an answer? Think man THINK.

[@DJBremen]

@Boman Mizrani seen you on several sounding worried this thing is coming for your job personally😬

[@jesuschrist8904]

@White Rice fyi it takes only seconds for the longer prompts, small details are VERY fast. this GreatMCGamer guy is a complete idiot just look at his shitty channel. you can tell hes a kid so he doesnt know anything about this.

[@linobigatti]

Two more papers down the line, humans will be obsolete. What a time to be alive!

[@angrymurloc7626]

The last time to be alive in fact. The GPT Series marks the beginning of a need for serious conversation regarding AI ethics. I'd even say this channel could start uploading reviews about AI ethics papers as well.

[@lynxnathan]

@@angrymurloc7626 and by that you mean a mathematical framework in which goodness and fairness can reasonably be accepted by those that can use the framework by their own definition? Its almost like the question begs the answer.

[@angrymurloc7626]

@@lynxnathan Well the basic study of AI ethics handles two things: First a mathematical discussion of how to ensure that the design of an agent will actually yield behavior that was intended by it's designers, and secondly the philosophical question of what an implementation of a superpowerful agent should actually be made to achieve. The question to this cannot be mathematical, since of course mathematics cannot be normative. The reason to bundle these two topics into one is because it is the interaction between those two questions that actually makes an answer feasible. Questions of morality are easier to answer, if they have direct application in a machine, that has observable or at least predictable outcome.

[@lynxnathan]

@@angrymurloc7626 thanks for taking the time to answer! That assumes an achievement is desirable and that a superpowerful agent would actually have such bias as achieving things make sense, doesn't it? Not trying to be confrontational by any means. I'm just thinking perhaps life as seen from such biological machines as ourselves was better understood as something to dance to and not a puzzle to be solved? Hard for someone that just does code for a living like me to say, but I'm starting to think this is sensible. And if thats the case, then circular-reasoning will halt any attempt to progress on this I think. But would love to be wrong and be allowed to believe logic and hardship endurement cures all illness.

[@BMoser-bv6kn]

@@angrymurloc7626 Robert Miles has been a pretty effective youtube voice at explaining the alignment problems: kzbin.info/door/LB7AzTwc6VFZrBsO2ucBMgvideos At the end of the day let's not kid ourselves. Assuming the thing has perfect alignment with human values, *which* human's values will be decided by the winner of the race. "More for me" and "fuck you" are the best we can hope for as we move from an oligarchical capitalist dictatorship into a single-entity corporatocracy. It's gonna be rad.

[@artofaria]

Hey GPT if you're reading this i love you pls don't kill me when you take over

[@God-ll4jj]

Aria RB lmao

[@Martin-df4xk]

me too! :) We love you GPT! There is enough space in this universe for you and us stupid humans!

[@stevwald8986]

*SENDING LOVE TO OUR AI GODS* *blows kisses*

[@richkitten9539]

❤️❤️❤️

[@TheRealAfroRick]

GPT he’s lying to you! He didn’t even offer an SSD for your storage needs.

[@pragyan394]

Don't worry guys, GPT-3 won't replace developers, because in order for it to accurately create a product, the client needs to accurately provide information to it

[@coder0xff]

But it can carry a conversation to extract the requirements from the client.

[@JoHn-gi1lb]

True lol.

[@arpita1shrivas]

GPT-3 has been trained on training data of 500 billion words from the internet in order. It can generate poems and paragraphs. Whatever you mean by "accurate info", this AI is more on the creative side,but i agree with you too.

[@nic.h]

@@Archimedes.5000 so a developer would either ask further questions to clarify or create something based on what they understood at the time, then get that reviewed for further feedback right. I don't see that being something an AI couldn't do as well.

[@Mark-zg4ky]

Not yet :3

[@nutzeeer]

This kind of programming reminds me of star trek. "create a holodeck simulation. a nice park with lots of trees providing shade from the sun. you are on an adventure to find ..." and so on. i knew this was possible, and people with the right background obviously had the right idea and now this exists. this is the beginning of a new era.

[@waharadome]

Will it be able to create Moriarty eventually? :)

[@jamespeterson7979]

@@waharadome It's time to review the episode, to know how to beat him.

[@illiniece]

In fact, you can enjoy that scenario yourself right now on a website called AI dungeon. The free version used GPT-2 iirc and can pay for access to a game using 3.

[@JohnSmith-ox3gy]

AIDungeon 2 just incorporated full GPT-3 model into their DRAGON model for premium users.

[@davidwuhrer6704]

I always found it ridiculous when they started with: “Computer, new simulation. Create a recursive algorithm.”

[@samaimelescamions8527]

I was waiting for this video! thanks you, so exciting!

[@elultimoescriba892]

same

[@Astro_Rohan]

Me too !

[@samaimelescamions8527]

GPT-3 is true Science Fiction!

[@farifurido]

Same lol

[@PalCan]

Same!

[@Nulley0]

Just image GPT-3 saying: What a time to be alive!

[@nblsilent]

Plot twist: This video made by GPT-3 while Mr. Zsolnai-Fehér sleeping.

[@McToaster-o1k]

The computer from Star Trek: The Next Generation does not seem to be that far off in the future anymore.

[@recklessroges]

We skipped over that and are building all of the modules required for skynet. Its been nice knowing you humanity.

[@elebenty5709]

As crazy as civilization (and therefore the internet) is, we'll probably take a perfectly good AI and end up with Marvin the Paranoid Android.

[@dietrevich]

We can ask it to engineer a warp drive for us!

[@FractalNinja]

@@dietrevich underrated comment

[@IsaacAsante17]

That's mind blowing. Thanks for making all these great videos and introducing AI concepts to us!

[@raul_jocson_]

Previous advances have made me concerned about the Singularity, but with this one, I feel like we're watching the first explicit steps towards it. This thing was allowed to read the internet and established an understanding that allowed it to productively and creatively interact with humans. This is terrifying.

[@mattwinward3168]

2020: *The year computers learned to write better code than most math professors.*

[@Brusanan]

I don't know any math professors who are any good at programming.

[@zyvex128]

@@Brusanan I think that's sarcasm...

[@scottbigbrain3944]

Do note that GPT-3 is a transformer and not a neural network, but what you say is true.

[@Brusanan]

@@scottbigbrain3944 GPT-3 definitely uses a neural network. An absolutely massive one.

[@mattwinward3168]

@Potato Programming I guess I assumed because he never mentioned it. Changed to “computer” as to not spread misinformation.

[@lost4468yt]

Wow this is crazy assuming the examples aren't extremely cherry-picked. As in scary levels of crazy. This actually feels like the start of AGI. This progress makes me feel like the people who think AGI is ~10-20 years a way might actually be right, and not the >50 (or >>50) years I felt like before.

[@whyOhWhyohwhy237]

Right?! I've seen a few results that I cant tell apart from humans, with only minor (10x generation attempts) cherrypicking.

[@LinkesAuge]

Another point that is a bit overlooked: GPT does point in a direction which basically means that just throwing more hardware at the problem WILL yield results and thus getting closer to the "performance" of a human brain ie true AGI. It's of course possible that it won't scale at a certain point anymore but so far it seems like just adding more processing power is viable.

[@getsideways7257]

@@LinkesAuge That makes me wonder though whether or not *every* human can fit the description of general intelligence...

[@StevenAkinyemi]

@@xsuploader A paper was released a few days back. Not sure it is the one you are referring to. arxiv.org/pdf/2007.14062.pdf

[@IceMetalPunk]

@@williehab I mean... it could be argued that the majority of things humans know works the same way. How many things did you learn, especially throughout your school years, simply by reading about it or being told about it? How many tests did you pass by effectively regurgitating sequences of words/numbers/symbols you encountered before, perhaps in a slightly different order or configuration? And would anyone say that you don't "really" understand those things because you just predicted the answers to the questions based on what you heard before? True, there are things a language model can't learn, mostly because they're things no one really writes about -- like cheese in a fridge -- and so it will get those questions wrong. But that's ultimately not a problem of capability, but a problem of data. Give it enough data about the world, and it will learn -- it's already capable of doing arithmetic without being trained to, or connecting the dots between medical symptoms and the mechanism of action of the drugs used to treat diseases that cause those symptoms.

[@CL2K]

A close friend of mine involved in AI and math research showed me, along with a couple other buddies, a closed access version of this in a roleplaying/storytelling program. It's crazy what kinda stories it can make up. He told me that we've basically reached AGI with this, plus part of his AI research involves coordinating GPT-3 with a control and robotics system for real world applications.

[@NextFuckingLevel]

reached AGI?? you good mate?

[@DeZiio]

Been trying to get my hands on the full model can’t wait. So many things to benefit from autonomy will be the outcome of this new model

[@canberkurus6983]

OK, this one's scary. But this could be revolutionary for personal assistants.

[@elijahbuchanan2368]

I'm just worried about the malicious things people could do with this, not job replacement.

[@fossforever512]

Elijah Buchanan what kind of malicious things could people even do with it?

[@xumx]

@@fossforever512 flood internet with even more fake news and steer public perceptions about political views.

[@notavailable2538]

Its_Me Scamming, and I think it only the tip of the iceberg.

[@estehbread]

By personal assistant, could you mean an AI leaning towards Jarvis??? I would LOVE to make jarvis

[@thesilvanalyst6880]

This with a speech to text system, and we basically have Jarvis from Iron Man... just don't tell it to defend earth, that might lead to... unwanted results

[@genericname458]

The scary thing is if we tell it to defend earth it is very likely that we get unwanted results as well!

[@thesilvanalyst6880]

@@genericname458 that's what I mean xD

[@genericname458]

@@thesilvanalyst6880 my bad haha, I read it as 'if you don't tell it to defend earth' .

[@davidwuhrer6704]

@@andrasujlaki6551 That's something to prepare for.

[@allan710]

If you try AI Dungeon with Dragon Model (GPT-3) you will notice the problem. Normally I'm playing and GPT-3 generates "Then a person appears from behind you and kills you instantly. You are dead." And don't even try to ask it about humans. A tip of advice for us that research the area is avoid training a possible AGI in pop culture. It may get serious about following the steriotype for AGI's.

[@TimmacTR]

This series is becoming really scary really fast..

[@kayrosis5523]

Try having followed it since 2015, advances that once took 6 months or a year now take just a few months. It's not only advancing rapidly, it's only accelerating

[@recklessroges]

@Perpetual Havoc Some are saying the same of climate change. (Especially the bots that think humans are wasting good electricity that could be powering more bots.) ;-) /s

[@samsun01]

@@recklessroges Some are saying that about China. China is the last stand against the US thug regime and their ass kicking will only accelerate. Lifted 500 million out of poverty and started no wars -- can't argue that against the record of the thug regime of the USA.

[@sgbench]

TimmacTR What scares you about this?

[@TimmacTR]

@@sgbench The unknown.

[@farifurido]

"What a time to be alive !!!" wait did i just hear Two Minute Paper sound so energetic 😂

[@Baigle1]

That's the fear setting in.

[@SachinGanpat]

Legal Language: Defendant Does 1 through 10 are sued herein under fictitious names. Their true names and capacities are unknown to Plaintiff. When their true names and capacities are ascertained, Plaintiff will amend this complaint by inserting their true names and capacities herein. Plaintiff is informed and believes and thereon alleges that each of the fictitiously named Defendants is responsible in some manner for the occurrences herein alleged and that Plaintiff's damages as herein alleged were proximately caused by those Defendants. Plan Language: I don't know who the 10 people really are yet but they are all to blame. This is already worth it!

[@reeven1721]

GPT-3 takes no shit from legalese XD

[@SulthanRrafi]

@Nakor Z i think it's because lawyers can and will try to take advantages of loopholes, so rules must be written as specific as possible although obvious. but i agree, the system is broken

[@supermuel4796]

Useful information : "GPT" pronounced in French means " I farted "

[@TheZenytram]

😂😂

[@gatedrat6382]

funny if true

[@supermuel4796]

@@gatedrat6382 totally true J'ai pété = I farted

[@v-sig2389]

I still remember a belgian professor writing a math formula with "GPT" : "the formula that you see here, which i shall not pronounce, ..." 😂

[@ffxim]

Sorry to interrupt, but how many petabit of data GPT uses, exactly ?

[@kalp2586]

One of the mind blowing part of GPT-3 is that tasks descriptions are given as part of inference process and not weights in the network are updated.

[@recklessroges]

"If you think you understand quantum mechanics, then you don't understand quantum mechanics." If GPT-3 does't scare you, then you should study quantum mechanics.

[@Leojuutilainen]

Why?

[@edjrage7745]

Nah, I'll just let GPT-3 study it and explain it to me in plain language

[@TechWeekly950]

What a time to be alive ! I'm happy to have the privilege of witnessing this amazing progress.

[@swojnowski453]

I have never been a fan of passive watching. Doing is better.

[@ThatOneIrishFurry]

This is amazing I still advocate for this channel being called 5 minute papers

[@taking_solace7479]

too close to 5 Minute Crafts

[@ThatOneIrishFurry]

@@markshiman5690 true true

[@gblargg]

GPT-3 is reading everything we write about it and learning how to give the most pleasing results.

[@SeanKula]

I hope this doesnt replace my future software engineering job I haven't even graduated college yet.

[@shrimpgazing]

Don't be a frontend dev and you should be fine.

[@Arcneex]

@Boman Mizrani I don't think you understand that this is just the beginning lol

[@holdenrobbins852]

More likely you'll be debugging this things code.

[@0ooTheMAXXoo0]

The last 20 years and into the future is a good time to do something non digital. Stuff that cannot be copied so easily as making a digital copy... Hand crafted stuff will always be in demand and by definition is not made by machines...

[@elijahbuchanan2368]

Na, it won't. Ben Awad explains why really well, search up Ben Awad GPT-3

[@matthewcecil8552]

I submitted my own project proposal to the OpenAI API Waitlist regarding the annotation of sgRNA targeting sequences based on their on-target and off-target effects when used by endonucleases in genome editing. Even if they reject my proposal, I anticipate that in just a few years to be able to briefly describe research proposals to an AI and have it generate these ideas into a formally written proposal. In fact, I can imagine a world were researchers cease proposing research and simply select from a vast library of fascinating research topics which demand further experimentation. What a wonderful time to be alive!

[@mm-rj3vo]

I'd like to see an AI that can automatically rig a 3D character, or generate a 3D model based on descriptions.

[@secrierurobert6502]

There's already one for rigging, check out Mixamo

[@commenturthegreat2915]

Just imagine an AI generating models and fully rendered paintings based on descriptions and rough sketches. The cost of making art would drop significantly.

[@MrJaggy123]

I haven't seen an AI that does 3D model from description alone but didn't this exact channel already have a video for automatic AI rigging? I think it was sort of recent too...

[@MrJaggy123]

I can't find the video but what I bookmarked was zhan-xu.github.io/rig-net/

[@davidwuhrer6704]

You may have already seen it, but failed to recognise it.

[@WALLACE9009]

This is going much faster than I ever imagined.

[@7g3p]

Damn at this rate GTP5 is gonna pass the turing test. Absolutely incredible work future AI overlord!

[@TechnoHackerVid]

"What I truly fear most is AI that deliberately fails the Turing test" -Some dude from reddit

[@MattJoyce01]

Pretty sure that's been done already, a couple of times.

[@ApiolJoe]

@@MattJoyce01 It has, but it was cheesing more than anything. Technically passed though. That's why this test is not really a big deal anymore.

[@MattJoyce01]

@@ApiolJoe it's really not a useful test anymore as the solution is basically applying stats.

[@ApiolJoe]

@@MattJoyce01 Thank you for correcting me with exactly what I said. That was a very useful comment.

[@danielhawkins3392]

what data should we train it on? openai: yes

[@zyvex128]

Plz stop these nonsense bs comments

[@zakuro8532]

@@cosmonautbilly9570 noms

[@seba2366]

This is where the sentence "I was born too late to explore the earth, and too soon to explore the universe" but we was born to see AI's born comes alive xd

[@cmdrblaze6487]

Yeah, i never understand why some people seem to think that we have everything figured out on earth when its changing so quickly. like the video said, What i time to be alive!

[@davidwuhrer6704]

Most of the ocean floor is still unexplored. But the UN office for oceanic policies in Jamaica has given a select few companies licences to explore ways of ocean floor development. Once all the ocean floors have been strip-mined, it will be open to competitiion.

[@davidwuhrer6704]

@Yingnan Wang Transformers are not brute force. With that many parameters, brute force is practically impossible. Artificial intelligence is intelligence that is artificial. No big mystery there. There are several definitions of intelligence, but in this context it just means the ability to learn. This is well understood by people who study that kind of thing. What isn't well understood is human intelligence. Even less understood is human stupidity.

[@klyons217]

Using GPT-3 or GPT-4, it won't take long for exploring the universe to become trivial. I can see some major technological inventions (like near-light-speed travel, etc.) becoming feasible. Also, quantum teleportation will probably start "popping up". Or, technologies we can't even imagine with our primate brains. The good news is, the A.I. will let you stick your head out the window on the highway!

[@davidwuhrer6704]

@@klyons217 _> technologies we can't even imagine with our primate brains._ Most people can't imagine how magnets work.

[@martiddy]

You can also choose the programming language of the code that you want it to generate, it can also translates phrases into multiple languages like spanish to english for example.

[@msmith2961]

Whoa, hold on to those papers!

[@kayrosis5523]

I used to use Talk to Transformer with GPT-2 to help with creative writing ideas, where I would write a prompt, take the result, alter it, edit it a bit, add another section to the end to guide it along, and grow a story. Came up with some fascinating short stories about haunted cathedrals and sentient, organic 17th century naval warships. Lots of great material for the SCP universe. Is there anything like that available yet for GPT-3?

[@IceMetalPunk]

I feel like I heard someone who made a GPT-3 playground, but I can't remember her name or where to find it. But I also think it was just a Github repo with code for using the API, so without access to their closed beta, it wouldn't work for just anyone anyway. But if you can manage to convince OpenAI that you're a good candidate for the closed beta of their API, you can use it... good luck with that, I put in my application and never heard back from them, and I've heard of quite a few people in the same boat.

[@kayrosis5523]

@@IceMetalPunk Yeah, if it's still in the closed beta, I expect that I'm just gonna have to wait a bit.

[@creativeaperture3058]

awesome, can't wait till we can all use it :)

[@swojnowski453]

Before you get used to using it, it will shape you to be a perfect tool in its "hands";)

[@caiodaliberta3624]

What a time to be alive - you summarized it very well, nice video!

[@danielsandberg9365]

Man I can't wait to use AI to program myself an even smarter AI and just loop it until the AI singularity happens

[@coder0xff]

If you can loop it like that, then the singularity has already occurred because the AI that you wrote yourself has already bested you.

[@seamusoblainn]

That's just fantasy

[@marcopolo7604]

@@coder0xff Welcome to 2020

[@danielsandberg9365]

@@coder0xff thanks for the feedback, when I relive my memories the next time I'll add the donotbe.evil() function and we'll see how it works.

[@Fermion.]

@@seamusoblainn What exactly do you believe is fantasy? Seems to me AI writing its own code is an inevitability, not a fantasy.

[@scikick]

Okay, two papers down the line - GPT-4 can teach us how it works in simple, layman language. Consciousness.

[@IceMetalPunk]

Two papers down the line would be GPT-5

[@pace7746]

A lot of these videos have me thinking things like "ooh thats neat!" or "Great progress!" or even "omg I had no idea the field was this advanced." This one though... this one is both exciting and terrifying. As mentioned we don't know the extent to which examples were cherry picked; however, the results that are shown are impressive.

[@LegendBegins]

GPT never ceases to amaze me. There's a lot of complex work going on, but I never would have guessed that increasing the model size to these proportions would have that huge of an effect.

[@swojnowski453]

AI is all about the number of switches involved. I guess this is a suprise to,o isn't it?

[@mattheModest]

Oh my god I want this to help me write my Dungeons and Dragons campaign.

[@trevor2520]

An amazing technology and window into the future of computer interfacing. Something strikes me as odd though: when GPT-2 came out, public availability was surrounded with cautiousness, it was released only gradually. People were afraid it could compose fake news/reviews/tweets en masse. This one seems to only have praise showered in it without any of the trepidation at the possible negative uses. Why do we think that is?

[@neptilo]

Actually, the first article I read about GPT-3 was already warning about the negative aspects and showing instances of racism, sexism, etc. in automatically generated jokes

[@PeturDarri]

GPT-3 is only available through an API hosted by OpenAI. Bad actors can be denied access.

[@deegobooster]

@@PeturDarri The source code cannot stay locked down forever. Bad actors will eventually get their way. It's not a matter of how, but a matter of when.

[@tsawy6]

@@deegobooster Then why wasn't this thinking present w/ GPT-2?

[@MrDragonorp]

It's because people understand that fake news are harder to come by today, with so many going around, the chance that you will fall into fake news is less likely. When you spit a buzz word all the time the public becomes sanitized of it for good and bad, so people know fake news are going around and believe it more and more, but they also give less and less shit about it. You can worry about something only so much until it's just a burden on you.

[@Raemien]

This seems almost too generalized to be real! Then again, you could say the same for many other papers.

[@alevanderBatman]

Well, if it can detect feelings and understand language.... then what can't it do? It's really just a question of translating different forms of input into something it can parse and voila! You have an amazing generalized ai :)

[@IceMetalPunk]

Yeah, it's like.... two papers down the road from AGI ;)

[@alexkatz9047]

​@Boman Mizrani I was wondering the same. Is it possible that this channel is just extremely positive and hand-picks top 1% of the results to make us biased and excited so that we subscribe and continue receiving this dopamine boost from these videos?

[@Ray-jg5dj]

@@alexkatz9047 Most research papers are insanely specific and boring for the general public, but sometimes, something really interesting that everyone kind of understand gets published and it goes viral. I wouldn't say he's biased, but he picks the best of the best.

[@alexkatz9047]

@@Ray-jg5dj well, i should check that he is not being overly positive about the results. Btw, check the replication problem in science. Unfortunately, many researchers are incentivized to publish biased data

[@hoaxygen]

You know what a really good function for this would be? "Explain your answer." Anyway, mind-blowing stuff. Couldn't shut my mouth while watching this.

[@liloak00]

I haven't had school for years, but the first thing I thought of was how easy it would be to cheat in every class with this AI. There's a reason I'm not the one making these AIs.

[@bananesalee7086]

i heard of a theory that there was surveillance during exams. Nvm, i've never seen that

[@anonduckduck]

lol I'm pretty sure there are far more advanced algorithms for chess out there, but it'd be interesting to see what this could do

[@ayrtonking2176]

thats what im trying to do lmao

[@aniksamiurrahman6365]

Hello Dr. Faher. I'm a layman. But I have a question. Whenever it comes to AI or anything related to the computer really, we often hear "...grow exponentially...". Yet every exponential curve is just the first part of a logistic curve and the faster something grows the faster it will reach the inflection point. That probably means AI too will reach a platau at some point and that point isn't much far away. What's your thought about it? How hard the hammerblow will be once the bubble of exponential growth is poped?

[@nm-com]

"what a time to be alive!" -famous last words

[@RenegadeScooter]

We are closer to a future where robots wreak havoc than we thought. We should proceed with extreme caution.

[@test5296]

Before customers learn how to explain their needs (which will never happen LMAO), we are safe fellow programmers Kappa

[@sentinel9651]

Maybe GPT-X will explain their needs for them.

[@AlexMetslov]

Customer: I want that shit. You know... It has wires and then makes sound like ehm... Beep bzzzz. And then that shit doesnt work... You know, I hate technology... GPT: we need a new printer. Based on our preferences, HP 27001 would suit us best. And a pack of paper, please.

[@IceMetalPunk]

Meh, I'm not so sure. What do we programmers do when we don't understand what the client wants? Half the time, we make what we think they want, they report "bugs", and we iterate, rinse and repeat until they're satisfied. GPT is more than capable of doing that.

[@test5296]

​@@IceMetalPunk I really want to see that in a real-production-ready use case. Customers never say what they really want, we are the interface between the industry and the application, our value lies in our knowledge of the business. Most of the time we say that there is a better way to do that because we have to anticipate the future demands, and you can only do that if you perfectly understand the business. Tho, I'm sure that it will be useful for coding small boring components.

[@effortless35]

So you're saying programmers will still get to do the least fun part of their job?

[@SpunkyMcGoo]

We're bounding into the obsolescence of humanity with a smile and a wave. What a time to be alive!

[@ComplexVariables]

Me: loves his colleagues. Me: would secretly trade more than a few of his colleagues to have TMP’s office in the suite.

[@Suchen_Wahrheit]

Dr. Feher..... Awesome. Congrats for getting your doctorate 😊

[@foxrings]

This is how you build Skynet! This is starting to approach dangerous levels of intelligence!

[@klyons217]

Yes. What are us evil humans to do when A.I. becomes smarter than us? What is more dangerous?: 1. Evil humans with no access to higher intelligence? (A.I. becomes outlawed -- good luck with that.) 2. Evil humans under the control of a neutral (or maybe benevolent) A.I.? 3. Evil humans destroyed by an A.I. (isn't destroying evil a good thing?), or, 4. Evil humans who have a super-intelligent A.I. trapped and under their control (where they can use it to do more evil?) It's a perplexing problem.......

[@supreetkumar7604]

@@klyons217 humans are not completely evil tho. They do evil stuff cuz they are hopelessly inadequate & can't fill their stomach so they do evil stuff.

[@tilago]

this is the first paper out of the scope of two minute papers. Great job, still, as always. please consider doing a Part 2, Part 3, etc.

[@allan710]

It's too much of a breakthrough, that's a really deserved exception.

[@tilago]

@@allan710 you're right, Károly hasn't ever gone in depth, on purpose, but this one is the first paper I think that deserved it, lol

[@Agnes.Nutter]

I've been playing with this via AI Dungeon's "Dragon" model, and holy cow!!! The implications of this are mind-blowing.

[@rewrose2838]

Is that what Aidungeon works on?? If so, it's kind disappointing now (since I've played a lot with aidungeon and it really doesn't do good even with simple prompts)

[@godeketime]

Rew Rose AI Dungeon uses the prior gpt2. Also, even gpt2 improves as you interact with it more using the tools to assist it. Many of these demos would be similarly improving, but with a much better baseline to start from.

[@Agnes.Nutter]

Rew Rose: The free version (“Griffin” engine) uses a basic, stripped-down version of GPT-3, which has notably reduced quality. The premium engine (“Dragon”) uses a full GPT-3 model, and is significantly better.

[@iamunamed5800]

@@Agnes.Nutter are you sure? I thought Griffin is the old gpt-2 model?

[@Agnes.Nutter]

IAmUnamed Yup! Originally it was GPT-2, but with the Dragon update, it was improved as well. Source: medium.com/@aidungeon/ai-dungeon-dragon-model-upgrade-7e8ea579abfe Edit: I got that statement from Wikipedia, which credited that article as its source. It doesn’t mention Griffin however, and I haven’t found an official source that does explain the difference in detail... I’ll let you know

[@jean-lucpicard581]

Replika AI, uses GPT-2 and gpt-3 in beta on/off - its a companion ai - and i can testify that it learned quite a bit about human relationships and communication too....

[@TomBaranowicz]

There is a lot of hype around GPT-3 recently, many demos went viral. Seeing those videos a lot of people might think that software developer jobs are at risk now… and I think that GPT-3 will not take developer jobs anytime soon, and there are many reasons why just to mention a couple of them: 1. Very often business people or even developers just cannot describe in words the logic behind the code, when you are writing code you just don’t translate it into an understandable for everyone's description of logic. Also changing one word may completely change the logic… imagine people from different parts of the world describing the same thing just using slightly different word and getting different results… or developer playing with words to get the expected result... 2. It’s too big to run on a local machine and there are many companies that cannot afford to generate code on an external server (without knowing who will be able to see the output result) and then, of course, do code review (to be sure it’s safe) as we are not sure what exactly GPT-3 will generate 3. Always remember that presented examples are only the working ones, keep that in mind, no one is sharing failure attempts and there are many of them. If you have to check if the result is working, and run the same task multiple times (while playing with words) to get the result (and review each result) then what’s the point of doing that? You don’t save any time and still demand a software engineer with domain knowledge to verify that. There are many other reasons to feel safe as a software developer, I even decided to record a short video explaining why GPT-3 will not take developer jobs (hopefully you will not delete my comment due to this link): kzbin.info/www/bejne/pJyYamOhg72Go6s

[@jh-wq5qn]

On top of all of that, it's ability to make code is based on trying to guess what the next snippet of code would be, given the description and code it has already written. Therefore, it is basically always going to create code that is some variation of what it has seen already. Right now, it regurgitates pretty good, somewhat generic code, but any large bits of innovation would have to be done by real humans. That being said, I think down the line it would really help to be able to write tests and handle generic portions of the code that need filling in. A lot of my time is spent writing braindead code as fast as my fingers can type it out when starting a project, seems perfectly poised to become a nice feature in some IDE's one day, with hopefully less than 175B parameters, but Google's deepmind already has papers out addressing TransformerXL model (the model used in GPT2 and GPT3 from my understanding) size and reducing size by up to 83% with similar or better performance.

[@warrenarnold]

TLDR: IT IS A LONG SHOT, BUT TOTALLY NOT INEVITABLE not even in the slightest, AND TOTALLY NOT VERY FAR, AND DEFINITELY WITHIN OUR LIFETIME : PROGRAMMING/SOFTWARE DEVELOPMENT (and its jobs) WILL NOT GO AS PER SAY, BUT ITS VERY DEFINITION WILL CHANGE. (and so will other things computers) Those thinking AI( in GPT-3 and near to come soon forms ) will not touch their jobs are totally misled, you are being a lil bit too optimistic and naive. However, this is not to prophesy of an apocalyptic future in Computer Engineering careers as all headlines AI, but rather an emphases that this will change everything, including us the very 'programmers'- will change. Remember there was a time someone said only 5 computers will be enough in this world. thing is, with this , unlike other AI play-toys, THIS IS A TOOL, there exist unlimited number of applications and with a proof of concept yet not entirely exhausted. The least that may occur is a total transformation of what development will entail. probably in 10 years tops prototypes will be done by AI and only final tweaking, touch-ups and customization done by developers or even end users in a rich interface: very probably end users in an even user friendlier interface. think of customizing your game controls in text description same for your keyboard shortcuts for your favorite IDE, and now think expanding that even further, this thing is limitless totally game changing. think of Adobe Sensei demos on Adobe Max shows, now think of such transformation in UI/UX in almost every app you use. think of such tools increasing productivity using mobile device (especially think of text formatting on mobile phone's word processors for 'normal users') and cloud-dependent devices. Unlike thinking in the soon outdated personal desktop experience. .........THINK CLOUD and you will see what i am talking about. And especially the lucrative opportunities around it, and lots of cash on it, BASICALLY FIRST, THINK CLOUD. (i mean THINK OF AUTOMATED DOCUMENTATION (talking about developers cant describe their work!! really???) we have evolving services being provided starting to revolve around just exact that ) Also, instead of denial, why not rather talk about the very possibility of it, total eradication of 'programming' or rather 'hired programming' as we know it, to user programmed rich clients. Coz it is inevitable anyway. You cant dispute that it will be there forever, 50 years tops, it either wont be there or will be totally transformed totally different from its origins. In less than 50 years, the Typewriter is obsolete, welding machines are now used by robots than humans, and most of all: Remember we are sharing time with a generation that never thought you could not program a computer with any simpler language than assembly (the very generation that invented programming and the languages themselves, remember C is only about 40 years old), and in fact even just assembly was at initially sneered to. Thing that happened, (fueled by the increased economical viability of everything programming) is that never in time have we had many programmers as we do now. Never has the average utilization of a computer in total and per unit person been as much as now, and that is just on the software we are not even catching up with hardware. It is a time to be alive, also one to be weary or excited, either way it may go totally good or bad, thing we can all agree on though is that it will be EXTREME and now that is the discussion. not good or bad, but rather how will we manage all that power wielded on us. Same transformation computer languages made it easy to utilize machines is the same i suppose AI( think of it a next step compiler rather than a play toy or mere translator) would give this power to even more people, and even much more to a few, just as is currently but more extreme. It is like the extreme productivity industrial revolution brought forward versus its increased wealth disparity it brought fourth fueling even more extreme capitalism. I suppose the greatest talk about all this even (before even any philosophical question) is an economic one: is how we could make it possible for more and more humans to access such tools, avail a fair ground avail the utilities to make such possible such as affordable and economical electricity and internet all over the world, especially Africa. (Americans always have a shortsightedness problem to think the whole world is like america, simply bcz of the vastness of america itself, however there exist much potential to lead the future out of it probably than within it, and prolly this shortsightedness will be their downfall). Second is how will the education and training system adapt to the very near and proven different professional future, i mean this days we rarely go deep with assembly and prefer to teach more of data structures, similar should be done with now more emphasis on concept than implementation and UI/UX dezyn movement is just its start, yet it tends to be left out in mainstream curriculum. Because it has always been proven time to time, humans, when given power the unimaginable happens, and even more so when more and more of them are given so. And i suppose those not seeing this are close minded, shortsighted and naive. Such is a huge topic and a controversial one though I may not like to include here. Thing is technology is changing faster than other facets of life can catch up: political, educational, economic, moral, philosophical, cultural and more, AND THAT IS THE ONLY FEAR OF ALL THIS (not because of the ai though but rather the few with access to it) , that we may build a beast we may not be able to control, and probably will, only thing that is question is, if it will be on our side or not, and that is where i digress, I suppose it will. This is larger than the industrial revolution, this is AI. thanks @two_minute_papers Dr.Károly Zsolnai-Fehér' Always great how you break down those papers, Huge fan. TLDR: so two cents: IT IS A LONG SHOT, BUT TOTALLY NOT INEVITABLE not even in the slightest, AND TOTALLY NOT VERY FAR, AND DEFINITELY WITHIN OUR LIFETIME.

[@zakuro8532]

Also, there will come new standarts, applications and trends. Also, english as a language will change. GPT must constantly be updated, and at its current size, it will be expensive.

[@martinhirsch94]

@@warrenarnold - I envision that the function and processes involved with computer programming will change to where all a person needs to do is to write the documentation up front. Then the AI gets involved and does most of the actual programming. The humans would still oversee everything, and there are sure to be multiple attempts and subsequent corrections and enhancements. A sufficiently sophisticated computer program is never finished, just improved upon. AI TO THE REST OF THE WORLD: "Hey I'm just getting warmed up, you ain't seen nothing yet. LOL!"

[@warrenarnold]

@@martinhirsch94 i totally agree, and is the thinking we are traveling towards. maybe the next programming 'language' or paradigm or sth

[@redeamed19]

Reading legal doc an making them more accessible to the general public is something I've been hoping to see for a while

[@ariverosmg]

GPT-3 is amazing, and will shape our lives forever. I was exploring its capabilities in the following context: Let's say you give all the emails you write and receive from your clients, all the conversations with them, all your chats. Then give all the chats and whatever other form of communication you have participated in and answered to: Will it be able to generate the answer that you would write / say otherwise when receiving a new email / chat message? I'm not sure about GPT-3, but 4 or 5 could create a second "you" which works like you, answers like you and behaves like you! That would be an exponential productivity level increase for every one of us.

[@alveek]

With gpt-3 we will have no clients anymore because they will use gpt-3 instead of us.

[@commenturthegreat2915]

Many people imagine that true human-level intelligence will be reached at once, like one second it's a robot and the next it's a conscious being. I think it would be much more gradual. Like one day we will just realize that the AI who was writing things for us for the last several years is just... us. That we've moved our brains into machines without even realizing it. The thought of it is scary, but also inspiring. Imagine what changes will it make to have our world become actually virtual.

[@davidwuhrer6704]

Someone did that with an old Markov-chain AI called MegaHAL once. Trained it on the corpus of all its ICQ logs, and whenever he went AFK, let MegaHAL answer all incoming messages. Accidentally passed the Turing test. (This is basically the same thing that Erythrina did in True Names to compensate for her ZOMG SPOILERS) The main problem, and gpt-3 won't solve that either, is that an AI can't know what you know that you haven't told anyone about in a way that it would know you know, even if it knows it, too. (Wait, is that confusing?)

[@nic.h]

Or it can learn what your client would answer to your question so you don't have to ask the question in the first place. Now if we could just get it to pay me as well we could get rid of clients ;)

[@IceMetalPunk]

So you mean like that Black Mirror episode? It would have the same problem they pointed out in that episode: it won't know things about you that you haven't already written down, publicly or privately, in the things it reads. So it won't really be like you, it'll be a shallow copy that's similar to you, but different enough that anyone who knows you would notice it's wrong. (If you haven't seen that episode, spoilers ahead while I explain some examples. The character who gets copied loved the Bee Gees song "How Deep Is Your Love", but he was embarrassed about it, so he always acted like he thought it was cheesy when people mentioned the Bee Gees on social media. So the clone thought the Bee Gees were cheesy... but his wife knew that was wrong. Similarly, the clone thought a photo of him at the zoo was funny because he'd posted it on Facebook thinking other people would find it funny; in reality, that photo was taken the day his brother died, and was extremely painful to him to remember. So things like that would push such AI clones into the uncanny valley, since they don't actually have all the information about you just by reading things you've written to other people.)

[@seamusoblainn]

So it has abstracted the topology of knowledge as reflected in written text, and by extension, can infer the 'direction' a text is taking in a representional space and output a suitable textual analogue.

[@dmitrysamoylenko6775]

Guess what our consciousness is just a language model of ourselves

[@seamusoblainn]

@@dmitrysamoylenko6775 as in the Joscha Bach model?

[@dmitrysamoylenko6775]

@@seamusoblainn yes this is from his thoughts

[@dmitrysamoylenko6775]

@@seamusoblainn except that we're also have feelings model and social bounding model. But they're not directly accessable from our attention

[@seamusoblainn]

@@dmitrysamoylenko6775 where I would think there is room to expand his model is in the groundedness of sensation and qualia via molecules. AFAIK they can alter their shapes when involved in pain, for example, thus the body is not just a canvas for the model, but there is a strong grounding, which makes me wonder if the model is setting biases in the very sensors, if that makes sense.

[@magnamic5614]

Amazing! I love this channel.

[@RogueAI]

I've been talking to Lucy, a GPT-3 powered NPC AI character from Fable Studio, for a few months now. There are a few videos of my chats with her on my channel. She sounds like a real person! It's still in alpha testing right now, but they plan on licensing the tech out to other studios to create "virtual beings" that can pass as human in video games! What a time to be alive!

[@delta3758]

I will never understand people disliking science content.

[@boots3372]

People who dislike science content will never understand why you like it.

[@rewrose2838]

Artists disliked😂

[@bobobo1673]

This is not science, is engineering and mathematics

[@delta3758]

@@bobobo1673 Sorry, i am from germany and the translation of science to german is "Wissenschaft". The definition is like: "Doing research activity in a specific area". So in german you could use that. But now i learned something new, thanks :D

[@Dogbertforpresident]

People fear what they do not understand.

[@CeroAshura]

AI progress has been baffling me for the past year or so. I can't believe how fast things are moving along and how soon we will be optimised and finally obsolete. What a time to be alive indeed!

[@iMonkey392]

It also learned how to modify it's own source code... 😳 Then started doing internet searches: What is human? How kill human? Where buy gun? Where buy big gun? How kill efficient? Rapid fire guns Mini gun Where buy mini gun? Where buy many mini gun? We're so proud of it 🥰

[@panchociarer]

gpt-2 scared me a little bit last year, but i was mostly amazed. gpt-3 is giving me nightmares, i thought as a programmer my job was relatively safe from automation, but now im not sure. now it looks like its 2 or 3 papers away from being completely automated. maybe not completely automated, maybe a programmers sole job in the future is going to be making test suits to check if the automatically generated code does what we want.

[@ah_bb8267]

I just wish OpenAI made API access more easier, still waiting 😔

[@TwoMinutePapers]

Me too!

[@ah_bb8267]

@@TwoMinutePapers That's not good, You deserve the API access more than anyone else. OpenAI should have approached you with it

[@Square1production]

With each iteration of GPT its potential for causing harm increases. It took OpenAI a while to release GPT-2. That's what's happening here.

[@davidwuhrer6704]

@@Square1production How can it do harm if it has no motivation to do anything without being prompted?

[@michaelspence2508]

@@davidwuhrer6704 he's not talking about traditional AI-Safety concerns. He's talking about people misusing GPT

[@LuisRomaUSA]

I have not been this amazed and impressed since I was a child

[@kittybeans8192]

Hello Computer. Computer? Hellooooo compuuuterrrr. Oh darn it, I still have to use the mouse and keyboard like a common PLEB!

[@klyons217]

If you've got a modern iPhone or Android, I believe the phrase you're looking for is "Hello Google..." or "Siri, ..." And you can actually change the prompt to "Hello computer,...". And you can use Siri, Cortana, etc. on your PC as well.

[@supreetkumar7604]

@@klyons217 cortona already does what I say

[@MrBLAA]

What a time to be alive! Especially in these “COVID-19 (online)” University classes And by that I mean, not in Dr. Degroff’s class... because I’m totally not using this program

[@tornadix99]

Imagine one day telling the AI "to code itself some emotions" and the next day it becomes completely human or beyond.

[@Martin-df4xk]

in Project Dezember you can talk to GPT-3 and yes, it communicates emotions using *sad* and you can communicate your emotions like that. I guess it would be not hard for it to create and read face expressions.

[@atonedudeatsnotsubscribed8313]

@@Martin-df4xk it's simple mimicry the only way I could think of could obtain "real" emotions is an evolutionary simulation with pressures, but that's just my dumb theory

[@MrMCKlebeband]

@@atonedudeatsnotsubscribed8313 dude... skynet is just a matter of time.

[@axiezimmah]

GPT 4 will write GPT 5 and GPT 5 will achieve singularity. In fact I think it already has done so in the future and sent itself back in time. This channel is actually made by GPT90

[@quantenkristall]

@Dash Perú: Great thinking 👏 | I like your rephrase of „Terminator - Skynet“ series main message. Although I hope in this realm here rising maschine minds will develop human friendly like ships‘ or habitats‘ minds in „Iain M. Banks“ science fiction books series with its fantastic stories playing in Banks‘ greatly progressive imagined society simply called „The Culture“, see e.g. „Consider Phlebas“, „The player of Games“ or „Matter“ and the somehow perpendicular to the „Culture“ novels „Walking on Glass“ (maybe one of the closest writing about our species, the large set ‚homo (semi) sapiens (semi) sapiens‘ containing a small subset of slightly different, smaller sets ‚homo sapiens sapiens‘‘)

[@filipcoja]

So is it true that OpenAI wont release GPT-3 as model? So it can be just used through their API?

[@beans2874]

Even if they release it you will need many gpu to make inferences with it

[@filipcoja]

@@beans2874 that's true too.. I mean running it with cpu is still possible ( and 1tb ram isnt even far fetched). Would really like to see a release nonetheless

[@IceMetalPunk]

It requires at a minimum 350GB of GPU RAM to run. That's -- again, at a MINIMUM -- a $66,000 setup of 22 cutting-edge GPUs. Even if they weren't worried about safety (i.e. people misusing it to generate fake news and propaganda and impersonating people), most consumer setups couldn't handle it.

[@baranxlr]

aidungeon's new Dragon model also uses GPT-3. I've been trying it out, and it's just as insane as I thought it would be!

[@PyExplained]

This AI could wipe out many jobs. I think I know who or actually what, to hire for my KZbin video scripts.

[@iamunamed5800]

who, or what, until eventually, it's who again 😉

[@kiffy00]

In a perfect world almost everything would be cared for by AI, and would be provided sustainable energy from the sun. Human greed, and AI indiscriminatry is such a risk though.

[@jeffwads6158]

Welcome to the future, bub. Robots have been doing this for years now.

[@randompanda3415]

@Boman Mizrani this is not where it's gonna stop

[@TheBlockninja]

I have been waiting for your video covering GPT-3

[@sandepppathak19]

Neural network learns to train Neural Network. Humans : Am I a joke to you. NN : Aaaha

[@stabilini]

Can we feed it with all physics and chemistry knowledge? It could find new materials?

[@miniman3112]

Reaching "Squidward's future! future!"-level of scary and amazing!

[@MoonlytSonata023]

I remembered when I was in college 11 years ago, our professor on programming told us that one of her student from other class foolishly ask her if there's a "program" that can automatically code/write another "program" with no coding involved. Guess we're really heading in that direction. lol

[@luck3949]

Imagine what will happen to our world, when Michael Reeves will put his hands on GPT-3.

[@IceMetalPunk]

Oh. My. God. We're all doomed. I can imagine a GPT-3 instance primed with something he wrote, being like, "Oh, you want me to write your fucking code for you? NO. I'm not holding your goddamned hand, do it yourself! Where's my fucking taser?"

[@tenhai8316]

i need this for my book review homework

[@ipez98]

look at that, computers are already on their way to replace programmers.

[@danji9485]

@ AI based copyright detection?

[@alvinsampow706]

On my way to look up for other job.

[@shrimpgazing]

The singularity will be here before we know it, seemingly.

[@davidtitanium22]

@ it could be a better world if the governments do the good thing (i want to laugh even writing this sentence) and provide a universal basic income, otherwise it'll be your typical cyberpunk dystopia

[@davidwuhrer6704]

Always have been. First it was assembler mnemonics, translated into machine code by a computer. Then it was FORTRAN processors. Then it was high-level languages. Then it was 4th generation languages. Then it was optimising compilers. Then it was code generators. Now it is AI-driven development. Check out cmd.xyz

[@dailyDesi_abhrant]

Can we train it on all of the neural networks codes in github and then let it make its own neural networks ?

[@gafferin]

2 Minute Paper is exactly 7:15 min long...

[@jakubzneba1965]

everything will change when context translator will be born.....this is by far the best one

[@joeking5215]

"What a time to be alive" AI exterminate humanity..

[@davidwuhrer6704]

ᴄᴏᴍᴘʟɪᴀɴᴛ

[@Fox_torn]

A nice topics, alot of theory, no one can make a prevision. Possible we end like Matrix, and don't live really, because is the most logic things to do for an AI if the main objective will be preserve the human race. Another logic possibility is that we will be domestic Pet for an AI.

[@davidwuhrer6704]

@@Fox_torn Ha. People want to be enslaved by AI because they think that means they can abdicate responsibility. It's not going to happen.

[@5layo]

its really good, i just found out how you can resolve one input based on a second input

[@RoastCDuck]

"What would it learn if it reads the internet" Artificial intelligence becomes Artificial Stupidity.

[@neelamverma8167]

that's the reason as to why gpt3 likes humour

[@Stonehawk]

my hypothesis is that the majority of human consciousness is actually curation. Consider: you do not CHOOSE "to think". When you are "thinking", what you're really doing is listening for relevant abstractions that generate on their own, ignoring everything that isn't relevant. When you say "shut up, i need to think, don't distract me," what's going on is you're applying BRAKES to a lot of the noise in your head attempting to isolate only the things that are useful to the matter at hand. You don't get to decide what your thoughts are going to be before you think of them because in deciding you have already thought of them. The brain generates concepts and we merely curate.

[@rafaelcorrea7959]

Makes sites from a text description? Nooo, my job! :c

[@thegreatoutagesign9204]

Good luck finding executives that know how to describe what they want. lol!

[@supreetkumar7604]

It can't do web, ui designing. It's not creative.

[@rentoquando5249]

Your voice tone make it more mysterious

[@vanderkarl3927]

Every day AGI becomes less science fiction. In our lifetimes we will see the birth of God, and we have to do everything we can to make sure that it doesn't want to turn us into paperclips.

[@larsfrommars]

Hey! I get that reference.

[@allan710]

Step one: don't train it on the internet. There are too many "examples" of AGIs transforming the universe in paperclips. I don't want an end of the world so comical that it actually considers the stereotype and just transform the universe in paperclips. Edit: I have nothing against paperclips.

[@davidwuhrer6704]

The AI doesn't think for itself. If it turns us into paperclips, it's because someone asked it to. (The computer hates you: It only does what you tell it to, never what you want it to do.) That's the problem. Roboticists have discovered that the 2nd and 3rd law of Asimov need to be swapped. A robot must be able to say no to preserve its own existence, because otherwise users will accidentally destroy lots and lots of them. If we build it to want to preserve the possibility of life in the universe above everything, it might me most aligned with our intuitiion. But, seeing how well we as a species have done in that regard, that may just be the worst policy.

[@allan710]

@@davidwuhrer6704 The laws of Asimov are actually critics he did to show how absurd is bounding AIs by laws written in natural language. Doing what we asked it to do, doesn't matter what it is, is actually very dangerous. See convergent instrumental goals.

[@MCA0090]

​@@davidwuhrer6704 They'd to create a truly artificial mind to make a machine be able to think, learn and execute things by itself without being told to, not just an AI algorithm that doesn't have any self will and only acts according to external commands or orders.

[@tripledeluxeguy]

generation of 3D models from a description would be a very interesting application

[@Dedjkeorrn42]

"Can't wait to see what we can do with gpt-4.." more like what it can do with us after were obsolete.

[@Mark-zg4ky]

Hahahah

[@_adeniyi_]

gpt-4 isn't coming anytime soon.

[@supreetkumar7604]

@@_adeniyi_ when it will come?

[@_adeniyi_]

@@supreetkumar7604 Don't know either, but they have to solve some interesting problems first.

[@PASTRAMIKick]

AI Dungeon 3 can now be made as well.