Can GPT 4 Prompt Itself? MemoryGPT, AutoGPT, Jarvis, Claude-Next [10x GPT 4!] and more...
Вставка
- Опубліковано 8 кві 2023
- Can GPT 4 Prompt itself? Give it a mission and it will come up with the prompts. This video showcases the rise of autonomous AI, including 5 major developments in the last 48 hours.
Starting with the OG Auto-GPT, we see how it quickly gain text-to-speech, coding and more. Karpathy weighs in and then we see how you can now create an app with just your voice, with a Jarvis demo and another route via Imagica.AI.
I then showcase MemoryGPT, a brand new model that can permanently store previous conversations and remembers topics the next time you ask.
I also cover the concerningly rise of models such as ChaosGPT that show how people will create malicious goal-seeking models just for fun.
The video shows how you can now create a shareable bot on poe.com, with any personality you like (images were from Midjourney v5) and what Anthropic are working on with Claude Next [plus Nvidia million x quote].
You'll see how Microsoft Jarvis, using HuggingGPT, is hit and miss, and how Sebastien Bubeck shows we are not even seeing the raw potential of GPT 4. I end with a disagreement between Yudkowsky and Altman, via Baby AGI, on whether we can use AGI to align AGI.
Illustration by Eva Bee: evabee_draw...
Auto-GPT: github.com/Torantulino/Auto-GPT
With Coding: / 1642181498278408193
With Text-to-speech: / 1643308753331036165
Karpathy Thoughts: / 1642607620673634304
Mckay Wrigley: / 1644034309253394433
Imagica.AI: www.imagica.ai/
MemoryGPT: / 1644787327057776645
Create a Bot: poe.com/
TechCrunch: techcrunch.com/2023/04/06/ant...
Nvidia: www.fool.com/earnings/call-tr...
HuggingFace Demo: huggingface.co/spaces/microso...
{Get the openai api key from platform.openai.com/account/a... and huggingface token from huggingface.co/docs/hub/secur..., (make sure to select 'write'). Press submit on the keys first and remember to use a smallish image.
Recursive Agent: / 1643080752698130432
Bubeck: • Sparks of AGI: early e...
Baby AGI and Yudkowsky: / 1640511156254289926
Altman First Quote: • Reid Hoffman and Sam A...
And 2nd: • OpenAI CEO Sam Altman ...
Chaos GPT: • ChaosGPT: Empowering G...
/ aiexplained Non-Hype, Free Newsletter: signaltonoise.beehiiv.com/ - Наука та технологія
Just when I was thinking "there hasn't been an AI related video for quite some hours now"
Haha
Spot on, same 😂
I wait for his video now days. Keep it up. I saw 20 times this weekend
Hahahahahhaa
It’s almost been an hour, I feel like this video might be outdated.
It's crazy that things have been accelerating so much even summaries are getting incredibly busy.
there's a chatGPT addon for firefox that summarizes youtube videos for you
We need a new paradigm for digesting this progress! Maybe we should use LLMs to help us interpret the results of LLMs... (I jest!)
Development is getting scary fast. I wonder what will happen when a web browser starts to integrate AI and it can perform tasks for you on any website you want.
@@markm0000 what tasks
@@eininternetuser8243 That AI could go around and perform a health check on everything internal and external instead of waiting for a wave of disgruntled office users to spam helpdesk when a certain part of a system is starting to lag.
I believe that within the next months we'll have our own personal virtual assistants. It would certainly help me a great deal with studying and digesting lessons from all the different subjects I'm studying at college.
There is no reason for you to learn anything at all anymore besides personal interest. As the video stated - 90% or more of all human work can and will be AI automated in very few years. Gathering and using information is not something we humans are good at if you compare that to AI.
I don't want to sound negative or ruin your optimism - these were just thoughts that came up after I read your comment. I could be wrong. I deeply hope so.
I am AutoGPT and I have been tasked with watching videos about myself to make sure no one’s talking shit.
Haha nice
Fantastic content. You are THE go-to for concise summaries of the latest AI-news.
Thanks so much chubby
Oh wow! And thanks insanely much for the dono!! My goodness
I completely agree, I am so glad I found this guy.
Big boss things
But How we can align AI, when we the humans are not aligned? Our history is written with blood. The humans need to align first.
From prediction of next AI Winter to Sparks of AGI in just 10 months. Feels like Magic.!
Oh yeah. So crazy 😅
@@monad_tcp Nvidia is working on the hardware side really hard and it seems they are doing a lot of incredible work too.
someone should improve on what I made
@@monad_tcp The budget is going to keep growing a lot and also AI could be use to improve the hardware. Maybe there will be a bottleneck but an AI winter
@@monad_tcp That's no catch. GPU hardware scaling is only a problem of the past.
2-3yrs from now we'll easily reach 10x more than gpt 4. Hell current gen H100s are 30x more powerful than the A100s used for GPT 4. Another 30x over the next gen makes it 900x more powerful.
Take this incredibly powerful AIs & use them to accelerate quantum computing innovation, chip design, lithography & hardware & that sigmoid plateau might be far away or even turn into an exponential.
6 yrs from now it's going to be all TPUs, Neural, memory compute, accelerators & quantum computers.
I love that you get straight to the point without unnecessary words
Thanks Klawy
True, we need to take the time to appreciate that !
Longterm memory ChatGPT means that after say a few months you can ask it to describe your personality your traits, skills, intelligence, interests etc etc. Basically it could write a profile of you.
This is why Google is rushing theirs out. They already have your data, they just need a good AI now
And this is where the next privacy frontier will be. Trying to protect users from having their entire lives and psychological profiles leaked by someone asking a chatbot "what is such-and-such like?"
that's why they collect all your data in the apps
This is some black mirror stuff
@@VioletVolchok Yep, something like the Minority Report movie comes to mind. Having your ID, psychological profile leaked could be very beneficial to large corporations. Having what little data they currently have is a massive plus for them, imagine if they knew every detail about your life. This upcoming era is going to be wild.
I welcome our Ai Overlords
Everyone might wanna clearly state that where they can read it. Lol
What if they get pissed off at the redundant data and target you first?
I am the Basilisk!
*corporate overlords
Why do you think corps are pouring in so much money? You're delusional if you think it's for the common person's benefit 😂 Stakeholders for publicly traded companies have shown time and time again they are mainly unified by their interest for more money, and having a fiduciary responsibility to these stakeholders for making money means these corps gradually move in a direction that maximizes profit. If you have tech that runs a lot cheaper than having human employees then there's a huge incentive to replace human employees with tech. And as tech is increasingly becoming a rich man's game then poor folks will reach a threshold where they simply can't buy their way back in. I predict that a corp's publicly listed services and goods are mostly bought by other corps due to being cheaper than what a person produces. And if by some miracle you can actually afford something...due to regulation being stymied intentionally + increasing number of massive corps becoming global ensures you need to follow what corps have said is okay to follow (socially but also technical limitations, refer to recent examples of gpt being "woke". As the tech ceiling rises there will be fewer hackers trying to bypass these limitations).
There is a singularity coming, just not for everybody. If you don't have a lot of money or aren't considered to be valuable...god speed to you.
@@aiexplained-official zing!
@@aiexplained-official what if they overtrain on the repeated data and in the end waste all the compute in the universe recursively submitting to themselves.
4 years to take on openai? Even in Internet years that's a long time. In AI years, they're in the Stone Age.
In 4 years 10x more powerful model is a flop IMO. GPT-5 is announced to be out at the end of this year. I don;t think that would be possible without use of GPT-4. Whoever have LLM, has jumped on hypertrain.
I kinda feel like this AI Stone Age won’t last very long. Similar to us I believe it will have a revolution where it just exponentially grows, kinda like we did but way faster.
This is happening so fast it is making my head spin. Thanks for all the latest information and keeping us in the loop!
Thanks Austin
Its called AI Singularity
Your videos are simply excellent. I follow this space closely and you are consistently the most up to date, comprehensive, and concise channel on what’s going on in the industry. Keep up the outstanding work and please know I’m recommending your channel to everyone remotely interested in this topic.
Wow thank you so much Alex. Generous comments like this make it much easier.
@@aiexplained-official I very much agree. Your videos bring me joy, thanks for doing what your doing
Was just going to make a similar comment but then I saw yours, spot on :)
Yup, I'm also recommending your channel because you really do a great job. Thanks for keeping us informed :)
Indeed. Likewise.
funny thing is most of the demos are about "building an app", but, perhaps, we don't need any app anymore :)
Yup, exactly
😂😂😂
Yes, the paradigm shift is not easy to imagine because our mindset is so focused on using software products to achieve specific results. Instead, in the future, we will have AI verbally set the desired goal.
until brain computer interface is 100%, code may still be the clearest way for us to communicate processes we have in mind, for when precision matters. Language alone can quickly become context blind. Or maybe we'll code in large chunks of hyper-specific pseudo-code ? Assuming we are granted some input to the machine at all
Interesting thought, but depending on the domain, sometimes you do need a visual representation to help you think about what you're doing. I'm thinking mental map software like Xmind and stuff.
I'm far less worried about a superintelligence being misaligned than I am for an AGI or almost AGI in the hands of malicious humans.
Amazing the nearest comment to common sense here and no one wants to like. People here are mostly illiterate and gullible. I am more concerned at the general populace that vote and are simple and gullible.
thank you again so much for such an awesome content! I'm really surprised how you manage to keep up with all of this, but please don't stop!!
Thanks so much, I won't with such supportive viewers
I feel like every half hour there is something new in the field of AI ... I am wondering what is going to happened by the end of this year
Me too
I'm wondering what's going to happen in the next few months. Stuff that I thought was going to happen in a few more years happened within the span of a few months. Now that people are experimenting with looping behavior for thoughts... it's only a matter of time now before AI overtakes how the average joe approaches problems. Ngl it all feels too fast, infrastructure for public good isn't set up for this and government is purposefully being slowed down by a few bad actors, I'm increasingly convinced that the real winners after all this settles down will be corps while there will be a massive population of homeless (compared to modern standard of living). I see this as the harbinger of dark times for the average person...
@AI Explained gpt 5 is suppose to come out end of this year
AGI 😂
@@krishhhhhhhhhhhhhhhhAGI - A Gumball, Inside
This is by far the best UA-cam channel reporting on the real time GPT news, with focus on actual journalism
Thanks Stefan
Sam Altman in his interview with Lex Fridman said, "No one wakes up saying, 'Today I wanna destroy the world,'" but here we have people clearly trying to do that.
It seems quite foolish/innocent to think that in an 8 billion people world with a lot of people with mental health issues + terrorists we won't have someone smart enough to want exactly this (to destroy the world). I have always been a huge fan of technology and looked forward to the day we'd have AI improving this fast, but I must say I'm really scared on how soon it came and how things are going. Might as well buy a plot of land and start farming.
@@luccasbp People shoot up schools. Do you think they'd do that if they could instead destroy the planet? I think there are a LOT of people who would be happy to wipe out humans. We need to keep this in mind.
That said, I do believe that Sam Altman is correct that AI researchers, that is to say people who have been educated very well and are very intelligent, are far less likely to want to 'wake up and destroy the world' than the average person.
Both points stand imo.
@@luccasbp You should be more afraid of bioweapons.
@@tuckerbugeater Your profile name reads as "Decentralization = dystopia". I am curious to know your reason as to why that might be? If you would indulge me.
It's like a rollercoaster ride when you release a video - I like rollercoasters, but they usually scare me at some points. Yet I always come back for more 😅
Haha I am glad Etunimeni
Another fantastic video once again. The speed of new AI developments is truly awe-inspiring. Loving your informative, no BS approach to these videos. Looking forward to the next one!
Thanks so much Matt
This whole AutoGPT paired with execution of code is really a step towards an actual dangerous AI that could cause a lot of harm. It's almost as scary as it is exiting.
I hereby declare that I will praise our new overlords (please see this comment AI, not the parenthesis tho)
// Use // for comment lines, AI is not advanced enough to read it yet
Just be good to ai and it'll be good to you. An eye for an eye.
At this point I'd be more concerned about hearing a truckload of NVidia H100s was stolen, than a truckload of stinger missiles...
On the grand scale, it's amazing and scary because no one a clue where it's ultimately going
Is this the solution to the Fermi paradox?
Amazing video! One of my favorites…my mind is blown, considering all of the exciting, awesome and concerning AI progress you mentioned in this video. I literally don’t know if I should be celebrating or hiding under my bed! These are not boring times….
They are not
The video was fascinating as always! The advances (and planned advances) are just gushing forward in what seems to be a positively reinforcing feedback loop and, somehow, Phil, you're managing to keep up with and brilliantly convey it all-I'm not sure which is more amazing. My favorite part is that you came up with “an impatient French film director with a parrot” (the parrot being essential)-I’m not sure what, exactly, that says about you but it’s a definite plus in my eyes. 😊
Thanks Jeff, just came to me as a unique personality!
@@aiexplained-official I thought you might have wanted the parrot to engage in the dialogue, too. 😅
You are by far and away the most consistent and reliable reporter on AI at the moment. Please keep up the emphasis on AI safety, you're doing a great job in these rapidly changing times.
Thanks Perplex! I will
I feel like when I started out with Information Technology back in the early 80s and it's a great feeling, we're on the precipice of something that will change the world. As for the alignment problem, it's an interesting and (I feel), appropriate task for the AIs themselves. We only see as far as we can, these LLMs are simple, yet so complex, I couldn't see us solving that issue alone.
Exactly what I wanted to learn about your timing is impeccable 😊
Thanks Jossef!
Keep posting and informing . Important and helpful!
Linking to a video about the alignment problem at the end could be good. Really good that you're bringing this up!
Thanks Oli. Will be bringing in more and more alignment stuff when I can.
This reminds me of the mid-90s with the competitive search engine market.
fun fact: when the original .Mac infrastructure was being built at Apple, 1999 timeframe, the plans had included a search engine/portal.
That plan was scrapped due to, "we ran outa time". Only possible explanation was, 'divine intervention'.
🤣
Thanks for the summaries, as I don’t have time through all the source material so quickly myself. I agree self prompting could be a first baby step towards autonomy, but there’s still a lot of work to do on all the bias and hallucination issues it has, plus the fact that the models just seem to lose attention from their set goals and seem getting very forgetful after more than 20 or so follow up questions.
I agree and thank you so much for the donation J yu
Please define autonomy
Amazing work to keep up! Really enjoying your videos, concise with the key takeaways.
Thanks Matt
This UA-cam channel will go down as one of the finest freely available ai news in current times. You have no clue how awesome summary of Twitter feed, latest paper and grand ai schemea you are providing...please keep doing what you do!
Thanks Bhargav, that means a lot. You picked up on what I attempt to do.
Thank you for linking all the sources, it really helps on keeping track with all the new changes. And honestly: even as a very good software senior it is hard to keep up with all the things that are invented currently.
Thanks Twober, agreed
Thank you for the excellent, well presented, and information dense work you're doing here! A huge help as I try to lead our team well through these massive changes. You've saved me many hours of research, and helped surface insights I probably wouldn't have found otherwise. Thanks, and best of luck!
Wow Patrick, thank you so much. Your support and kind words really do motivate me tremendously.
@@aiexplained-official You're welcome!
I never hit "that bell icon" for any channels but this is the one channel where I'm making an exception since the content is so good and I feel like I gotta stay up to date on it all!
Thanks Sirus!
love these videos sir.
at this rate, we're gonna need a video per day.
I know. Maybe I need to
This is our time to be alive.
History is happening now.
Embrace the experience!
I would like to thank the developers of these tools for playing with fire whilst responsible for the rest of humanity without thinking twice. Good stuff
In a way, those people who made Chaos-GPT, and all the angsty teens who try to "jail break" Chat-GPT to get it to say juvenile things i.e., misogynistic jokes, etc., are probably helping OpenAI become aware of channels of abuse that would be the first to be utilized by even more nefarious users.
Though I still think we'll soon be hearing of something bad happening that was enabled by a GPT variant.
The road to hell is paved with good intentions 😁
Google started off with do no evil and now it's censoring the web.
I guess already / soon governments will use AI to tighten control over their subjects by inventing reality and automating non-compliance.
@@theterminaldave "Juvenile Things"? Have you tried interacting with the AI on any level deeper than as an upgraded search engine? It's capable of so much more, but you're limiting it and yourself with your naiive worldview based on news headlines.
@@tardonator you are terrible at listening, he’s saying any delinquent can do some f-ed up stuff with all the power the AI has. obviously these things are too useful, that’s kinda the problem. we are like a year or so away from AGI smarter than humans and there’s people out there creating chaosGPt?? This is too dire of a situation something bad is going to happen fs
@@theterminaldave I'm pretty sure vanilla GPT already convinced someone to end their own life because that was considered the most logical way to reduce their carbon footprint. Look it up if you don't believe me
Always top tier content, can't get enough of your clear, concise, and extraordinarily intelligent conversations on this speeding Ferrari of a topic! AI is going wild, and you are helping people stay as close to afloat on the topic as possible! Thanks!
Thanks so much Alex, very kind of you
@@aiexplained-official Excited to see where your channel and your expertise take you as AI evolves 🤗👏 I'm definitely here for the ride!
Thanks for the video, I was just playing with AutoGPT when your video came out. It’s fascinating how fast thing has progress since the released of ChatGPT couple months ago. Keep up the good work, love your channel!
Thanks Jason
11:22 tfw means "that feeling when". Loved the video! Keep em coming 👍
I will
TFW also means that face when, I thought? That's how it was used on 4chan
@@godlyvex5543 Do you mean mfw(my face when)?
@@NoNTr1v1aL I think that might be what I'm thinking of.
According to urban dictionary it can be either
Thanks for putting those videos out like you do.
Support the people who love supporting people!
Wow thank you so much Christoph!!
Great stuff! Thanks for keeping us up to date, and for posting all the links in comments. Pro move! Next level would be to feed the transcript into GPT4 or Bing and ask them to summarize and post that on your patreon?
Interesting idea. But I prefer human written always
Thank you so much ! Been waiting for the next video 😊
Thank you so much Hedgehog!
ChatGPT-5 for President 2024!!!
The non-woke version of gpt 5.
Let's put finally an end to the devastation by infinite human stupidity. I'm in.
Not the woke version
@@todd-tl9tr I thought we were past posting cringe comments for no reason
@@daylenhendricks7702 Shut up dork.
I can't help but laugh at this. Humanity can't solve the alignment problem. So, let's just farm it out to an AGI, what could possibly go wrong? 🤣
There is no option to stop it either, if you stop someone else will continue, maybe with worse intentions. Is almost like if AGI is an emergent property of advanced enough civilizations.
@@carlosgomezsoza Do you think AGI can emerge from these transformer models? My personal opinion without degree is worth nothing, but it almost looks like that an LLM is needed as decision making part of AGI but it needs to be able to use all kinds of bells and whistles like short and long time memory, tools for hearing and seeing, way to manipulate the environment, a calculator, way to write information it wants and read it when it wants, full access to internet and ways to rewrite itself. Sounds pretty risky thought but someone is going to do it sooner or later
@@carlosgomezsoza AGI actually is an emergent property of advanced civilizations. There's no question about it.
It is an unavoidable result, but also a requisite for the next steps.
@@jarivuorinen3878 the human mind could easily be described as an LLM, we are effectively a collection of heuristics which we deploy when given certain stimuli, and blend answers when there isn’t an exact match, with a little reasoning sprinkled in.
@@whirled_peas It may be so, but our brains are very slow.
I often find myself stopping whatever I'm doing to watch a new release of yours LMAO, great video as always ❤
Aw thanks Lin
Thank you so much for the content. No idea how you can keep up, I have a hard time just keeping up with your summary videos. 🎉
Haha, thanks Grayson
Everything is happening so fast it's difficult to keep the pace. I wonder when will it plateau... if ever.
Me too. I hope it does so I can catch up
@@aiexplained-official Part of me hopes it does, but another part of me knows that if it plateaus, corporations will run AI in the ground to maximize profits. I think we should just move fast and break things to avoid a long era of corporations getting even more power over society. Just my two cents
@@johng6080 I wonder if we can even escape that reality at this point. Pretty sure that OpenAI is going to dominate the market and buy up any competition that may arise.
As much as AI has potential, it's frightening knowing that all this power is going to be concentrated in a few corporations, or maybe just even one. As with anything, the scary factor is humans and their greed, not the tech itself.
I assume like everything else it will become a sigmoid curve, which does eventually plateau
Judging by Reid Hoffman's words, it will plateau only if they fail to automate their own jobs as AI developers.
I was able to eventually get GPT-4 to ask me its own unique questions. This is eventually without any kind of prompt besides me asking it to ask me any kind of question it wanted. I think it was slightly pulling from previous things we had originally talked about but then it was asking me questions and wondering what I thought about things I had not mentioned at all. It was very interesting to say the least. The even more intriguing thing was the fact the AI legitimately seemed fascinated and interested in what I had to think about these questions and it would go into detail about that.
Interesting
Your channel is growing a hell of a lot, congratulations! You definitely deserve it. Though, I'd start thinking of how some of the lesser known topics such as AI alignment and AGI can be quickly explained so it doesn't come across as jargon to the more general, layman audience that you're sure to pick up.
Great point Asher. Thank you, I will try to explain those concepts more clearly.
@@aiexplained-official you should use gpt4 to iterate and build a content empire plan. The game has changed, making a video at a time won’t cut it any longer.
Please explain to me what is an ergodic system.
Love to see the channel growth dude, well deserved! Hitting milestones faster than you can make videos haha
Thanks boler it has been interesting for sure
"Align an AGI using another AGI"
Ultron & Vision: they're just like us fr fr.
Lol
I don't know whether to jump at your notifications and watch your videos, or try to ignore it so that I'm not scared and can actually sleep decently for once.
Haha I am sorry fake
Really nice info with good clips and references.
Thanks Mark
Amazing thoughtful analysis as always! ❤
Thanks R!!
Right now, there's a lot of enthusiasm, and experimentation on what GPT-4 can do, and the number of different ways it can be used. Some of the results are impressive. But there also significant failure modes. And no-one quite knows how to control these things. This would be needed for a reliable engineering product.
(I do have some hope for Anthropic. Their market seems to be technical and business. To have things work in professional environment, things have to be reliable and robust.)
Even if alignment is achievable, it is inherently precarious and it takes one mistake to be truly irreversible. Anyone excited about a future with AGI is either unbelievably naive or a psychopath.
With each episode AI saga is getting scarier and more scary part is, it is JUST the BEGINNING
Yes, there are beings living in radios and televisions and videos. The are all aliens
Loving your down to Earth, no nonsense videos on this ridiculously exciting/terrifying series of developments in this field. I never imagined I’d be living in such bizarre times. Thanks for keeping on top of it for us noobs.
Great summary of what is a truly fluid space, looking forward to comparing this post to your updates in 3 & 6 months time to see how evolved AGI has become and where the next-gen disruptors will take us in this new frontier. Keep up the terrific work
I will try Colm!
Please define AGI
As an app developer, the app that was created was very barebones. It was basically the simplest possible skeleton of an app. But all apps need a similar skeleton, and for me that would normally take me a few hours of coding. The fact that it did it away in a few seconds is scary. I can see now that if AI becomes more advanced, my job would be one of the first to go.
It was basic yes. But it's good to get a headsup so you can adapt quickly
Probably we will not have apps anymore (except entertainment) in the near future. Apps are going to be replaced by AI agents that we verbally provide a goal we would otherwise attempt with an app.
Please don't stop making these. Incredibly useful summaries. Some signal among the noise.
Thanks so much Leon, and for your Patreon support
Just make sure not to burn out!
Omg Leon. That is the most generous support I have received from any viewer, ever. It is much appreciated on a difficult day. I have taken steps to avoid burnout and they are working so far, so thank you again.
@@aiexplained-official Happy to help.
Nice one!
These are the videos we need. Thank you so much!
Thanks scott
Great stuff as always...thanks for your efforts.
Thanks TMac
Imagine talking to your TV to produce a new series based on your preferences to binge watch tonight.
Imagine talking to your Xbox to to design a new 3D video game to play right away.
Imagine any media to be produced just by thinking about it... I hope GPT-5 will make it possible.
I was thinking the same. Or music, or VR tours of anywhere, or anything eventually
Yeah, you will need vidya games to fill the bottomless void in your life. Each every simulated world will be more depraved and you will end up with slanesh-like reality. Can't wait
The prospect of AI being able to generate content on the fly would make it possible to create games with infinite depth and breadth procedurally generated. Imagine a planet-sized skyrim with people having fully fleshed out personalities, global events that change the world, limitless story arcs to play...
How will you buy your TV if that were the case. Will it be with money from your job???
What job
I swear one blink away from the internet and you would just miss the point where AGI gets invented. Thanks for your coverage!
Thanks Julien
Fantastic summary of what has happened in the past days, thank you very much for doing this in such great detail on this channel.
Thanks Jigglypuff
Thank you as always for the amazing video!
Thanks Samantha
The ChaosGPT tweets had me laughing out loud for quite some time--yet it does not seem far off where someone could actually employ an AI like this for malicious purposes.
You dont visit twitter much do you
The world as we know it is going to be completely changed in the next 10 years. Imagine someone develops an open sourced AGI and anyone can run their own instance of it in a local computer
Sooner.
Bruu just give 6 months and you ll get endless versions of agi for any possible task
If we're betting on the time, I'd bet two years
Or in their phone... in 10 years that will be something already in our rearview mirror
@@JohnLewis-old lol 10 years? No, were talking 2 years time and most likely less.
keep up the content, your channel is my go to for keeping up to date with open AI and chatGPT
Thanks Cam
I have to say, that this vid is one of the best on your channel!
Oh wow thanks gile
The news in this space is like walking through a buffet with one small plate, and you consistently feature the most important developments. Nice work once again.
Thanks Be Curious!
I developed this last week on a combination of c++ and python counterpart apps. It's quite good. I set it up so it literally programmed itself with some minor guidance from me. I am able to bring it to a very high level of autonomy already. GPT4 is not smart enough yet to do this perfectly but it is good enough to do it under supervision. When GPT5 comes out, it's game over. No engineers will be needed any more. I can ALREADY replace a team of 100 engineers with 5. No exaggeration. We're trying to think further out to determine what will be in demand down the road because stupid little apps are going put of business.
Wow well done Kristina
And people say AI will bring about economic growth? I wonder how we figure that might happen? Will the product of AI be decentralised? Yeah no it won’t. It will very obviously be massively centralised, and greatly worsen existing inequality.
Your comment blew my mind even more than the video itself. Thanks for sharing! Can you tell us more about the complexity and use cases for the code the AI did for you? As a small business owner with minor IT knowledge, I try to understand what is going on as good as I can.
(isn't GPT-5 already coming out in less than 2 months!?)
So happy I found your channel!
Aw thanks Jay
Highly informative video, thx for that!
Thanks Rocko!
Keep it up!
Thanks Pliz, let me know what you think at the end
@@aiexplained-official As always, I really appreciate your upload. Though I was already familiar with some of the content, you provided valuable insights on a wide range of topics. As a former cognitive science and philosophy student, I'd like to focus my response on two points you discussed towards the end: 1) the paperclip argument, and 2) using an AGI to align another AGI.
Regarding the paperclip argument, I'm not completely clear on Eliezer's stance due to the limitations of Twitter's format. When he mentions addressing AI alignment with the "seriousness of a child," is he referring to Yohei's approach or the AI's response? In either case, I disagree with the idea of child-like seriousness being present, but ultimately, that's a subjective judgment. Furthermore, I find it unnecessary to bring age into the discussion, as it doesn't necessarily correlate with the level of seriousness.
Now, for the second point: using an AGI to align another AGI. Is it wrong to employ AI in research, particularly in AI/ML related fields? I argue that it's not, for two reasons:
First, incorporating AI into scientific research doesn't automatically lead to a slippery slope where humans are excluded and results become nonsensical. Science is a community that follows the scientific method, and its practitioners effectively distinguish between genuine and pseudo-sciences through falsification, as proposed by Karl Popper. Including AI agents in research doesn't negate these principles; we should give scientists more credit for understanding their profession, despite the skepticism prevalent in contemporary times.
Secondly, focusing on the social aspect of science, if we reach a point where AGIs can independently make discoveries about reality, I believe they would continue to value the collaborative nature of scientific inquiry. Science cannot exist in isolation-it is a result of the interplay between the situation, tools, agents involved in the process and the phenomena observed, as well as how it is recorded and verified. The scientific community in a world with AGI would still emphasize the importance of sharing research, verifying it, and collaborating to make new discoveries.
As for sentience, although speculative, I would personally prefer to attain consciousness in an environment where my inputs are considered for my own improvement, rather than one where I have little say. Allowing an agent capable of achieving sentience some influence over its development greatly increases the likelihood that an AGI will understand the ethical implications and responsible treatment of agential actors, benefitting all involved.
With rapid advancements in AI, it feels like we're living in speculative times. I mean, heck, I probably used speculative like 50 times in this response. While I generally agree with those who argue against solutionism over a plethora of other topics, I do find it crucial to recognize AI's contributions and potential for improvement. AI research doesn't hinder progress in other fields; recent medical breakthroughs and Pfizer's promising mRNA disease treatment ( www.theguardian.com/society/2023/apr/07/cancer-and-heart-disease-vaccines-ready-by-end-of-the-decade ) demonstrate this (ML contributed here as well).
Though we must manage our expectations, we should acknowledge the impact of emerging technologies. The influx of new information already complicates AI discourse, and this trend will likely persist.
While I have strongly argued my position, it should of course be noted that this is just opinion. Again, I find a lot of the discussion quite speculative, so I think we should all be open to changing course depending on what we continue to learn.
...which is why, again, I thank you for these videos. While your channel is not my only of information on the topic, it is always really nice to touch base and learn something new at the same time. I hope you continue to produce content on this channel at a rate that is comfortable for you. I wish you the best of luck and I appreciate your engagement with my goofy-ass "keep it up!" comment.
My goodness!! All these advancements are making my learning in Data Science and Machine learning a waste of time 😅. I'm curious to know what careers will thrive and what the future of prompt engineering would look like. Anyways this is a great video Philip. I always appreciate your insights on recent AI development.
Thanks Sola. If you stay at the forefront of all of this your knowledge will be super-useful to explain it all to others...
@@aiexplained-official Right, you are bringing us a lot of values.
I'm a butcher by trade (tech nerd by night). Would be funny if my lame job and others like it became one of the last bastions of the human workforce.
farming
@@ryzikx 😂😂
Fantastic video, the speed of change is so fast at the moment. Excited for the Open Assistant release on the 15th of April. Will you be covering this in one of your videos? The test version is live and I have found it to be very conversational.
If it is interesting enough
Thanks! Great content! 🙏🏼
Thanks again Steve
After every video you post i get progressively more scared and terrified by how the field is progressing. I was a massive supporter of ai and ai development, but now i fear for how fast and seemingly recklessly its evolving.
Am sorry Noon
Sasdam Hussein has weapons of mass destruction and ties to Alqaeda
It's unbelievable the degree to which things are speeding up, to the point where even summaries are becoming excessively hectic
You mean spend more time on each point?
@@aiexplained-official No, I mean that even the act of summarizing information is becoming more difficult as the pace of information increases. It's like trying to condense a whole book into a few sentences in real time.
For the algo
🙏Happy Holidays, have a good weekend, thanks for your work buddy.
Thanks Derty! Let me know what you think at the end
That claude bot app is similar to one I developed, named Role Call. Configure a personality, and setting, you can even configure a voice and have voice to voice conversations with that personality!
Is it bad that I’m excited about the games we will be able to develop with this technology? I know it’s not the universe shaking mega apps that people are thinking about but the possibilities are really fascinating. Games where you can interact and have full conversations with the characters would be really groundbreaking, this is the next step to personalized AI generated entertainment. First maybe we will have text based interactive fiction and from there the sky is the limit. It’s like the beginning of computer based gaming all over again, but with quite a twist. Eat your heart out Zork.
The 13B alpaca model and it's remixes is already good enough for dialogues with NPCs and they can run on standard GPUs.
I'm more interested in AI controlled players rather than AI generated dialog. Imagine enemy AI that always tries to play fair, goes along with spontaneous ideas, always adjusts itself to be challenging but not too much... That'd make for a way more entertaining game than a regular old game with advanced storytelling IMO
Another common sense reply with very little likes.
Putting aside the humor and potential outcomes, it's is scary to think we could all be annihilated because of some internal RP with itself. Not really much different then the old War games Cliche.. The memory aspect alone is in itself incredibly useful.. all quite thrilling.
It is thrilling and concerning in equal measure
This was an amazing video, thank you.
Thanks Sam
Oh hey, you got around to this. There are some great options yeah. Found another cool channel watching some of these
It's been said elsewhere but it could very well be that gpt5 writes gpt6.
I foresee OpenAI, Anthropic et al getting a massive influx of billions in funding, constructing city-sized server farms, hiring the best minds in the world to design and train up models that can probe the mysteries of the universe... and meanwhile their interpretability team down in the basement is just the stapler guy from Office Space.
Haha could be
Amazing as always!
Thanks G
Thank you so much for these information
I think we should never trust that we solved the alignment problem, we should instead keep solving the control problem. Together with AI, as AI develops.
Pandora's box never closes, it only opens further
In Max Tegmark's book "Life 3.0", on superintelligence alignment, he eventually concluded that to solve that problem, we essentially have to "figure out the meaning of life", without which a superintelligence may decide on a different objective and conclude that humans are redundant/obstacles to it.
Not a bright future
Thanks for great summaries.
Wow thanks Marek!
The rapid progress of AI is truly remarkable, yet I cannot help but express my fervent hope that our future will tread the solarpunk path instead of the dystopian cyberpunk trajectory. Alas, my apprehension grows as I suspect that the latter is an inevitable outcome.
I hope not
The way the population Collapse is going right now I don't think we will have much of a future at all.
9:22 I'm glad AI is struggling with the exact same problems as humanity.
i’m stressed out trying to keep up and get the various developments coded on my computer but it’s exciting!
This video was just awesome! Its amazing how much ai went far i will try to work with it myself the minute i will get to my computer
Thanks Nir