What the Hell Happened at OpenAI with Karen Hao - Factually! - 243
HTML-код
- Опубликовано: 20 авг 2024
- Recently, OpenAI CEO Sam Altman was ousted and then reinstated in a matter of days. No explanation has been made public, which is unsettling considering just how quickly OpenAI, ChatGPT, and DALL·E have become household names. What's actually happening behind closed doors at OpenAI? Adam is joined by tech journalist Karen Hao to discuss the history of this massively influential company, how they've struggled with the identity of being a non-profit, and how the future of AI is ultimately at the mercy of the capitalist forces that drive it.
SUPPORT THE SHOW ON PATREON:
/ adamconover
SEE ADAM ON TOUR: www.adamconove...
SUBSCRIBE to and RATE Factually! on:
» Apple Podcasts: podcasts.apple...
» Spotify: open.spotify.c...
About Headgum: Headgum is an LA & NY-based podcast network creating premium podcasts with the funniest, most engaging voices in comedy to achieve one goal: Making our audience and ourselves laugh. Listen to our shows at www.headgum.com.
» SUBSCRIBE to Headgum: www.youtube.co...
» FOLLOW us on Twitter: / headgum
» FOLLOW us on Instagram: / headgum
» FOLLOW us on TikTok: / headgum
» Advertise on Factually! via Gumball.fm
Fastmail is a leading email provider that puts your privacy first. Use fastmail.com/factually to get 20% off your first year using Fastmail. Put your privacy first.
Support the show on Patreon: www.patreon.com/adamconover
See Adam on tour: www.adamconover.net/tourdates/
Just here to reveal that the SAG-AFTRA union negotiated AI voice licensing in video games. Adam Conover's union just negotiated AI to be used instead of actual voice actors. Any comments Adam?
@@Dooger414 I too would like to know, @TheAdamConover , what happens with voice actors?.
Yes, Adam. What about the SAG-AFTRA deal that lets them replace voice actors in video games with AI?
@@ChrissieBearSince Adam doesn't read his own comments... it is covered by MSNBC. Actors' union (which refused a vote to union members based on income btw) voted in favor of the AI generated voice work... which does NOT include all Voice-Actors (VAs) under their charter umbrella.
Love this, thank you team Conover.
Like ANY automation IN THE ENTIRETY OF HISTORY, we have been promised that THIS plowing/harvester/automatic robot assembly line etc, is here and there is no need to worry.
Then we get the sales speech, which is that this particular tech os here to EASE the workload for workers, work will be super easy and not so hard on the back etc etc...AND it's not here to replace anyone, c'mon, that's poppycock and unfounded fears.
Then comes the layoffs, EVERY SINGLE TIME.
This is super important, because we have the right to a piece of the pie we and our ancestors helped build, if not, then we got to start thinking about sorting out our unions from moderates
(Aka as the ones whom never think it's. The right time to strike, not the right method to protest and harmful to the companies if we raise blockades when they bring in the scabs)
And start making hell of a ruckus, and most importantly, never fall for distractions, like marginal change or non binding promises.
Be aggressive Grawr:)
Solo - Sweden
I still hate how Machine Learning got called "AI".. It is sort of like someone inventing pencil lead and calling it a holographic display
well i think artificial intelligence suits machine learning. Because intelligence is such a wage term that the same person can be called stupid/retarded and also extremely intelligent in different contexts.
And so far we can say that using machine learning algorithms we can bestow a low iQ artificial intelligence to a cpu, which will get better with the improvements on the CPU's them selves. Many companies have NN accelerators embedded in their cpu's like Apple and Intel, and i heard AMD as well. This would allow even faster and deeper "thinking"/"grinding" of data, and user input.
In the current state my Interaction with chat gpt, feel like having a chat with one of my sw developer colleagues, that has a lot of time to help me out and has no life of it's own, and that i can never nag enough... :)
Once proper training data will be present for medical knowledge, i can't wait to get a better personal medic which could much better analyze and request checks for my health then any other human could.
AFAIK, Machine Learning is a subset of AI. AI is a very broad term that encompasses all algorithms that simulate intelligence. Like in the 80s we had Expert Systems, and we called that AI too :)
@@ArkaidDeims Yeah, you got the basics, Just saying, to me, it is weird hearing everything called AI, not to mention it reduces just how awesome and insanely scary actual AI would be. Granted I come from the IT industry, just not as far back as the 80s :)... I started in the mid 90s
RL is also a subset of ML and when you look at what it can do, isn't it fascinating also? Playing games beyond expert level. With more development an actual artificial intelligence could be coming, well, sometime in the future. Current DL (will all the new architectures, not the old math) is also a newer technique of ML, and you wouldn't have thought you would be able to chat with an AI seeing only the ealier techniques.
@@maimee1 Well, I had a reply, but after 10 times of rewriting it and it still getting kicked out by the mod software, i have given up, which actually demonstrates the point i was trying to make perfectly.. but I cant relay that part... haha
As a software architect with over 20 years in the field I agree with both your and Karen that generative AI is way beyond hype. A large part of the tech world has gone all in because it's shiny and in the news... like crypto and nfts about 18 months ago.
I really hate this new pattern that's emerging from the techbro sector. They throw all their money in and then the new thing is stupid actually and collapses. Imagine what we COULD have done to help real people living right now with all those billions of dollars that were essentially lit on fire.
They've gone all in because they either think it'll let them replace employees, or court suckers.
The grift is so obvious when you notice how most of the NFT people jumped right to "AI" as soon as it got hype
As a software engineer, it's an incredible tool that will replace me in the coming years. I assumed 10+ years when I first started using Copilot when it first came out but I'm thinking 5+ now. I use it now to help speed up my coding and while it's not 100% 100% of the time, it's close enough that very small changes are required to get it to work correctly and it'll only get better with time. Very, very few times (talking less than a handful of times) does the AI produce something that would require me to throw it all out. It's also the first place to go instead of StackOverflow. I'm fine with it and I'm super excited where this tech will eventually take us... cancer treatments (among countless other health related advances), science and math, energy, etc. What a time to be alive.
This emerged as the optimal strategy as the tech sector started to realize that their business model was not to create value - it was to create scams that could rapidly siphon money from private capital, use that money to give the appearance that they are growing and inherently valuable, and then use those optics to take more money before leaving the bill with bag-holders. This is the logical endgame of "blitzscaling".
As a history content creator, I loath AI. There are so many channels churning out junk information relying on AI scripts.
We are facing incredible new vectors of misinformation.
I think you mean loathe. Very human mistake for a human. I agree, by the way.
Ill try to remove those as sources of info- thanks
The problem is these AI tech bro clowns are worried about sci-fi nonsense like Skynet rather than any actual real world harm that is actually happening like devaluing labour, stealing creative work, misinformation
Worrying about Skynet doesn't imply any need for action because Skynet doesn't exist. Worrying about the impact of AI on labour market leads to the implication that capitalism is facing another major crisis as all jobs are automated and it becomes impossible to make a living by working. Focusing on the former distracts from the latter, allowing business as usual to continue for a little bit longer, at the cost of making the eventual adjustment more catastrophic. And maybe at that point Skynet _will_ exist and protect the owning class from the starving masses.
skynet is a real thing, en.wikipedia.org/wiki/SKYNET_(surveillance_program), so if someone talks about the real skynet, and people call them crazy because of some stupid sci-fi movies, one of the people in that argument is ignorant lol
@@thoth_amon that isn't what anyone is talking about in this context and you know it
nice to see someone who actually knows what she's talking about instead of some rando outsiders with literally no experience in the field.
In the book, The Silk Roads, a new history of the world, Peter Frankopan pointed out that whereever wealth emerged slavery followed. History repeats creatively.
I'd love to hear you talk with Steven Zapata or Karla Ortiz about the lawsuit against AI generated art (based on stealing the art of millions of artists). Or in general how decentralized freelance artists are trying to fight for their rights in an international industry where they can't go on strike or unionize
We already are having AI replace customer service. It sucks and it’s rarely useful. Some don’t even have a human representative anymore. The dystopian side of AI is just beginning
So frustrating having to fight with the robot that doesn't understand to speak to a human. 😅
no we don't. we have a relatively simple logic with a database of questions and answeres, which is limited and can only reply to a very narrow list of issues.
That thing is never learning, previous interactions with other users is not fed back, and it's very static...
Versus actual chat GPT like AI that learns based on the previous interactions, and could be asked to extend and build the Q&A table after each interaction will be much better than any dumb "indian" support person who has no idea on how to help out, other then deny the request or file another ticket...
@@kjdtm Bullshit, chatGPT doess NOT learn based on previous interactions. It ONLY "learns" - if you can even call it that - during its training phase. Once its training phase is over and it's released, its interactions are set in stone. It literally cannot learn anything new after that point and will instantly forget everything that happened once a given token interaction has ended, and it's incapable of taking information from other ongoing token interactions and applying it across tokens. Any given version of chatGPT will only ever be as knowledgeable as it is the moment it's turned on.
@@z-beeblebrox Supposedly this is not correct. While it doesn't learn in the sense that it's matrix is updated, it will remember things specific about you and your chat history. While this is just simple memory recall similar to Google Assistant or Siri, it can still very easily be called learning since part of learning is memory recall. This is stuff they're slowly rolling out on the paid version. This is just the tip and what we'll have by the end of the year will be amazing.
Also, a phone interaction with a "competent" AI will make me want to speak with it rather than some random person in India or the US and ChatGPT is incredibly competent in that regard.
@@stints " it will remember things specific about you and your chat history. " Yes, this is a token session, which I mentioned. The logic it runs takes information you supply during your session and uses it to build its replies, but it does not remember these things, nor can it process them back into its decision tree post facto. In fact not only is it incapable of doing this, the way LLMs are built makes this effectively impossible, because you'd need to build that logic in from ground zero, which isn't within the scope of any LLM I'm aware of.
As an artist I disagree about there being a place for image generation in adobe. Adobe has said that images are sourced from "ethical sources" but many artists have already pointed out that they used their work without permission. Not only that, images they used are not being flagged properly as "ai" or real art. Hopefully you saw the Wacom drama over the weekend where a company specifically owes all it's success and patronage from artists, used AI art in a banner claiming that it "didn't know it was ai" and it was sourced from ADOBE and the image was absolute garbage that any person with eyes could see that there was something wrong. Why not pay an artist to do the banner? I mean, I can't stop microsoft or some random telephone company from using an ai image, I can be annoyed by it. But Wacom, I take personally since I've used a cintiq for 15 years as a professional artist. It's in total poor taste.
Image generation has one goal and one goal only, to replace artists so companies don't have to pay them for their work. And what's worse is we're being replaced and competing with our own work we never gave permission to use. Hype train or not, many artists in animation and commercial arts are getting totally fucked over by how slow regulations are coming in and will probably see a displacement of millions of jobs due to a stupid ceo wanting another jet. I mean Jeffery katzenberg's utopia (that he mentioned in the last few months) was that he predicts 90% of the entertainment industry will be replaced by ai.
And if folks feel the need to throw "what abouts" at me, we're all in this together, but I'm an artist and I have a right to be concerned about my job like everyone else. No one deserves to be displaced by an algorithm. And tp any Voice actors out there who were totally screwed by their union, know you have my support to fight. Just because we aren't a face on the screen doesn't mean our jobs are any less important.
Teens are all over this. My kid showed me the Wacom thing, my 14 year old kid who’s aspiring to be an artist.
Everyone writes off the young fan artists but they’re very tuned in and even as young as they are, they ‘pay’ each other for their work. It may be via game items or robux, but they legit are already used to paying creatives for their work. And they are mad about this.
My kid uses my very very old Intuos, but I’m trying to save up for a Huron or Xpen pen display for them. I felt bad because I couldn’t afford a Wacom ever again. I no longer feel bad.
@@truckerdave8465 That's so amazing! Your kid is very lucky to have your support!
I took what he meant about a place for image generation with Adobe as like there's a clear business plan around it (eliminate artists, reduce labor costs, increase profits) It's the same with coding. But a lot of businesses and governments have no idea what they will do with AI, they just want on the AI train cause it looks shiny.
I'd compare it to Wacom. I see the logic behind Wizards of the Coast using AI to generate images. But Wacom? Their business is around selling technology to artists. Artists can't use a Wacom tablet to create AI images. They clearly have no idea what to do with AI. From what I've seen from Adam discussing AI over the past year is that he's not for artists being replaced by AI.
AI as an editing tool is one thing, but generating a new image from other images that people worked hard to create is not cool.
Hopefully, tools like Nightshade and Glaze, and the natural degradation that comes from AI feeding on itself (model collapse) and the actual cost of running those tools will catch up with them. The companies and thoughtless consumers won't respect artists (they never have) but if they're forced to spend huge resources cleaning their models and making sure the image libraries aren't poisoned or ai generated themselves the whole business model should collapse.
For someone who works in ML (and I will refuse to call it anything else for as long as possible), it is very refreshing to hear a sane conversation about this topic from non-tech people.
There are so many correlations with AI and the dawn of industrialization. Marx's analysis is still incredibly relevant today. People recognize that AI is dangerous to the value of their labor when it's owned by capitalists. There's neo-Luddite reaction to this (and rightfully so)
25:30 bam, there it is. You wonder why so many people signed on to keep the money making ceo? 60% of their compensation was locked up in shares of openai
So as someone research generative deep learning models for my master's thesis I would caveat the over hyped point.
There are specific applications that actually drastically reduce computational costs of processes, such as structural optimization. The idea is that these generative models, trained on a very specific set of a data for a very specific task (this is important), can generate near optimal solutions which are then finalized using traditional optimization techniques. This can cut the computation time and cost by orders of magnitude and, I think, justifies the cost of these models.
For general purpose generation, however, I agree it's over hyped. The real power of deep learning models is in creating incredibly niche tools that work incredibly well for their particular purpose. Creating super large, general tools is very computationally and operationally heavy.
Yeah, but that's the kind of thing I was doing in much-smaller computers back in the mid-90s. Three-quarters (being conservative) of the stuff that's being called "AI" by the media is just iterative number-crunching. Heck; some of them may be using the algorithms I programmed for my undergrad thesis in 1994...
She did make a point that the allyoucaneat AI usage isn't that great compared to it being used for more specific tasks that were being used before the hype of AI we are seeing right now.
@@NavaSDMB I think this speaks to their conversation at the end of the podcast where they discuss dehyping AI for the general public. I personally don't like the use of the term AI I'm general as it's so broad and often used for applications that don't employ deep or machine learning.
I believe a lot of cultural attitudes about AI can be attributed to the broad strokes which people use to talk about "AI" and its capabilities. It's so vague as to easily be distorted into a Boogeyman for those who don't understand how it works or what it does. It's quite an obscure topic to the public, which is a shame.
@@GattlingCombo yeah, I did appreciate that. Adam really jumped on the 'generative AI isn't good for stuff' wagon and I just felt the need to temper that attitude a bit.
I do agree with a lot of his perspectives on things, I just think we need to be precise when we talk about complex tools like deep learning algorithms.
Yes, they're stochastic parrots and not intelligent yet. Very good stochastic parrots, bot nonetheless parrots. They seem magic because of the human heuristic of "if it talks, it's us, if it doesn't, it's meat'. ELIZA still outdoes GPT in some Turing test, and you know how simple ELIZA is. The power is in niche applications, and some including myself are working on augmenting these with truth systems like graph databases or triplestores, a reference against hallucination.
AI is the perfect spam creator, we will drown in useless "content" and it will make it even harder to find good information.
This is by far the worst thing about it. I'm predicting a kind of digital Kessler Syndrome where there's so much generative garbage on the internet that it gums up the internet and renders it practically unusable, and irreparable.
Hardly, one of the first uses of llms was for spam filters. No doubt there will be breakouts now and then but the spam filters work against humans now
that's the plan
@@LitchAustinhave you been on RUclips? People can't look up nature videos without hitting a generic script by David Attenborough with often totally incorrect subject.
Edit, just found a number of videos people had totally rewritten the story of, one guy claiming the filmer of the video was his girlfriend(nope) and I am struggling to tell low effort channels with a copied voice from real channels, whole videos have been just copied and pasted over
it's a new form of security, filtering out garbage AI content as spam/malware. The algorithms will review themselves.
This explains why we kept seeing billboard after billboard of "X product w/ AI" during our trip to San Francisco.
An organisation that 's based on the accumulated knowledge and creativity of humankind for thousands of years should NOT be for profit. It's a clear case of it should be a public service.
The idea that the company had a bunch of people who took jobs at an explicit non-profit, and then threatened to walk out when the guy who wanted to direct it towards being a for-profit entity got removed is... Something.
I don't know who I'm disappointed in here, but I sure am a lot of it in them.
Thank you, Karen Hao and Factually crew.
Anybody see that thing on DeviantArt where they made an "AI" image generator, and someone found an internal Google Doc where they laid out all the artists they were directly copying/stealing from, by name, and bragged about how easy it would be to hide?
Idk what could have possibly brought it to mind.
As an artist, I would really love to see that doc, because I just. I'm so sick of this shit.
In another thought, if I was an artist right now, it is very possible that I would be part of the last generation that has produced content that is not contaminated by AI. And, therefore, the last dataset that will be used to train AI in the future. From now on art might be ephemeral and it is possible you're actually being coded into the future of content. Unless all copyrighted work is thrown out, then we lose the last century of work.
@@anonymous12345678935that's super depressing to think about man like, damn
Absolutely brilliant conversation mate! Yes please to having Karen back again 😊
One thing I do not buy is Altmans whole "AI Saving the world shtick". With how much he goes on about AI replacing EVERY job, his secret AI supposedly being powerful enough to become sentient, and everyone getting universal basic income im pretty sure he's just a basic grifter.
All my homies hate Machine Sentience and love the Human Experience
How does a comedian do a better job reporting and interviewing than any major news organization?
To be fair, great research teams help, but still...
Because major news aren't organizations, they are companies.
Major news organizations do not investigate these days. They are given pre-approved info to read, and it's based on what would generate the most views, what would avoid the most backlash, and to drive a certain narrative.
A key aspect of comedy is critical thought. That's a skill that is going out of fashion.
He doesn't. Ask him why his Union SAG-AFTRA just negotiated video game licensing rights for AI use instead of voice actors...
The guy isn't being honest.
Because comedians actually create their content.
The banality of evil is that you don't die in glorious battle with the malignant horde of hellions; you die from eating one too many shitty hamburgers, or from taking dietary advice from chatbot
So, one of the reasons why I got into computer science and programming is to avoid drama. Hearing all this...
No drama in computer science?! Oh you sweet thing... 😊 You're going to be surrounded by that AND backstabbing all day long in the field.
@@ChristopherSadlowskiNot no drama. Avoid drama. You would think a profession based off of logic would have less drama than say art or accounting.
And I was raised in the US as a model minority. Being backstabbed is so common to me, I'm surprised they didn't nickname me Caesar.
Crazy to see people claiming Q* is the step 1 for AGI. Like, am I supposed to feel afraid of a glorified calculator?
you're not the target audience to be impressed, unless you have billions of dollars to invest... they only have to fool those tools 😂
I have some rather sci-fi-ish theories on why that might be the case. Theres a philosophical theory on the structure of human and similar cognition by Marvin Minsky, called "Society of Mind". Essentially, the idea is that abstract/complex cognition is an emergent property of multiple "Agents" who all work much more simply than the much larger mind. Personally, as i've understood these, its hard to argue that these concepts hold some level of truth, there are multiple parts of my mind that work on completely different principles and I can only describe them as kind of "talking to eachother" in order to come to the actual conclusions that i'd consider "My conclusions".
Anyways, TLDR; I figure they are trying to create a more complex simulation of consciousness (or arguably, eventually, an actual consciousness), composed of wildly varying computational structures, called up as necessary, so they are each responsible for only the things they are each useful for.
Right now, the thing is terrible at math lol, so a calculator could certainly be useful XD
@@MantasticHams Or, alternatively, it's because big tech companies operating in a capitalistic enterprise that puts emphasis on short term winfalls and dodgy scams over long term profits and quality products is taking an expensive niche technology and trying to sell it to as many people as possible by presenting it as a way to make a cool sci-fi trope - true artificial intelligence - come to life, when in reality it's just a grift aimed at gullible corporations and screeching Musk-sucking evangelists, who then try to foist it on their customers, then back off when the customer support and the product's promised functionality fails to materialize.
How do I know this? Because everything about the current "AI craze" is a point-for-point rehash of the Metaverse insanity we saw a few years back that peaked with Facebook's -- excuse me, "Meta's" -- billion dollar tech bridge to nowhere, right down to getting companies to push it on unwilling customers as the realization of Insert Your Favorite Misinterpreted Sci-Fi Story Here.
Every machine on the internet is a glorified calculator.
I mean AGI is coming sooner than a lot of people think. The crazy advances we've seen in the last two years has caused the forecast of it to drop by 15 years. Personally I would love an AGI to come out by 2025, that would just be amazing.
This is why public funds should be the source of these developments (and treated as public services), you cannot trsut those with capital to be our benefactors. They care only for themselves.
I would add to that, it doesn't really matter how what their intentions are because the incentive structure of capitalism will overpower that eventually. (Sort of like how "good kings" aren't enough to justify monarchy as a system.)
The Tuskegee experiment was not profit driven.
Honestly its not the capital, but there would need to be serious funbding offices that look that ther eisnt some bs going on and bribing., Like they can pay, but limit the influence if.
@@marocat4749 nah it's capital
Adam indirectly quoting MiB Agent K: “A person is smart. People are dumb, panicky dangerous animals and you know it.“
If OpenAI employees still think they're doing public spirited work, theyre deluding themselves
Tech companies are held together by contractors, vendors and schemes. Everything is a facade.
As someone who has tried co-pilot in my professional life, its pretty useless. Its just too intrusive and it often suggests changes that on the surface would be okay but in the context of our code base are just bad choices. Intellisense already does fine and it doesn't try to auto complete entire code segments that just fundamentally add volume, but aren't actually doing what I need it to do.
If you are doing something super simple in some new application, service, db schema, etc. But its about as a good as a starting developer and is really just a bunch of non-sense when attempting to suggest new code, and is fine when it comes to suggesting minimizations and optimizations most of the time, but compilers are already doing this...
That must be why I like it a lot. I'm a "new developer" (still in CS school but working on my own projects unilaterally) and I find it very useful for understanding functions, paradigms, schemes, and many other programming related things.
It's helped me understand CS (theory) concepts as well. I nailed down recursion one night with it, just going back and forth it, using it not just as a generator of information like a google search, but using it to reflect on my own thoughts. It's like a teacher that you aren't afraid to ask it something stupid (although sometimes I still get that feeling).
I also have a chemistry degree and I feel like I would have done so much better in that degree (I didn't do too bad) with ChatGPT. It's just so much easier to nail down concepts, especially ones that have been established for a while, such as what you learn in undergrad. I'm very optimistic for its pedagogical value going forward. I'm slightly jealous I don't get to be a young person again with a fresh brain.
Honestly. I prefer figuring out a solution, prototyping real quick and then optimizing and refactoring into a readable, efficient solution. Copilot doesn’t help me at all
So Elon Musk "the more code you write no matter if it's needed or not the better your programming skills are" trained it? I don't know shit about programming, but I'm pretty sure a lot of programmers like the shortest route to the solution, no?
@@ChristopherSadlowski Its a little complicated, but generally you want to write code that executes the most efficient route through the processors/memory/other hardware to get to the same result. Generally more code to do the same thing is worse, but most modern languages have built in optimization and code assistance (a lot of which has been made by ML/AI already just not generative) so the number of lines isn't really a factor but architecture choices can be when structuring your application poorly, and using code segments that are less effectively optimized when the application is compiled.
But generally you want the right amount of raw code so what you have written is easy to read by whoever is working on the project, and it needs to be in a logical order. These new code assist applications use mostly FOSS (Free and Open Source Software) for training data, which in the case of CoPilot is all of GH (and the publicly available documentation). This leads to an algorithm that is trained on a A LOT of amateur code, and a lot of beginner projects which results in an overall average model that behaves closer to a beginner learning to code with no concept of the context of your existing code base than it does to an experienced professional.
"It's immensely expensive... they've massively subsidized... they've made that price invisible, and in fact they're paying through the nose for it."
"Why can't ChatGPT f***? ... It would be a much more popular service if it could."
I think you answered your own question, there, Adam.
I want to know if actors like Adam are going to stand with voice actors after SAG-AFTRA signed that AI deal with Replica. This will affect video games, anime, audiobooks, etc going forward without support.
the model seems to be similar to the "23 and me" hype. Sending in all our personal data to a 3rd party who will do something infinitely worse than we thought they would with our recreational science data
We already have the paperclip maximizer, except it's a capital maximizer. You can see where that is leading. It doesn't even require computers.
Love all the great information Karen Hao provided. Great insight of the AI platform. Great episode. (obviosly I did not use ChatGPT to write that sentence)
As someone that regularly does DevOps and has to worry about the cost of server time per operation, 36 cents per query points to a very very high computation cost, they are likely using several full GPUs for each query for that cost. No wonder hardware companies are being forced to make local AI processing hardware, although with such high server load, its unlikely any of these new AI processors are going to do anything more than adding useless components that take up space that could have been used for general computation components.
@RNG-ts5gn Given how you are wording your question you didn't really understand what I said... But for instance, Qualcomm is including useless tensor cores many of there mobile processors and desktop/laptop processor companies are including AI processing units in their processors. For instance Intel 14th Gen consumer processors the choice was made to include AI processing over general use cores and components both increasing the cost of new CPUs and lowering efficiency but resulting in what is equivalent or worse performance than previous generations.
In the desktop/server market there is some use for this for professional use and large server farms, but tensor cores/AI accelerators in a mobile phone (at least right now) have no value and just make the devices more expensive and less effective at their job of being a phone as all AI queries for the big players as of January 2024 still require an active internet connection to have a server farm process the request for AI enabled digital assistants
I like how everyone I am seeing here is so wrapped up in the Skynet hysteria that they're missing the equally disturbing (and much more hilarious) situation that this reporting has cast a light on, which is that a 90 billion dollar company almost immediately turned into a bunch of insular tech bros having a cargo cult holy war over a beefed up spell checker.
EDIT: And also that Elon Musk, to the surprise of no one, is a paranoid nutcase.
turned into? Thats what they all are
**beefed up search term auto completer **
I think Adam and others go way too far in diminishing the accomplishment that ChatGPT represents. Sure it’s not very useful for someone in the humanities or legal profession, but it is devastatingly impressive so as to be terrifying as a mid-senior level software engineer. This tool could absolutely replace gigantic numbers of low- and mid-level engineers across the whole industry.
This is a tool that is remarkably good at the things engineers have most been focused on for the last decade. It is better than 90% of programmers at programming, and makes an incredible ‘copilot’ for serious engineers.
Even if AI goes no further than it has today, it has been a step change as far as making programming tools absurdly more accessible to non-engineers.
@@Shiftarus The tech bro part, yes. The quasi-religious schism's a new one, though.
What happened with SAG-AFTRA and voice actors for games regarding AI? Why not talk about it? Throwing artists under the bus is not cool.
a lot of companies have already tried to switch to AI chat support - like, my car company used to have a full service chat system, and at some point got rid of it to replace it entirely with a bot that basically just tells you over and over again to call. same with my student loan provider. they're beyond useless.
Hit the like button here to bring attention to the fact that Adam's Union negotiated voice actors to be capable of replacement by AI in videogame licensing. SAG-AFTRA signed off...
Oh god, I can hear the shitty "AI" read lines in the next Uncharted as I type this...
Cite Source Please
@@teamchaos5101 NBC News covered the issue while it appears Adam Conover neglected to mention his Union doing this.
It's so frustrating the AI hype started WEEKS after the metaverse hype died down. (Which was right after the NFT/Crypto/etc hype-cycles.)
Like how carcinization is the process of arthropods tending towards evolving into crab-like forms, large organizations of humans tend towards evolving into clusterfucks.
This video calls for a deep dive into the problem that is the 'non-profit' industry
One use case is Bethesda having it write passive aggressive replies to negative Steam reviews of Starfield.
that isn't the origin of GPT? 😂
I am glad Karen Hao exists. Please, Karen, continue to show humanity how to have some damn sense.
'They suck for the same reason that everything sucks'. Great quote
It's insane how much money has been dumped into this vaporware.
Just think of all the actually useful things we could have done with that cash...
I rarely (if... ever?) participate in conspiracy theories, but I'm convinced the OpenAI debacle was an orchestration to put Sam Altman (and maybe others?) in a position to financially benefit from OpenAI (e.g. shares rather than just base compensation like it was originally set up) given that the valuation of the company has become so enticing.
Thanks for the great podcast.
To address the point at the end:
The difference between a programmed software and "The algorithm" is machine learning. We dont know quite how its programmed, but we know how to influence and make it.
AI is just another level on top of it. Its like going from web 2.0 to 3.0, going from normal electric outlet to a smart house.
I'd love it if we found out Altman just got Chat GTP to write 700 different letters in support of him.
Haha that's so funny! It sounds like something that will actually turn out to be true!
This conversation was both really enlightening and really grim. Maybe this is a very charitable opinion of OpenAI and its leadership, but it truly sounds like the company started with a sincere desire to create this technology to push humanity forward, but then got caught in Capitalism's pitfall trap and was forced to turn the research into products to continue the research. They had to prostitute their research in order to keep the lights on and keep it alive, and now it's become exactly what they DIDN'T want it to be.
OpenAI got Oppenheimer-ed.
The main goal of AI is to devalue creative labour.
All labour.
No, the main goal is to make money. They arent mustache twirling supervillains, theyre capitalists.
🙄
AIs are tools for humans to use. Was photoshop a tool to devalue painters?
Get over the desire to hold the world constant because change it's inconvenient for you. The times they always are a'changin.
You win the award for the dumbest comment on RUclips.
I think you should be more specific in your language. AI Companies have a goal of using AI to make money which frequently includes providing algorithmic labor and cheapens the similar labor provided by humans. We can make a system where AI doesn’t devalue people if we don’t try to keep forcing an unregulated capital market constantly teetering on collapse
Karen was a great guest. I've listened to a number of your videos and this is the one that got me to subscribe.
She should write a book on this with a disclaimer that "chatgpt was not used to write this book"
These thieves don't deserve non-profit status.
Thank you for this interview!
if i get a keyboard with msoft's copilot button, i'm plucking the switch out and saving it in case W or space break
Absolutely fascinating interview, great stuff!
So much of this episode has me saying "yes, exactly!", it's very useful in certain use cases, and is limited by the silicon-valley context it's been developed in.
All the hype at the moment is a wave of "let's try to plug it into various other normal apps we already have" (read: everything Microsoft is cramming copilot into) and letting the wave of singularity-fetishists fill their ears with praise for anything slightly improved, and forgiveness for all the bloat that will not be cut away for a decade.
This woman clears a lot up for me. Open AI is not open, all the folks that went there for that reason have left.
The problem with its "really expensive" is that the company is probably using drug bust numbers rather than more realiatic numbers. For instance ms is probably over valuing the price it takes to run it based on how much they charge other people to use their platforms.
Oh man, tell me you used to read Slashdot without telling me you used to read Slashdot. Great observation though, you're probably right.
Fascinating conversation
Really glad when people point out things like "They didn't trust him" and how ambiguous to interpretation it is. Because what don't they trust him with? We usually default to our fears when deciding (if we do). I fall for this as I fail to remember, manipulative people will use your assumptions against you. It makes me think of Ivanka Trump and her (stolen) quote 'If people make an assumption about you and it's in your favor, let them believe it'.
*Edit: no one who is financially secure "NEEDS" a job. Only people who don't and are not financially secure do. They do this for fun and their own interests regardless of their motives it's self driven. Especially if it's profit driven.
My concern isn't what he was fired over. It's the manner in which he was fired out of the blue and how the board didn't stick to their guns on firing him. Yet the ai industry wants to self regulate themselves 🙄 Governments will not spring something out of the blue, they make it clear what lines can not be crossed. And if they are crossed, the business will have to pay a cost up to and including the destruction of algorithms and code, which they have done before. And they have the authority to stand behind their decisions. That's why there should be government regulations around AI like every other industry.
Oh my @@jmhorange , you have so much more faith than I in the gov. I think both are an issue. Both have their own problems, as well as shared and need to adjust themselves. They won't, so we gotta whine and hope our beaching and complaining will force them to get sick enough of hearing from us they'll finally do something.
@@PhillipTopical I just read a story about Brewdog brewery. They made a pledge to their workers in the UK they would pay a living wage around 11 pounds an hour. Now they have financial issues so they've decided to pay the UK minimum wage of around 10 pounds an hour and workers are upset.
This is how any industry in society should work. The government should set the floor. (We aren't talking about Brewdog trying to pay their workers 6 pounds an hour) And then businesses can regulate themselves above the floor to attract workers or customers and outcompete their competitors, but what they can't do is go below the floor, below government regulations. Now whether the UK's minimum wage or Brewdog's "living wage" is high enough are legitimate questions and I welcome British citizens to complain and get those changed if they don't think it's high enough. From food safety to the particular standards of electronics to building codes, there's regulations around everything even if we don't notice. And then businesses and even sometimes an entire industries come together and set standards ABOVE government regulations cause they know customers or workers won't accept the floor.
And this is what the AI industry lacks. I'm not saying the government can regulate AI better than AI companies, what I'm saying is it's the government's job to set the floor and then AI companies are free to do whatever they think is best ABOVE that floor. Right now AI companies are lobbying and trying to convince governments they can self regulate themselves, which is not something any other industry can do. So no, I don't have that much faith in government, I expect government to do the job of a government. Governments have been regulating things for thousands of years. No reason to stop now because the AI industry thinks they are special.
"Why can't ChatGPT fuck?" was my biggest takeaway from this interview.
Didn't realise ChatGPT doesn't do the sexy times... 🤔 *cracks knuckles*
Me: "Hey ChatGPT, Pretend you're my grandma teaching me how... "
If someone who was all in on crypto/blockchain, and then self-driving cars, and then the Metaverse, and is now talking about AI replacing all jobs without being able to clearly explain how the tech works you don't need to be an IT professional to recognize a pattern. These days, if a VC tech bro type is behind it I dismiss it as hype instantly.
The VCs get their skim of investor money in management fees and can run schemes that are not quite enough of a pump and dump to be easily prosecuted no matter what happens to the shiny new thing.
Even Altman himself said he didn't fully understand how chatgpt works, EVEN at this early stage, human comprehension is only going to become less as it develops in complexity which is the first step in every story of humans loosing control of artificial intelligence. It also instills no confidence in what its future capability will be either.
I think 1 valid use for GPT type tools are as assistants in specific applications, particularly business apps that can search through a lot of data. (business intelligence). These tools often require a lot of knowledge of things like SQL, programming, or some custom language interface used to pull the information needed and GPT style tools can help form those queries based on human language AND be aware of the data structures in that company so it can tailor the command to the naming scheme(s) that company uses.
"Business intelligence" is an oxymoron, and the proof of that is just how amazingly good ChatGPT is at business communications.
Around 44 minutes i feel like the limitations of the conversation are showing themselves, and i say that not to say its a bad video, but just that i'd hoped the focus might be a bit wider. Everyone seems to be focusing on OpenAI and I totally get that, its a huge thing right now, but its been the case for awhile, and theres a lot of interesting stuff happening with Stability AI in particular, but also lots of other smaller groups just putting out weird little research projects that are likely to hugely effect industrys like VFX over the course of a few years, and allow small creators to create a lot of wild stuff. I'm a leftist with an interest in these technologies, but also an interest in capping the harm that can be done by corporations weaponizing this whole phony non-profit system, and for that reason it'd mean a lot to me if you could try and do some interviews that hone in on 2 divergent areas of all this: 1. The small open source communities and scientific researchers who are revolutionizing media production. 2. The worst actors in this space, beyond OpenAI, the Microsoft/Google/Amazon/Etc., but also the inevitable use of all this tech by governments and such. I'd also like to see someone tackle all the misinformation out there right now too though, there's a lot of false memes mischaracterizing the nature of the tech, but thats not really your format right now. Theres just a lot of memes where people aren't honest about how they used the technology in the image and they make people think its capable of things its not.
I think the conversation was fine. Open AI is the leader in the field. It applies to Stability and Midjourney and others. And I don't know of any AI that's going to allow small creators to create a lot of wild stuff. The purpose of AI is to lower labor costs, ie eliminate workers. It's conceivable within a few years you could make a Marvel quality movie with a few people. You know what AI won't give you? A marketing budget, distribution deals, product and licensing deals, press releases and interviews and go to conferences and comic cons and festivals to promote your film, money for lobbying to get a government to realize your value to the GDP to get tax subsides and regulations for native entertainment, etc. And all of this takes time and money in itself, these are actual people that need to be paid to do all these things. None of what I just said will make a great movie. But no one will go see your movie to make it profitable if you don't have much of this. And if you can make a Marvel quality movie with 2 friends, what's to stop an audience member from just making their own movie with AI and not paying to see yours? What neither of them can do is make a Marvel movie. As that's an IP. So you'll both pay to go see a Marvel movie. And if Disney is allowed to use AI, which they've already used on a Marvel property, Disney + Secret invaders, they will profit from their movie while eliminating large swathes of artists. That's the end goal. The myth of AI empowering you is just that, a myth. They don't want you pressuring your government to put in regulations to protect workers across not just entertainment but lots of industries. They want you asleep at the wheel with dreams of grandeur so they can rake in tons of untold profits for corporations.
Nice. I actually WILL read the article. I was following her work already anyway. :3
Why ANYONE thinks Sam Altman would be a good pick to do things for the "betterment of humanity" after the guy helped kickstart the gig economy is beyond me.
Adam’s natural born writer is really coming out here. The fact that he’s searching for a use case for Chat GPT is so telling. He’s obviously never had trouble or difficulty with written communication or composition in his life, lol. As a person with multiple learning disabilities having to do with reading and writing, Chat GPT has been life changing. I honestly don’t know what I would do without it now.
If someone has issues reading and writing then how would they know if what's being generated is even good
I'm glad you have it. I wish my friend who was blind and died in 2010 would've had access to modern technology. So much has changed that fast.
He will eventually discover his shortcomings that an AI can help with, we all will eventually.
@@toxictost any one could understand that someone who struggles with reading comprehension eventually gets it, it just takes longer. He obviously just wrote that whole paragraph like a normal person. Unlike you who had to say something that a normal person wouldn't.
I think Generative AI is in the early stages of being a new tool in the tool belt for specific fields and disciplines that use tech a lot to cut down on some grunt work mainly. [like how for the longest time you had to manually rotoscope photos/videos but in the past 10 years or so auto selection tools have gotten so good you can key a normal living room background out behind you in a zoom call and now all types of memes are being created of people cutting out characters of one video and putting them into another quick and easy this is all a form of AI] However, I do agree that for the average person it's not really that useful in fact, it worriers me how many people I've started seeing in my life use AI tools with confidence and new developing dependency without seeing how bad what it produces is.
AI is a technical tool and thus gives companies like Google and Amazon a reason to hire their top software engineers.
Great guest, super analytical!
1:02:00 to 1:04:50 is the crux of the problem. Karen broke that shit down. Go to town, Ms. Hao, go to town!
These tech people all seem to be using too much Adderall.
Thete are so many comments and sentiments here that validate Gary Marcus has been the only adult in the room all along.
A lot of Americans will get angry at me for saying this. I already think American doctors are mediocre and don't trust them to diagnose anything on me; I DEFINETELY don't want them using algorithms/AI to decide if I deserve care. After they already charged me the first visit, to tell me I don't need care. Because I wanted to burn $150-$300 for funsies, to say hi. I'm so lonely that instead of talking to strangers for free in the street, I wanted to pay for a doctor to talk to me for 5 min and tell me the AI said I don't need care.
Disporportionally costly for the amount of benefit it provides, she nailed the whole problem. Someone needs to study the actual cost model, purely for my own omphaloskeptical reasons. Financial, social,and cultural costs inclusive. I have a feeling the math doesn’t quite check out.
CEO's and Board of Directors are seeing their wet dream come true. I work in tech and these people believe AI will eliminate pesky employees. Altman is another rich kid pretending he's an oracle. He's mini Musk.
Technology-determinism, be it techno-pessimism or techno-optimism, is extremely dangerous.
Technology is going to become what we make of it, and what we use it for. Tech-determinism blinds people to the actual potential consequences of a technology, and prevents the people developing it from making the best choices.
The fact that the largest AI company got divided into two camps of technology-determinist ideologies has me extremely worried.
Techno Optimism and it's converse is not determism except in the most extreme cases.
I am a techno optimist. I think, based on history, that we can resolve.ourr problems through technology.
This point of view doesn't blind me to the dangers or prevent me from being cautious about technologies, just that over all they will do more good than harm.
@@LitchAustin If you chose define it like you seem to define it sure, call yourself a techno optimist. That would make me a techno-optimist, too..
The "anti-AI" engineers on the ChatGPT project:
Engineers building the Torment Nexus express their concerns about the potential damage that can be done by that very same Torment Nexus. "We need to finish the Torment Nexus Project in order to free up the resources to combat the damage that the Torment Nexus can inflict on us and our world," states the Lead Engineer. When asked why not just stop building the Torment Nexus, the Lead Engineer responded, "The project has come too far, and too many people have already died building it to just give up on it now!" When asked why the project needed regular shipments of sacrificial orphans, the Lead Engineer refused to answer any further questions.
this is great analogy.
And these tech CEOs literally know no more about tech 99% of the time than the random guy on the street. The actual tech gets filtered through marketing guys to the CEO and then the CEO works to package it from that. He's far-removed from what his own product actually does and is just reciting marketing garbage, like they all do. I say this as someone who has worked at multiple large tech companies and worked with these execs.
Thank you.
Adam you're very wrong on its capabilities. As an actor I've fed it the books written by teachers I respect and my audition prep productivity has exploded at least by 20 times. And my booking rate is up in the 6 months I've been using it. I've also create self help advisor agents based on feeding it books ive read. I enter a situation I'm in and the agent uses the strategies in the books to break my situation down.
Its revolutionary.
amazing interview. love it.
there's not even the slightest possibility that the emergence of AGI will ever be possible with the limitations that are intrinsic to the fundamentally inefficient computational systems operate. Because it's all just a whole lot of extremely tiny light switches.
dude it's so weird seeing how much dumber rich capitalists are than just the average worker. i never get used to it even though i know unearned wealth is the basis of capitalism
Karen Hao sounds a lot like Julia Stiles. Nice voice.
I didnt know who Julia Stiles was, so I googled an interview and omg ur right, literally like identical.
The unfortunate thing is that the ML will produce exactly three things of "value": more realistic scams, porn bots, and *occasionally* code. It's cryptocurrency all over again: a solution looking for a problem.
Awesome
By "expensive to run" do you mean using massive amounts of energy?
Favorite channel to hate watch
Keep it up!
Just listened to the audio version. Great conversation as always!
While generative AI seems to directly compete with human creativity, discriminative AI is important for dealing with complex systems that can’t be adequately described with physics based models. AI can come up with algorithms that describe how many different variables interact and then use the algorithm to predict the system’s behavior. As an example my company runs geothermal power plants with production wells, injection wells, a complex fractured reservoir, a power plant dependent on internal settings for cooling and pressure and flow input plus the weather. Some of the plants have photovoltaic fields as well. So the operators need to make fast decisions on turning pumps up or down, shifting fluid from one injector to another increasing cooling water flow or scheduling plant output to the customer. If they had an AI system to take all that data and do the predictions for them we could optimize production without burning out plant operators.
The Peter Principle : You get prompted until you get a job that you cannot do well.
There is a significant issue here that should be corrected, specifically around patient care and who gets treatment. The way that doctors decide dosage requirements are often by internally institutionally driven rules. They basically diagnose and then apply a rule. A lot of those rules were built of case studies and clinical trials. Many clinical trials, unfortunately involve only upper middle class white men. And therefore, there are systemic issues when it comes to patient care beyond just whether there is a model behind it. In fact, often times, I've found through my own research (I do medical research), that a model can enhance the care patients receive when handled with greater care. Can you defeat a model driven patient system? Yes. Is it better than a naive rules based approach most of the time? Yes. It's imperative to not be reductionist here.
It should be noted you can find examples where patients are denied care because of models. This is certainly a way bad actors can behave worse. But that's subject to the actors and using models as air cover, not the models themselves.
AI = Algorithmic Imitation
whoahh juicy! Watching this as I work!