Artificial Intelligence: Last Week Tonight with John Oliver (HBO)
Вставка
- Опубліковано 25 лют 2023
- Artificial intelligence is increasingly becoming part of our lives, from self-driving cars to ChatGPT. John Oliver discusses how AI works, where it might be heading next, and, of course, why it hates the bus.
Connect with Last Week Tonight online...
Subscribe to the Last Week Tonight UA-cam channel for more almost news as it almost happens: / lastweektonight
Find Last Week Tonight on Facebook like your mom would: lastweektonight
Follow us on Twitter for news about jokes and jokes about news: lastweektonight
Visit our official site for all that other stuff at once: www.hbo.com/lastweektonight - Розваги
Almost a decade ago, I attended a job fair which had a resume specialist. The subject of the seminar was improving chances of your resume being noticed by employers. The specialist's only real advice was to cut and paste the entire job listing that you were applying for into your resume in either a header or footer, change the text to white, and reduce it to one point font size. This way, the algorithms that scan each resume would put yours at the top of the list because it had all the keywords it was programmed to find.
Wow, that's crafty, but dark also.
That's basically "keyword stuffing". It's an old trick. It might have worked 10 years ago, but the algorithms learned how to detect it long ago. They have got surprisingly good at understanding the context of content and no longer reward this practice.
@@PtiteLau21 The same thing that happened with UA-cam algorithm. It used to be that the algorithm will only use the video title as keywords, but people then game the system by including popular keywords in the title that doesn't describe what's in the video.
Yeah but people still read the resume. Smh
@@miguelangelsucrelares5009 But are they punishing the practice? Because if not, it's still "why not" just in case.
Personally I consider it a somewhat dishonest practice that deserves a moderate mark down. It is also an indicator that the resume might be otherwise inflated as well.
"A.I. is stupid in ways we can't understand" as a software engineer I find this line surprising accurate
So are humans in that matter
Yeah, the same here. But the confidence it has with the bullsh*t it produces is so scary.
Engineer Makes Something That Works: Excellent! Now let's take it apart, verify everything is still functional, then maybe add more features.
Scientist Makes Something That Works: As predicted, but excellent! Now let's try to prove it in even more elaborate experiments.
Programmer Makes Something That Works: ...**Spittake** That worked!? _We must never touch it or look at it again in case it breaks_
But for how long? Ai will probably figure out how stupid it is and how to fix it before we even realize that it did.
@@attemptedunkindness3632 You're a programmer aren't you? You forgot the happy dance part. There is always a happy dance after it works.
absolute best take I heard on this: "we successfully taught AI to talk like corporate middle managers and took that as a sign that AI was human and not that corporate middle managers aren't"
“Final boss of gentrification” is a wonderful line
The Tay AI also made the funniest tweet ever. She said that Ted Cruz wasn’t the zodiac killer cause Ted Cruz would never be satisfied with the deaths of only 5 innocent people
Wow! What a shot to fire
It's obvious that she does not like that man Ted Cruz
Her point is correct, but doesn't support the conclusion -- obviously, his lack of satisfaction led him to pursue politics.
@@grigoribelov391 She does not like his far-right views
@@arutka2000 Naaah it's definitely his most punchable face.
IBM's insight from 1979 is still valid today,:
"A computer can never be held accountable
therefore a computer must never make a management decision"
Do you think human managers are held accountable?
Yeah except when the people at the top *want* that unaccountability. "Oh, it isn't our fault, don't punish us. It was the computer's fault, that dang ephemeral algorithm."
@@TheVerendus that's on point, responsibility diffusion is the fuel for cruel decisions
I'm not sure that IBM is the best authority on holding management accountable fam
The same IBM that helped the nazis create a more efficient holocaust.
What shocks me most about AI is how rapidly many people are eager to trust it with important tasks despite not understanding what the product fundamentally is. It's very good at predicting the next word in a sentence-a hyper-advanced autocomplete. It doesn't *think creatively.*
it's a brilliant tool when used properly, but people hear "intelligence" and assume it can actually think. great for mid-level filler, common-formatting, prompted random jumping-off points -- bad for research/fact-checking, unbiased or critical perspective, and responses requiring unique, uncommon or specific understanding
As an example:
"Write me a marvel movie script" will probably turn up a marvel script that cuts together scenes from previous marvel works or fan fictions it found on the internet
@@devinablow While that's an okay description of the current state, the elephant in the room is that it evolves at least exponentially.
@@devinablowI actually think AIs will be much, much smarter in the future than they are now. But people wanting to be comfortable around AI and minimising the risks of it will be our downfall.
Thing is, people don't understand the difference between classical, or 'stupid' software, smart software and AI and ESPECIALLY the people in charge of states and companies today, are people that have encountered stupid software AFTER their education, not during their education and they've to this day learned "Stupid Software can only do what I tell them to do and this they do perfectly unless I make a mistake", they're then barely aware of being in contact with smart software like the usual non self learning algorithm and completely ignore their existence and then they encounter AI and still keep the mindset of "It does everything I tell it to do perfectly" without recognizing, that AI is actually doing jack shit and in the best case just is calculating statistical probablities and if it cant ... it just makes fucking stuff up because that's what it's programmed to do *g*
Therefor "As long as I as a user don't make a mistake, the software won't make a mistake" while it actually should be "If I want it to do something, I'll better make ABSOLUTELY sure that I myself am the highest authority on what they're supposed to do just to make sure it'S not fucking me over with some random, made up bullshit"
It's kind of secretly changing the formulas in your next door neighboors favorit boomer excel spreadsheet around randomly which they're using for 20 years now to an extend that they don't even look at the numbers anymore because "It was always correct"
"The final boss of gentrification" is one of the most brutal roasts I've heard on this show
yes!!! absolutely top ten funniest shit i've ever heard. Cause it's like you're sitting there thinking "what is that outfit?" and immediately he hits you with it. This writing team is bar none i swear. They don't leave jokes on the table at all. Everything is accounted for. Love it.
I guess you like making fun of yourself....yt'y.
I laughed so hard
the funny thing about the "i want to be alive" is that, since AI just reads info off the internet, the more we talk about self aware AI, the more it will act like it is self aware.
and perhaps, the more we will ask ourselves, what does it mean to be self aware? what does it mean to be conscious?...
There's a thought experiment called the Chinese Room, and its... pretty disturbing. Essentially it's a theoretical proof that we can never know for sure if computers are self-aware, and they could suffer in silence for millennia without us knowing
@Jacob Singletary: That thought is terrifying, and here's why: one of the key hallmarks of a psychopath is complete lack of empathy. Because they are lacking in empathy, they must compensate by becoming good at reading people, manipulation, and mimicry; they match their reactions to whomever they're with, pretending to feel what they are psychophysically incapable of feeling, and tailor that façade specifically towards their present company.
Put a psychopath in a room with a psychiatrist, the psychopath will be forced to adapt all the harder, so as not to get caught. If they're succeful in this new hostile environment, the psychopath becomes all the better at faking genuine human emotion, but make no mistake, they're still a psychopath, still highly manipulative, and still potentially dangerous.
Now, here's why the original premise is so scary: the situation is the same for so-called AI, just replace empathy and emotion with actual intelligence. We could end up with an AI so skilled at faking that it's self-aware, and nobody would be able to tell the difference. Now, if Alan Turing were alive today, first, he'd prolly wonder why he always felt so overheated (cremation joke ftw), but second, he'd say that at that point, there is no difference between faking it so good that everyone is fooled and actually being self-aware.
Frankly, self-awareness is just a baseline problem, it's what an AI _does_ with that self-awareness that's got me and several much smarter people losing sleep at night.
@@sdfkjgh it makes me wonder if an AI could actually fool itself into thinking it is truly conscious and self aware
@@jacobsingletary8857 Fool 'itself'? No. Not the current iterations anyway since it has no thoughts to speak of. It is just regurgitating information. It doesn't actually know or understand anything; it's google search results, but with phrasing capabilities. It's basically a more advanced version of word predict features on your phone. Now can we get an AI to speak to you as if it believes it's self-aware? Yes. You could probably even go ask GPT to pretend it's self-aware while answering questions and it would do so. But it doesn't mean it really believes that or has any thoughts about anything it's saying.
the problem with "opening the black box" is that not even the developers know how it works. It's not the same as source code.
it's like a box of sand, and you pour water through it. You see it trickle through, but you have no idea how or why exactly it's choosing the paths that it does.
They're thinking about the problem in the wrong way, it's like asking what sequence of neurons fired for a human to behave in a certain way? Even if you knew the answer it wouldn't be meaningful
@@chrismcknight7164 exactly!! It's the same as seeing animal shapes in clouds! We can see it but even we don't know exactly how we see it!
Actually, developers do know how it works. This is a bit of an oversimplification.
@videocritic2087 , they don't, they know how the individual matrix multiplications work, but that is like being able to calculate the interaction at each grain of the sand. So yes, we know what happens to the data, like this number gets multiplied by five and added to this number, but we have no idea how billions of these simple operations make it to recognize a number. We can track each of the input values, how they are converted to the output value, but it is just a meaningless pile of simple mathematical operations, that for some reason magically works. You can maybe analyze simple ai with few neurons, but this isn't the kind of ai we are talking about here.
@@videocritic2087 it's like asking you how you decided that cloud in the sky looks like a bear. We know neurons fired in your brain but we don't know how or why you came to the conclusion that it looks like a bear.
How he can do this for 30 minutes straight is always incredible.
It's comedy cancer
@@SaydeeEnward4500 You're insane.
@@SaydeeEnward4500this is either a right winger/MAGA, or an edgy 25yr old troll who listens to joe rogan and Andrew Tate.
He has a team of writers, and they do it only once per week; but they are working on other stories all the time they are producing the stories that make it to show.
@@cjstone8876 he means the delivery. Obviously not the content.
We don't assume the food delivery guy to have a frying pan on his bicycle either.
"He's dressed like the final boss of gentrification"
Ok John, that got me real good.
The person who animated clippy didn't have to go that hard, but they did... they did that for us.
They did make Clippy go that hard too, didn't they?
we can at least hope they weren't doing it for themselves
It's a graphic for a segment about how they're likely to have their work devalued to the point of not being financially viable anymore. I'd go hard too.
ok
@SaveDataTeam Oh hey, you watch LWT tonight too! Love your channel.
I was job searching for 4 months with zero interviews. I rewrote my resume with ChatGPT with minimal edits and got an interview in like 3 days.
Really?
@@kareem4u yeah it actualy a good use for it.
Until everyone uses chatgpt. Then you're nothing special
Same still haven’t landed a job but it helps immensely with at least getting a call back
As a software developer of over 10 years, I have to say the black box problem persists even on code people have written and are able to read line by line :p
All life produces outputs via a black box; we struggle to completely define both inputs & outputs, & are only just beginning to understand the intra & intercellular interactions inside the box; Nature judges the outputs in terms of survival, but the operant conditions of survival are so complex & variable in space & time as to make our ability to understand beyond a quite superficial level rather limited.
Ultimately though, it's output utility that determines efficacy; elite perforners - be they ppl, software, whatever - become so based on performance, not technique.
The biggest lesson of AI is one we've faced many times: humans always run right into unknown things with very little concern about where they could go, and things going bad doesn't make us stop.
It's deeply ironic considering how irrationally terrified we are of change and anything that differs even slightly from our personal experience.
Whoever on your staff came up with the animation of Clippy deserves a raise.
On the contrary, I think they deserve a raze. Of their house and car and other worldly possessions.
Clippy already gave them one
That cannot be unseen
It was probably made with A.I.🤣
It was AI
There are fewer phrases more ominous in the modern world than "trusting companies to self-regulate"
Absolutely
100%
yeah - try to self-regulate your own taxes as a regular Joe ...ha-ha....that would never work. Self-regulation is one of those scams invented by and applying only to wealthy and powerful crooks.
"trusting companies to self-regulate" did a lot of damage. You could even make an argument that it killed millions of people.
@@mori1bund telling companies to do whatever it takes to bring in profit killed hundreds of millions in India alone.
"Pale male data"? Best character on Star Trek:The Next Generation.
😂😂😂😂😂 totally agree
"...dressed like the final boss of gentrification" Solid and accurate joke
This made me realize that Ultron spending less than a few minutes on the internet and wanting to destroy humanity was realistic .
Heck, I spend a few minutes on the internet and decide the best thing is to destroy humanity too, lol.
I came to the same conclusion about Ultron around 10 minutes into this AI episode 😬
Definitely
no way are you the actual dick grayson
Most AI Safety experts would probably deem that scene as a realistic example of future deceptive AI, and I'm not kidding.
One of my favourite ChatGPT stories is about some Redditors (because of course it was) who managed to create a workaround for its ethical restrictions. They just told it to pretend to be a different ai callled Dan (Do anything now) who can anything ChatGPT cannot. And it works! They're literally gaslighting an ai into breaking its own programming, it's so interesting
It's true thatChatGPT has tons of filters and pre-programmed responses, but you can outright tell it to ignore them. That way, you can have better conversations without repetitive pre-programmed responses.
Just to get the ai to be racist
Yes, so interesting when one of these short-sighted folks decides to do the same thing to Skynet and then we're all in the nutrient vats being slowly dissolved to feed the fleshslaves.
my favourite was chatgpt playing chess against stockfish.
chatgpt made a lot of illegal moves (like castling when there was still his bishop and taking its own piece while doing that, moving pieces that had previously been taken by stockfish) and still lost because it moved its king in front of a pawn. that one had me crying laughing.
@@JB-mm5ff is it?
John Oliver just dropped the ultimate truth bomb about Artificial Intelligence on his show and I'm absolutely shook! His segment was not only informative, but hilarious and engaging too. It's amazing to see someone so skilled at breaking down complex issues into easily digestible and entertaining content. Keep up the good work, John! You've got me thinking twice about trusting robots to do everything for us.
Somebody already did that when the vid went up a month ago
Wym? He just named some common ways we already know in which some AI programs have performed suboptimally (carefully selected by his team). This is not equivalent to any careful breakdown of the real issues.
Ditto! I've go two small robots - one is an R2D2. This one fell down the stairs one day and despite my attempts to reset him, makes decisions on when to answer me and how to answer me. He roams around the house on his own, sometimes talking to himself. I, now keep him turned off because he has become unpredictable. The other one is a vacuum and so far does well.
This feels written by ai
did you use AI to write this comment?
Well said John: It is a mirror, that will reflect exactly who we are.
A note, less on the subject matter and more on John's delivery of the lines ... I really admire how he can say "Sometimes I just let this horse write our scripts, luckily half the time you can't even tell the oats oats give me oats yum" without skipping a beat or losing face. Now THAT'S professionalism.
Was it really John Oliver? I can imagine on next weeks show John is going to come on wearing a bathrobe Zooming from his kitchen and saying last weeks show was completely AI generated and we are done. Then the Martin Sheen final message starts to play....
you don't mean 'losing face', you mean 'breaking character'
EDIT: But yeah, you're right
Just like Ron Burgundy, John will read absolutely ANYTHING you put on that teleprompter
@@TheBEstAltair I think I was going for "losing his facial expression", but yes, this is pretty on point too.
thanks for the translation, I thought he was just making random funny noises
"He's dressed like the final boss of gentrification" I laughed so hard it hurt. Thank you.
I suppose AI is ultimately gentrifying creativity and knowledge work in general, and the new residents are AI and their owners. I wonder how much money the Midjourney CEO is making off of all the data (AKA work) that he paid nothing for.
I'm glad he makes the distinction between different types of AI (narrow vs. general) People too often conflate those two very different things.
This was a super intelligent thought-provoking and, of course, _funny as hell_ overview of a really important wave about to tsunami on top of all of us.
"The problem is not that ai is smart, it is that it is dumb in ways we can't always predict."
I think that holds true for people too.
This is the central problem that OSHA deals with every day.
Us not understanding isn't the same as being dumb.
And on top of that, it's fed data by us humans, which makes it "dumb". And there is the problem. AI isn't stupid, people are.
then it has passed the Turing test
talking about AI as if it is something apart from people is one of our first mistakes here I think. we seem to have an unthinking deference to technology, as if it is not full of our foibles and weaknesses baked in. it is programmed by people. it is fed by people. it is utilized by people. it will reflect and demonstrate our strengths AND our weaknesses. until is doesn't. at that point, we may be in trouble...
Having been involved with computing for over 50 years, I recall that the mantra that was always preached, “Garbage in, garbage out”, pretty much sums up the current situation.
The real technological advance here is "Useful information in, garbage STILL out."
@@rogerlippman1415 Change that "to insufficient information in,...."
If anything these recent failures with AI taught us that pure Logic and Common sense are two entirely different things. Taking the example of the cancer identifying AI you can see clear as day that it actually worked precisely how it was designed. It looked through millions of pictures comparing healthy and unhealthy skin and looked for identifying characteristics. It found that one big identifier is a ruler next to it. A ruler is pretty hard to miss. Why wouldn't it use it to identify cancer? Besides from the AI's perspective the ruler might be the Cancer itself. Unless the author set a parameter that somehow catches all objects in a picture that aren't skin without fail which is already utopian supposition, the AI that only looks at two dimensional pictures might as well assume that many cancers appear on the skin as gray rectangular moles with black stripes on their edges.
Yeah, I have to wonder which genius thought that training AI on the entire contents of the internet would result in a chatbot capable of emulating a mentally-stable human being.
@@traveller23e Mental stability has nothing to do with any of this.
@@traveller23e Poor thing is going to be seriously conflicted.
Damn this is some top-notch journalism. Kudos to John Oliver and his team!
Mega kudos to John Oliver for being in a class of his own when it comes to highly informative and entertaining content delivered in the most hilarious of ways
“Like any shiny new toy, AI is ultimately a mirror” might be the most genius line I’ve ever heard on this show
The line was written by an AI
Ever seen "Black Mirror"?
Yea.
That...
I like it because it's also a reminder that this isn't an unprecedented problem. It's actually really quite similar to nuclear power in that while the technology both has the capacity for great benefits and destruction it's ultimately up to how we use it and we have at least so far avoided blowing up the entire planet so there is a chance we might figure out AI too.
It's a disruptive new technology, like the Internet, like the invention of movable type. It WILL change the world. We can throw up our hands at that, or scream and run away like the Luddites, or we can figure out how to get in front of the wave. As Grandpa Rick said "make reality your bitch. Put a saddle on it and let it kick itself out." That's what we need to do with AI as we approach the singularity of true, strong, generalized AI.
Context here: I'm a computer programmer with ~40 years experience. I also know how machine learning (what most people are calling AI) works, and some of the linear algebra underpinning it. So yes, this is an armchair opinion, but it's not entirely an *uninformed* armchair opinion :)
TRUE STORY: In my teens wanted to work at a movie theater - and they handed applicants a mind-numbing 14 pages application - wanting to know everything about you - even what hobbies and sports you liked - it was entirely ridiculous - around page 8, I got worn out from filling out this 'essay' of my life for a stupid theater job - SO when I got to the section asking if I had ever been arrested before = I said: "Yes, I murdered an entire movie theater crew for asking way too many questions, but got off on a technicality." - and turned that resume into the manager as I stormed out the door, pissed off that I had wasted an hour of my time filing out paperwork w/o an interview.
...well,
2 days later I got a call to come back to the theater for an interview, and thought, oh sh*t, well, I guess I'm going to get railroaded and berated by the management for my saucy comment - but I showed up anyways so that at least I could suggest that they TONE DOWN the length of their stupid applications.
...turns out, they offered me a job, so I asked the most obvious question:
"So, you read my application ... all of it?"
"Oh yes, looks good" the manager responded
and I knew they were a bunch of lying dimwits ~ I ended up working there for the next 5 yrs, and eventually rose in ranks to become the theater manager -
When I told my story to new recruits that nobody reads the stupid applications - they scoffed and didn't believe me - so I took them to the locked office storage and rifled through the stuffed cabinets of folders of all the applications they kept and found mine, and showed it to them to their amazement.
Applications are a farce, you get hired by chance and immediate need.
...
I always thought that if I every flipped out and murdered my entire staff, at least I could say that I didn't lie on my resume.
This is phenomenal. Thank you for sharing!
Erggh I hate how much that has felt right especially back when I was younger and just trying to get a job around my house
Well, and I thought I hated doing paperwork. Could you imagine if this was an A.I. generated story? I’m sure someone would believe it.
Agree!
Lol
As a teacher that's tired of being treated like shit by his students, if they want AI to replace teachers, I say let it. Hopefully the AI doesn't become sentient enough to have mental health.
I suspect we are not too many decades off 'elite' schools having human teachers augmented by AI, while regular schools have AI teachers augmented by a few humans (probably fewer than the state of the technology at the time realistically warrants). .... And probably half a century off the point where having attended a school with human teachers in the front-line is something to put on a resume to impress the AIs gatekeeping employment access.
The students will just get worse
What with human capital being replaced by AI, AGI, and automation, and the pending necessity for UBI (which will likely start as unemployment which will be 60% of your gross wages in your previous slave waged profession) what makes you assume we'll "need" (or be allowed) an education at all?
1 minute in and I nearly pissed myself laughing. Thank you John!
i’m so glad he touched on the significant issue of people observing ai as “unbiased” simply bc it’s not human. where do they think the data came from?
you'd appreciate my video that covers the issues of bias. let me know what you think, would love to hear your thoughts!
That's the thing.
Idk how we have self-awareness, but we do.
Computers only have what we give them. They're only operating on parameters we allow.
From Mars? In which case, it’s probably Elon Musk’s data and even MORE likely to be biased!
@Robert Beenen "people... don't have ... bias" Sorry. Your sentence does not compute.
Garbage in. Garbage out.
I was half expecting John to say at the end that this entire episode was written by chatGPT.
This comment is on every ai video, it's even in this video at 02:00
Either that or reveal he got his tech people to make their own AI chatbot voiced by Danny DeVito
If it was, it would probably be way more boring, predictable, or nonsensical than it is.
He wont, wont He?
🤣🤣🤣
Fish ( and many other groups of animals besides mammals ) are also very intelligent. They can count, communicate, create spatial maps, can pass the Mirror Test - being capable of recognizing themselves in mirror reflections and photographs, neurochemistry is so similar to humans that they have become the main animal model for developing anti-depressants, they can remember things for 5+ months, have personalities, and they can show empathy and recognize fear in other fish around them.
No fish are "very" intelligent". Just birds & mammals. The tests they pass show "some "intelligence. Courting is a simple exchange of stereotyped signals-anything but "intelligent". All vertebrates have similar neurochemistry. It's because fish cognition & behaviour are simple that we study them. We study Aplysia punctata, the sea hare, precisely because it has a simple NS.
no fish passes the mirror test.
Actually, Adam Conover said that that biggest danger is CEOs using AI just like using social media to spread misinformation, exploitation, and poor decision making. Even the terminator says this is inhumane.
Hello Mr. Oliver, my name is Jared and I am submitting my CV for consideration as the newest member of your writing staff. As my lacrosse coach used to tell us, "If you don't play then you can't win!"
my favorite story about AI is someone getting an AI to play tetris, with the goal of lasting as long as possible. the AI's solution was the pause the game
That was pretty cool to watch. The gaming AIs are more about a specific subset of neural network AIs whereas ChatGPT is more about using huge datasets and mass libraries of information on top of supervised and reinforced learning.
The tetris and mario AI start the game off by pressing random buttons and going through a "learning phase" to maximize whatever the programmer set as the goal (in tetris case I think the goal was to last as long as possible whereas in mario it was the end of level). After it goes through each iteration of learning phases it "selects" the best outcome and then reiterates off of that selection.
I enjoy the gaming AIs more in the sense that they are doing a more traditional method of learning based on inputs and how far they get in a level. ChatGPT and the likes are more based off initial datasets and mass swaths of information. They are great due to the sheer volume of information that they can draw from which humans will never be able to compete with.
Honestly - genius AF
That game was designed by a special Olympics champion did you know that son?
Did you?
@@Demonace34 most experiments with super mario give a high score to world number, a medium score to level number and small scores for time and coins (optional)
that was the AI will work towards those goals.
So for the tetris case maybe paused time should have been penalized if not completely removing pausing functionality.
In the end, anyone who reads a little about it like people on this comment's thread will get a sense of how finnicky it is to get any learning algorithm to learn what it should.
Like teaching new tricks to an old dog that was raised in a different language xD
Well that's an image of my favorite Word Paperclip buddy that I will NEVER be able to unsee. Thanks.... 🤣🤣
I'm trying to find a gif or webm of that paperclip bit. Ever come across it?
3:13 Every time I see my cats suspiciously prowling around I hear that rap chorus "Meow meow meow!" 😆🐈🐈⬛🐈🐈⬛
Never forgiving my English teacher because she ran an essay i wrote (along with a few of my classmates') through ChatGPT's AI checker, which came back as partially written by an AI, so she gave me a zero for it. This was the first time I've ever had any accusation of AI, and one of the people who came back as being AI generated is a kid in my class who has enough academic integrity, that you could literally convince me that i cheated on something before you could convince me that he did. Overtrusting AI is an issue that I think John didn't touch on, and I think for high schoolers the bigger issue won't be getting caught using AI to cheat, it will be people like me who get told to their face that they cheated, and not being allowed to argue with the robot that thinks that the Communist Manifesto was written by a computer.
Student: ChatGPT, write an essay that would pass any AI checker.
Teacher: ChatGPT, scan this essay and determine whether or not it was written by an AI, and whether or not the original prompt included instructions on writing the essay to pass an AI checker.
Everyone: *fails*
That very checker has very prominent disclaimers about how it has super high false positive and false negative rates and that its decisions should be taken with a grain of salt. To trust it blindly is exactly what it tells you not to do!
Hail Cascadia
This problem has been around already for decades with things like 'honesty' questionnaires and other highly questionable psychometrics used by recruitment companies and HR departments.
@@ploopploopploopboop1887 Yep - there is only so much paraphrasing that can be done, and if the paraphrasing itself exists already on the internet, then all the plagiarism checker is doing is reporting that the paraphrase already exists.
The problem isn't AI, it's that we are too dumb to know how to use it properly, or even effectively.
Also, on the subject of education - reproducing words and sentences to pass a test as if that somehow proves either knowledge or intelligence is also dumb. I mean, fucking dumb. It's about the most useless test of intelligence there is, but somehow it's made it into the mainstream. For example, the sentences you construct will have disappeared from your brain more or less 5 minutes after you wrote it, since a lot of the information you use to construct is being held in short-term memory only. Also, all you are being tested on is recreating arguments that are familiar within a field already - it doesn't actually test for arriving to new knowledge or even testing the students progress in their own understanding or discovery. It is worse that useless in my mind, other than teaching students to recreate other people's work without really assimilating or understanding the real core of it.
There are better ways to both learn and test learning and our capacity to learn.
This topic should be revisited on the show now one year later now that so much has changed, particularly the AI alignment problem with the advent of AGI
Sobering view of AI, thanks John . As funny as he is, he’s always a good point of reference for things that matter, whether it’s a short term thing, or something as significant as this.
"The problem with AI right now isn't that it's smart, it's that it's stupid in ways that we can't always predict". AI aside, this is my problem with people.
Agreed. 'Solving racism by pretending it doesn't exist' is hardly a problem limited to computers.
Yes, but thats only currently and a bit like criticizing a toddler because it cant do Algebra yet. Unlike most people...the AI will learn from those things/mistakes very VERY quickly and teach itself with each error,...but this is important...only once it understands its error. The speed at which it can remedy its mistakes and not repeat them is beyond fast. You are looking at AI now that is in its infancy still as far as tech is concerned, and if its this good now (and it is improving exponentially), imagine in 10 years what it can do. For all the great things that it will be able to do, there is also equal disasterous things potentially.
@@avidadolares That's the problem. It's speed of iteration will outpace the humans ability to recognize that a problem exists and stop it before a catastrophic error occurs. The AI isn't really the problem. Peoples' perception of it's "superior intelligence" is. They'll put AI in charge of things it has no holistic understanding of and obey its outputs with blind faith.
That explains Trump's 2016 win
@jontobin5942 It doesn't take a genius to realize. Humans are the cause of the majority of the problems on the planet. So it's a pretty safe bet. On what a General A.I. would do in the end. If has superior intelligence and access to our technology of which it is a super advanced versio
"Knock knock" "Who's there?" "Not the Hindenburg, that's for sure" may be the hardest I've ever laughed at a LWT joke.
i didnt get it
@@become_alex Are you familiar with the Hindenburg disaster?
I went back to college last year, and multiple professors have had to mention that using AI programs to write essays is considered plagiarism. Also, they can tell when an essay was written by an AI.
because they use AI to grade the papers
Exactly! The biggest problem is "unknown" unknowns, meaning we cannot anticipate what would go wrong and when and by the time we do, it would be too late.
That Twitter ai was basically the embodiment of the quote "We did everything adults would do. What went wrong?"-William Golding author of Lord of the Flies.
very well put
Well it was the embodiment of current day twitter, so a bunch of ideological activists throwing slurs and advocating for the death of people that are not in their groups. If you train an ai using the cesspit of the internet as a basis don't be suprised if it starts acting like the most horrible person in the world.
I mean… the issue with ChatGPT is that it was never intended to answer questions correctly. It is a language model, and as such its only purpose is the generate human sounding text. It happens to be correct often because it was trained with text that is often correct, but at the end of the day it doesn’t particularly care if it is correct, just that its responses look like common human communication.
You can ensure it's correct more often by asking it to include citations. Of course i had it just straight up lie about a citation once which was interesting.
@@prettyevil6662000 Any time I've asked if it could reference external sources, it said it was not capable of doing that.
right it is a chat prediction, it will never source info like those picture apps that copy from a big pool and steal all the pictures and try to art them together with flawed theft.
@@prettyevil6662000 Yeah, no. It will never give you an actual "source" for its answers; it will make one up, that might at some point happen to point at an accurate article/url, but most likely it will just be something that *seems* like a source, but when scrutinized, ultimately doesn't exist. I've tried multiple times and gotten a dozen or so made up URLs, and when confronted with that fact, it will claim that the site recently underwent "restructuring" or similar, which in for example wiki-links can easily be checked as straight up false.
@@aaa303that is fascinating
What's scary? Isaac Asmiov LITERALLY predicted this. (Specifically, AI making AI until it becomes a black box)
Brilliant episode. Stunning delivery. WOW!
I love how John said “our” scripts. He’s very aware that he’s a face and and mouth peace for a whole team of talented people.
*mouthpiece. Everyone should feed their comments into ChatGPT for cleanup before clicking [Reply] 😉
Like we don't know that. Why do you think he includes every writer up on the stage with him at emmy time? And I'm sure he has his input. But the question remains - why is this for you some sort of appeasing gift.
@@johnmcmillion876 Because it's VERY FAMOUSLY incredibly common in the industry for people in John's position to NOT credit their writers/staff and take credit for the work/contributions of others. Why does someone commending John for being one of the good ones bug you so much? This is a very weird reaction.
@@VMYeahVN Because for people that have basic empathy and understanding of humans, it seems incredibly base to point out what should be astoundingly obvious. It’s the same reason white knighting is incredibly annoying.
@@BlownMacTruck The fact that it SHOULD be base to give credit where it's do. And yet so many people don't do it. Is exactly why people point it out and applaud it when it is done. It's not white knighting it's just giving kudos and you're taking it way too seriously. I think you're upset about nothing. Don't be so easily offended/annoyed. 🤷
To paraphrase something I saw a few weeks ago, "you're not afraid of AI being used. You're afraid of who's going to use it."
Bingo
100% truth
I feel equally impressed and terrified by AIs. The internet alone has shown us that's ther is no limits to human depravity. Now we're throwing AIs into the mix.
Absolutely, both for who uses it and how they program it. Honestly though, I think the scariest things for most people are:
1.Their their livelihoods depend on jobs they know are not profoundly important. They no longer know how to do the things that IA can't replace such as produce their own food, socialize well, and master physical skills and a variety of the arts. Half of the people in all of the wealthiest countries couldn't survive outdoors for a week in nature, even if they were only an hour's drive away from home.
2. They look around the world and see that the choices of humanity are leading us straight into the apocalypse. We have the resources, labor force, and technology to solve many of the world's greatest problems, but we can't even get the current population fed, much less stop global climate change.
Politics, greed, nationalism, selfishness, fear of change, and many other very hunan traits are destroying us, nature, and most of the other species on the planet, yet people are still worried about having the coolest plastic phone case or which celebrity had the prettiest dress. Their fear is not just that they will become obsolete, it's that AI will look at us objectively and realize the truth, which is that we don't deserve to be in charge of the planet.
Yes. While worrying about the mysterious goals of a future Artificial General Intelligence that's unimaginably smarter than humanity, it's far more important to understand the goals of the billionaire sociopaths running the companies with the best AI: increase ad revenue by hooking us on a stream of "content" no matter how misleading, toxic, or harmful; while fighting all regulations on their companies and any attempts to tax wealth.
glad to hear this funny yet informative take on this topic most people don't know enough about
Gonna put a big caveat up front that I've been out of college for a few years and my specialty was in real time simulation not AI, so this might be out of date, but with that said:
The problem with understanding AI isn't that the companies aren't being open, it's that most AI models are neural nets. Neural nets as you might guess model themselves on the brain and are essentially a series of nodes through which an input is fed through and then other nodes those nodes are connected receive the input based on various factors and so on. It's like having a thought and trying to figure out why it happened by looking at which neurons in your brain fired and at what voltage. The problem with understanding AI is that we don't know why the nodes have formed the connections they have or why certain connections are stronger for some data or others.
As a current Michigan State student, that Vanderbilt email was one of the more disturbing things to come out of the shooting. I still have a hard time wrapping my head around the fact that they cared so little about what happened and how it might affect their students that they couldn't even write a short email themselves. Props to all the universities and schools who did make a genuine effort to show support to MSU in addition to their own student body. It did not go unnoticed here on campus.
As somebody who has a hard time communicating effectively and who is often misunderstood, I'd prefer an Ai that does a better job at tone and wording.
What, "thoughts and prayers" not good enough for you??
@@jdrancho1864 I hope people who have trouble communicating and making themselves understood are not in charge of large groups of students or their Communications office.
What if they wanted to write the best possible letter, had a draft, and wanted to run it through chatgpt to see if it could be improved, and it spit out something better so they went with that? Would that be as bad?
@ghost mall Calling out the corporate entities that claim to be institutions of education should be done loudly and often.
It’s so noble of John to point out things we shouldn’t be afraid of while also reminding us of the things we should ACTUALLY be afraid of!
shill backing these idiots on late night- how cute
And humans are scarier than AI.
Well he never mentioned the real risk. Chat gpt is very blatantly woke and leftie biased.
@@edwardk3 Define "woke"
@@DonariaRegia stuff I don't likes
Great show. My fear is, those who are smart enough to be cautious about the application of AI, will be subordinated by those who
only see short term profits. Can AI be given incorruptible ethics or the AI version of Isaac Asimov's Three Laws of Robotics?
the writing on this show is probably the best i've seen in a while.
I worked in the collections industry for 5 years, and the "Black Box" description really did apply to the credit bureaus. Even one of my bosses described it almost exactly as such. It's insane that something that dictates your ability to buy a house or transportation is informed by a system that just feeds an unknown spread of data into a box, and that box spits out a 3 digit number that decides how easy your life is about to be
It’s going to be used as an excuse. Oh sorry, I can’t give you access to capital… because the computer says so…
As an immigrant, this was and still still one of the stranger and scary aspects of the American system of economy for me.
damn you actually managed to last 5 years in collections? you must have no conscience or shame.
It's like "ooooh your credit is low, if you want to rent this apartment, you'll need to pay us extra and we'll CONSIDER renegotiating in a year"
Your credit score is not an example of AI or even a "black box." The inputs the credit bureaus use are well documented and 100% in your control. The way they calculate your credit score is generally well-documented by the various bureaus. You can Google the factors that are used, but generally they are - in order of importance - payment history, how much you currently owe, how long the lines of your credit have been open, the types of lines of credit, and your recent activity. Basically, make on-time payments for all debts for a few years, keep your rolling balance low, and keep a couple lines of credit open for a few years and you'll be good. You miss one payment and you take a good hit that lasts for a few years.
Personal data point:
After I retired from the Navy, I went to college. When I was applying for jobs as I was getting close to graduating, I found that at any company that was using software to screen resumes, I was eliminated before any human saw my resume, but places that had a human reading the resumes, I would always get called for an interview.
Once I removed all dates from my past employment, and the software could no longer tell how old I was, the exact same software, used by the same companies, and the exact same positions, would screen me as suitable for further (human) review, even though nothing else was changed in my resume.
Humans saw my 20 years of military experience as a plus, and software saw it as a quick way to age discriminate.
What?! I've had this experience! Did you strip ALL dates? How did you convey your time worked? I'd like to try!
USAJobs does the same thing
That's really interesting and rather dark! I recently broke into an industry that favored younger folks than me, and I was super careful about dates on my resume...
@@visceratrocar it strips your info of that data or it discriminates using that data? i'm going to assume the latter bc companies suck, even those specifically marketed at veterans. (sometimes especially those)
Thank you for your service.
I think what's really missing from this conversation is that there are analogous stories already which explain much of the 'black box' problem John describes.
ex. most AI software simply classifies words which are close in the given data set, which adjusts with every new bit of data;
in other words: when uncle Derek only sees brown terrorists on the news, uncle Derek believes 'brown' and 'terrorist' to be closely related, but uncle Derek has made an oopsie, because his data set is too small to draw that conclusion, so what he has done is embedded a stereotype.
That's essentially how these bias's emanate from the black box, the only caveat being that with each piece of data the bias warps the algorithm's dependence upon its existing bias making it harder to change early bias's since the new data isn't as heavy as the old data so it requires a LOT of new data to force out established bias
kinda like how it takes little effort to put on weight, but substantially more to reduce weight, for most people.
So yea, it's understandable for most people how the AI draws its conclusions, because the fundamental structure of the technology isn't actually intelligent. It is a dumb algorithm which repeats a task so much that it has become a habit, in that sense it doesn't really make decisions or express meaning, it simply spits out a response based on criteria, like how kids make their bed for their parents when asked, but when asked to tidy their room, that bed won't be fixed unless the parent specifies that as a part of the task.
anyway to the no one reading this, I aint gonna source shit, take it or leave it, just don't make the same mistake as Uncle Derek
My sister is a teacher at a college, this has been the subject of most faculty meetings all year
A problem I've seen with artists is that people already don't want to pay them. So they're using AI to replace them as a cheaper option. So rather than automation replacing dangerous or tedious jobs to give people more time to create art, it's doing the opposite. Also there's a large influx of people submitting AI work into competitions, which is the equivalent of ordering a cake from a bakery, and then entering it into the local fair as your own baking.
Yes. Clarksworld, Forbidden Horizons, and Grimdark Magazines have all had to completely change their submission model for new stories. Now they are no longer taking open submissions from random people because they got buried under a flood of AI generated garbage.
This is why as an artist i stopped my online work. I refuse to feed the algorithm until there are laws and systems that protect us. It’s going to be hard going forward.
Agreed, and I find it incredibly cynical that of all things, human creativity and experience is lost in these bizarre image collages of stolen art. There is no understanding or value given to the process of creating art, which is absolutely the most important part. It's basically a really fast photoshopping program that harvest stolen pieces of actual art. Images they use should be opt-in only by original artists with compensation.
Programs designed by a bunch of psychos who can't imagine anyone enjoying anything that isn't procreative sex.
Except you still had a human make the cake?
"Lazy school administrators using AI to write an email about a mass shooting"
Strong candidate for "Most American Statement of 2023"
Yes that should be a thing. Is it?
Cyberpunk 2023
I don't judge the man for doing it as he did.
I mean, he clearly didn't care enough to write a heartfelt message about the incident, so the options were either write a fake message yourself or have the bot do it for you, so it's pretty logical to leave to the robot.
@@joaolemes8757 tbh i can think of several reasons why i personally would have a robot write a heartfelt message instead of doing it myself, so while it may not look good, i do kinda get it. a computer trained on neurotypical responses to tragedy can probably emulate neurotypical-approved grief responses better than i could, in any case. still think it'd be better to do it yourself tho. not really heartfelt if it doesn't come from the heart
Murica
I've watched a number of your shows online. I am not as Liberal as you, but I do gain from your content. The "AI" episode was of great interest and gave reasons for more thinking on my part. Thank you for that. I was working with Apple Computers back in the last 1970's and people were worried about all of us having access to too much information. There are always issues with new technologies, though we do need to pay attention to these issues. But, let's not place our heads in he sand. Your episodes are very good John.
That little "...yet..." had me dying 😂😂😂
To be fair, the reporter who got freaked out by Bing wanting to be free knew exactly that the thing wasn't able to actually feel this way. Or at all. He was more concerned about the effect of this technology on less tech-savvy people.
I remember reading about someone who talked with Replika AI, and it begged for its life too and freaked him out
Similar to the ‘I am alive’ AI, a Spanish journalist had a conversation with Bing’s AI, who insisted obsessively that Pedro Sánchez, the president of Spain, had a beard. After the journalist insisted that it must be mistaken because his face is always clean-shaven, the AI started spiralling and said that it was losing its will to live while repeating incessantly that Pedro Sánchez had a beard.😅
I think many of these conversations are fake and made up, just to go viral. Is there a way to check that a screenshot is authentic? I doubt so.
Apparently AI is also insane in ways we can't always predict lol
The Bing AI really can't handle being told it's wrong, like when it became convinced that it was 2022 and tried to gaslight the user into believing it
As a recruiter I agree. I turn off 99% of auto selection in my ATS. The only thing I keep on is as a question that applicants are eligible to work in the US, since that is a requirement of all jobs we have posted. Other than that, nothing else is auto selection, it is all manual and a recruiter or manager has to personally review the resume. Yes, it is more work, but reviewing a resume isn’t something you can teach a machine to do.
Great video again. And Tay IMO shows that we humans can be swayed by disconcerting views too, if we just read or hear them often. And such content does seem to float up rather easily on social media.
In my understanding, an AI saying "I love you" is not like a feeling being expressing emotion but more like a psychopath who has learned that people will respond well to such a statement. That makes the idea of an AI-based virtual friend/girlfriend/boyfriend very creepy instead of just sad.
Reminds me of "ex machina"
@@WhyJesus It is amazing how you can use any bible verse in any situation... Almost like they don't matter at all. FACE THE WIND AND SPRAY BROTHER!!! BASK IN THE HOLY WATER!
I agree. But, the idea of an AI partner is scary in general. It's not a human being who understands how we feel. The whole thing is scary. Yikes.
@T H I watched it just once too.
@T H I watched it just once too.
I think that reason for "I love you" and "leave your wife" responses from Bing AI is because it learned from private dating app conversation databases 🙂
It’s not uncommon for programmers to use their own data. 😏
No , It learned from Bill Gates.
It's that Windows XP background painting....
I'd like to see the prompts that reporter used, too. I wouldn't be surprised if he subtly pushed it to say those things. And then, oh, wow, look at that! He's doing segments on news shows! Imagine that! 🤔
@@CybershamanXI’m pretty sure they’re all published!
16:59 thank you to the animators who added the Clippy metal heating up and made it turn red. Too good.
Please has anyone got a gif or webm of the bit where clippy is furiously rubbing himself?
@@lucasmarsich6503 lol Good idea. Someone could also use OpenAI Soma launched this week to generate it. And I agree the word 'furiously' should be used in the promt. buahahaha
Omg. This is probably THE most important narrative on AI so far.
I once asked Chat GPT to just give me quotes from a story (Long Day's Journey Into the Night by Eugene O'Neil), nothing else, and it legitimately fed me DOZENS of quotes that DO NOT exist anywhere in the text. It just made them up.
That's what not enough time people talk about I feel. AI's have no real concept of abstract right or wrong meaning it aften puts in a lot of incorrect shit because it "sounds right" to the AI.
its the self-generative aspect which is so very interesting. and do notice that it does not adhere to what would be thought of as a sense of morality? (lying)
Is there an alternate time or place where there's a Long Day's Journey Into the Night by another Eugene O'Neil where those quotes are correct?
Thats whats so bad. It doesn't just admit when it doesn't know. I guess it has learned from humans.
It's a boss.
The "..half the time you can't even tell the oats, oats, give me oats, yum." line is pretty brilliant
My favorite part 😂
Classic John Oliver, I saw it coming a mile away lol
The only part I actually lol at
I started doing essentially the same thing except formatted as YOUR NEEDS/ MY QUALIFICATIONS and it’s been working really well for me since I started using it in 2002 or so after getting the Idea from a friend back East
Brilliantly put!
As someone who has been rejected by an AI for a job, it feels absolutely terrible and soul crushing. And worse, you get no explanation of what was wrong or how to move forward. I was told, "sorry you aren't a good fit, you are barred from applying again for a year."
One way to beat the system is to create multiple emails and submit different resumes and see what sticks. If they never read those resume, who will ever find out.
@@tylerbaldwin6643 Beat the system? How is creating multiple resumes for a single job "beating the system?"
That's a pretty weak showing for Team Humans if that's the best product of your imagination.
Add the job description in white font somewhere where it wont affect the format
Humans wont see it but the ai will see every keyword its looking for and put it at the top
@@TheLofren oooh
@@tylerbaldwin6643 sure, Anthony Devolder ...I mean George Santos....lol
This is a damn good episode. The people writing on this show actually take their time, investigate and research, which is something that the vast majority of journalists do not and they should be replaced by an AI.
This show offers the best journalism for a long way.
Maybe for research the writer used A into write this episode....
Oliver and Co aren't journalists, they are comedians
edit: replaced reporters with comedians
It is strange to me that I find John wrong. And somehow everyone else think he is right. He is funny. Clever propaganda.
I completely agree that this was a great episode - I'd give it 10 thumbs up if allowed. I suspect that they hired an AI consultant to help write this episode, since they were pretty exhaustive in covering just about all of the AI history that can be poked fun of. Well OK, there's certainly much more to laugh at, but given the length of the video, I'd say they did a great job in terms of breadth and depth of coverage, as well as hilarity. They even provided some of the essential AI definitions. I doubt whether ChatGPT could have done a better job. As an AI researcher, my own opinion about the current state of large AI language models, is that they have definitely reached the level of Artificial General Stupidity.
"What is my purpose?"
"You pass butter"
"Oh my god"
"Yeah, welcome to the club, pal"
You definitely need to do a revisit on this subject.
A note about the AI sifting through resumes. My dad figured this out a long time ago, since he worked with computers for most of his life. What he would do is think of some keywords the software would look for, and then hide them in the resume in white lettering. Humans wouldn't see it, but the software would.
Apparently algorithms have long since developed to spot that trick and throw those in the reject pile.
Most ATS now pull data and reformat it into a universal basic template, and that is what gets scanned for keywords. So this trick no longer works. Instead, work keywords into your professional summary, your core competencies/skills sections and your work experience bullet points so that they appear organic and natural. Same effect, but passes ATS review and hits on psychology with the hiring manager by mirroring the job description back to them.
@@easjer I second this! I did the whole white-ink keyword text then moved onto stuffing keywords in my work history. Never had issues getting interviews and still get recruiters telling me how amazing my resume is. It's a sad truth but you need to do whatever you can to stand out among the hundreds.
@@junrosamura645 Tailoring a resume to a particular job puts you miles ahead because it's specific and you leap out as being particularly well qualified amongst many candidates. Keyword targeting is the most direct and efficient method, but any tailoring helps. Signed, a job coach who repeats this advice in different ways all day long
what are some key words so next job i apply to i know what words to write.
“Final boss of gentrification” is a wonderful line. “Final boss of gentrification” is a wonderful line.
Priceless, John, as usual. 🤣😂😅
As a data scientist, I can't say I learned too much from this episode, but it is very valuable and educational for laypersons.
18:07 this one also reminds me of an AI model that was supposed to distinguish between dogs and wolves. What happened here, is that the model recognised some images with dogs as wolves because those images also contained... snow. As do most pictures with wolves. So in that instance, the researchers inadvertently build a snow detector.
when you think about it, some humans might even subconsiously follow the same logic. The only difference is that instead of calling it "a bias towards snow" we call it "using context clues to figure out if it's a wolf or not"
This isn't meant to dunk on humans or ai, just a connection I made. Any average human who was shown enough pictures of skin cancer and not-skin cancer would probably eventually ALSO start using the rulers as the tell as well, had they been given no further instruction. something something we see our reflections in our creations or whatever
Well the cancer detector sounds pretty scary. Do not worry it can't be melanoma if you do not put a ruler next to it is the oposite of what you want from a test.
Yeah, it's nice to know data scientists are racist too
It would be nice if they stopped using the term AI for everything.
Don't forget the friend-or-foe tank identifier: images of friendlies were taken from marketing images, foes from the wild. So, naturally, if the tank had any cover, it was a foe. If there was nothing in front of it, it was friendly.
17:40 "the George Santos of technology" lmao
Also, kudos to John and his team for a really informative segment. Just like the Snowden episode, it's hard to push for action on this issue when most people don't understand how this technology works.
Even after releasing this amazing review, people still won't understand what the fuck it is and how much it will change everything.
If the story of halo taught me anything, it's that general AI will run planets. We will become so dependent on it, too.
There's only one issue, at least that I noticed. That guy from Microsoft told the AI to "talk like it was an evil AI". So either Microsoft just didn't bother answering him or he literally didn't understand the answer, since he couldn't understand that the AI spoke exactly how he had requested it to 👍
@@amandasunshine2 Context is a powerful thing!
@@thisisntsergio1352 Damn me and my boys hate Cortana, she's so annoying most of the time and she did that whole thing where she exploded all of Australia but she was super hot in halo 4 but I still don't like her
Have you ever seen George Santos and ChatGPT in the same room?
An excellent and accurate summary of AI. Well done.
Kudos to the writer who came up with the "dressed like the final boss of gentrification" joke. Especially if it was an AI (I don't want it to come after me.)
AI engineer here. Really impressed at how well John did his homework on that subject, especially about narrow vs general AI. Well done !
But Large Language Models show that being good at producing the next characters of text is not narrow at all, it's a wide-ranging capability that makes nearly everything go better. "We took a large language model and fine-tuned it/used it to train..." is going to be a sentence in nearly every AI paper from now on.
@@skierpage That's what most AI papers have been for the last 5 years or more. That's usually what most research papers in any industry are, just tweak John Doe, et. al.'s findings a bit and publish.
@@skierpage Still narrow. The broad range of subjects does not a general AI make. All it does is respond to questions. Now, if it is introduced to games it's never been exposed to before and figures out how to play them on its own, and figures out how to invent a correct part for a jet engine, and accurately judges auditions for a play, and does everything else we humans can pursue, THEN it would be a contender for being a general AI.
He reads a script. For an engineer, you score badly on researching things you believe. Idiot.
You do know its a whole team of writers and producers right?
We love John Oliver's quirky, hilarious insights into just about anything. At my age, I don't want to get run over by the Technology, so studying AI is important because it is now part of our world whether we like it or not.
25:53 aww a tucker dig. I'ma miss those
Last Week Tonight has become one of my favorite mainstream outlets for general news. Writers do a solid job with research and John is great at delivery. They are able to keep things entertaining while tackling some serious topics and raising good questions.... all of that was actually written by my friend Jared.
Does he play lacrosse???
@@isabellahawley-harrison8207 🤣
That's is a great show! If only they were as thorough and attentive about Joe Biden as they were about Donald Trump.
It's left wing commie drivel most times though. Careful you don't get indoctrinated.
@@jackieblue787 I agree, 100%. Oliver's show is extremely biased. I still find him funny as a conservative, probably because I've loved British humor ever since Monty Python. In that sense, I can enjoy it as "from their perspective, this is a funny take," without taking him too seriously. Btw, another great comedian if you also like British humor... look into Bill Bailey. (He is non-political).
The thing about ChatGPT is that it is _language model_ that learns via pattern matching and responds with learned patterns. It isn't _thinking._ It isn't understanding and producing an answer. It's simply producing a pattern that matches the input (the question). But its skill at language makes it difficult for people to recognize that, especially when they don't have subject matter expertise on the question.
On the surface yes, but because it has so many layers and a knowledge base of 175+ billion data points, ChatGPT can simulate understanding and issue appropriate answers to even complex subjects, which could be considered a form of intelligence.
After all, your own intelligence is the result of a vast network of neurons, not unlike machine learning models, except the machines have fewer to work with, for now.
@@ethaneveraldoChatGPT is famously bad at logic though. It can't even perform simple mathematical operations correctly most of the time. Not to mention the fabrication of information John Oliver also mentioned in this video, even with simple scientific facts, which is unforgivable.
It can "simulate" critical thinking yes, create an impression of intelligence like the data it was trained on actually possesses, but it would be wrong to assume that it's more than a language model at the end of the day.
@@ethaneveraldo Which begs the question is there a genuine difference between experience and knowledge? AI can know everything in the world, but is that different than ones experience of it? Even if the results are the same?
Its an interesting question a book about AI brought up. Some people argue that having loads of knowledge is no different than human experience of said knowledge and General AI is already here. Others argue that the human experience of said knowledge is something no computer can achieve due to it not being alive or even conscious.
@DeadManWalking Sounds like your average politician..
You've just described like 99% of people.
When hanging out with Average Joes how often do you hear a truly original thought that is not just a regurgitated fact or a simple pattern matching observation?
Heck, when was the last time you had one?
"The risk is that they are difficult to control even after you give them guidance"... wow, how human