AI HYPE - Explained by Computer Scientist || El Podcast EP48
Вставка
- Опубліковано 16 тра 2024
- Join El Podcast Host, Jesse Wright, in a thought-provoking conversation with special guest Dr. Emmanuel Maggiori (computer scientist) as they dissect the reality of AI applications amidst the prevailing hype. Drawing from his extensive experience in the tech industry, Emmanuel sheds light on the nuanced challenges and ethical considerations surrounding AI implementation. The discussion navigates through real-world examples, illustrating the importance of practical, problem-solving approaches over exaggerated claims. Listeners are encouraged to reevaluate their understanding of AI's capabilities and explore ventures that offer tangible value. Tune in for a candid exploration of the intersection between AI, business, and genuine human needs.
Subscribe now and join us for this engaging and informative episode!
=========📚📚📚=========
CHAPTERS
=========📚📚📚=========
00:00 Intro/Start
01:01 Tech Jobs are Overstaffed
10:02 The Boundaries of AI, Machine Learning and Self-Driving Cars
19:22 Bill Gates, Elon Musk & Decoding the Motives of Tech Giants
23:57 From Chat GPT to Skynet
30:01 Career Paths in the Age of AI
40:26 Unpacking AI Research Biases
43:36 AI Girlfriends
48:34 Good Enough vs. Excellent Work: Thriving Amidst AI Transitions
58:44 AI Fears: Surveillance & Censorship
01:05:30 Amazon Fresh and AI Deception
01:12:13 AI…More Fantasy than Fact
01:19:48 Investing in Real Solutions
Special Thanks to Dr. Emmanuel Maggiori
BOOK: Smart Until It's Dumb: Why artificial intelligence keeps making epic mistakes (and why the AI bubble will burst)
a.co/d/6jt4V9E
WEBSITE: emaggiori.com/
Linkedin: / emaggiori
LIKE AND SUBSCRIBE TO EL PODCAST! 🙏💖
=========📲📲📲=========
FOLLOW US ON SOCIAL
=========📲📲📲=========
EL PODCAST CLIPS 👉 @elpodcastclips
RUMBLE 👉 rumble.com/c/c-2180860
INSTAGRAM 👉 / elpodcastmedia
TWITTER 👉 / elpodcastmedia
=========🎙🎙🎙=========
DOWNLOAD PODCAST
=========🎙🎙🎙=========
SPOTIFY 👉 tinyurl.com/elpodcastspotify
APPLE 👉 tinyurl.com/elpodcastapple
GOOGLE PODCAST 👉 tinyurl.com/elpodcastgoogle
AMAZON MUSIC 👉 tinyurl.com/elpodcastamazon
#elpodcast #elpodcastmedia #TechIndustry #AIReality #PracticalAI #EthicalAI #RealWorldApplications #TechHype #AIChallenges #Entrepreneurship #InvestmentAdvice #BusinessSolutions #ProblemSolving #AIInnovation #EmergingTech #TechEthics #HumanOversight #AIvsHype #AIInvesting #PragmaticAI #AIAdvancement #AIUnderstanding #artificialintelligence
I have worked in Tech all my life and I have never seen someone be so honest about what the industry is actually like as an engineer or developer good work
But this guy knows nothing about AI. With all due respect, he sounds like a fool. He does NOT understand how LLMs work. they are not coded liek expert systems, they are grown from data, we have no idea how they actually operate and hence why they make certain decisions
@@TheManinBlack9054and your reasoning is why we can't ultimately trust their output beyond providing entertainment value. If it has any level of importance you have to double check anyways.
Why is everyone complaining about the "shortage of good developers" if the job is so easy? Why are companies putting candidates through 3 or 4 rounds of technical interviews if you only need to work 6 hours every month? If this job was so easy they could save everyone a lot of time by simply hiring anyone on the spot to work 6 hours in 5 months.
Also, it takes years to hone your skill at almost anything, whether you're a mechanic, plumber or architect but the tech industry is the only one where people with tons of experience and knowledge are considered "overpaid". You can't be serious?
You know he does not understand software development if he thinks you can change two paragraphs of code? in ten minutes. Did he run any tests, does he understand the system? probably used to altering code on Jupyter Notebooks.
Nah, he took one (let's assume not made up) example of him in investment bank and extrapolated it to entire tech industry.
3 hours of work in 5 months - oh please tell me the name of that investment bank, I'd love to work there because over the last decade all places I worked in (all of them big and well known companies) I and folks around me were constantly overworked and on the edge of burnout.
25 years in IT and I confirm 100% what this fellow has said. The stuff I have seen....1GB spreadsheets that require guys working 24/7 to make sure it doesn't cras. AI is still a very very far fantasy for most businesses. In the 90s UML tools were supposed to replace developers...yeah right.
Only 1 month later, and your comment has aged like milk. Your 25 years in IT just show that you were wasted expenses, because you definitely have no idea about the field or AI.
@@OnigoroshiZero I studied AI while you were still wetting your diapers kid. AI is a fraud.
What a sober, mature approach to these developments ❤
I doubt the next breakthrough in AI will be discovered by some guy in a garage - the problems that need to be overcome are massive and not even really well understood. A lot of the hype around AI comes from anthropomorphism and sci-fi fantasy.
"sci-fi fantasy" you're talking about AI systems on your computer using the internet, you're already livin in sci-fi by a wide margin
@@TheManinBlack9054 The technology we have now is more appropriately called Machine Learning and not AI - but I get it's definitional. I've never once felt that any of the tech I have used is intelligent in the sense it can reason or act with any agency.
That breakthrough comes from anyone who manages to persuade dogmatic idiots that AI does not start for "artificial intelligence", because it is not intelligent.
So, it can be done by guy in his garage. Then industry starts to focus on meaningful research. To understand what "human like" principle is represented by model instead of "intelligence". And once they do understand, they'll realize that tasks like driving cars are not suitable for this kind of self-arranged spaghetti code. But there are tasks which are suitable.
2nd breakthrough which may come from garage is network-collapse into tiny one doing same thing as big one, but with lower computational requirements.
Point taken but it neglects human creativity (free!) which, I think, is key to the next steps in understanding the problems. Going back to the 1890s, the next great breakthrough in physics came from a patent clerk. He needed zero investment dollars.
This is a problem. We make negative predictions which seem OK until one of those unknown unknowns comes along.
@@zotriczaoh7098 I get the analogy, but all discoveries build on knowledge from before, and subsequent work builds on that. Even Einstein's theories didn't solve physics - we still have the elusive "theory of everything".
This interview is just great. Not only because of the AI analysis but because of all extremely bright statements on Scrumm, Self driving, how the soft industry worked last decades because of low interest rates, etc. This content is a sample of genuine natural superior intelligence. Thanks for that!
I work in tech as a Machine Learning / AI Engineer and I gave up looking for fulfilment after my 5th job role. I earn 6 figures, work from home 5 days a week and only work ~3-5 hours per week. No joke. Nothing new. Like seriously, the most little task that can be done in 1 hour takes 5 sprints (1 sprint = 2weeks). Like bruh, it's 10 lines of codes..
and what's the problem? lol or are you braggin'
@@ricardogarciarevilla6922I feel a sense of "Is this it?" An easy job can feel like bullshit and devour your soul, material compensation is just one aspect of job satisfaction.
@@ruffethereal1904 You are probably depressed or have some sort of mental illness. It's not normal to have a huge profit low risk job and feeling quite down about it.
@@ruffethereal1904 that's a problem of the individual then. I've been working in tech for almost 3 decades now and I live in this awful parallel universe where compensation is shit , expectations are high, and the workload can be overwhelming. Software "engineers" in particular seem to be so mentally dysfunctional that they don't even realise they actually live in paradise. The US corporate structure also seems to amplify this by a lot. I mean, come on, what do think working at a factory production line, a cashier job, data entry clerk, or being butcher in a large-scale slaughterhouse feels like? Sometimes a job is just that - a means to an end, something to bring food on table and pay your bills. It being "fun" or "fulfilling" is just a bonus. If you want meaning or fun - that's what hobbies are for and if someone claims to only work a couple of hours per week, there's plenty of time for fun projects, self-improvement, education, etc.
@@totalermist agree 100% with you. Lol if I was earning 6 figures working a couple hours per day, you bet I'd be learning new crafts, trying out new sports, and other hobbies. IMO if one wants fulfillment one can help others by volunteering for local causes.
What about a "self" driving car that is really just a large front camera where someone in India is "virtually" driving?
oh yes 🤣😂
@qwe😂qwe9678
I've seen how people drive in India.
This should be fun.
Also, we'll finally be able to die due to lag irl.
Ahh, the Mechanical Indian, like the Mechanical Turk.
@@gregrice1354 jesus
"you need to be either highly positive or highly negative" So much of Tech at the moment.
Having worked in the semiconductor industry for 40 years I can say it was very different where I was. Of course I was writing software and designing hardware to test products under deadline and once one project was done there was another to be done.
we have a self driving floor cleaner at work. we have QR codes posted all over where we can drive up to the QR code and scan it in training mode and manually drive the scrubber as it cleans to learn the rout for the next time it scrubs. The downside of driving on it's own is it does not know the difference between a more saturated dirty spot on the floor or a mild spot on the floor. Or the difference between dirt or a rug and can run over the rug and get it caught in the drivers that scrub the floor. I link my phone to the machine so when I run it on auto it gives me a play by play. To put it simply... A machine I have to chase around multiple times to hit the reset button because it went off track or it thought something was in its way and does not know what to do.
You will be charged and found guilty of not utilizing AI probably, since now every idiot out there believes it (whatever AI it is they speak of I do not know) scores so called Einstein level of IQ. But, people who are trying to utilize AI to do something meaningful already experienced AI, already sees AI is no magic wand at all.
We had a similar thing at a warehouse I had worked at. We supplied parts to a single customer that was attached to our warehouse, and management decided to get these automated robots to drive orders back and forth. They ended up having to keep all the people the hots were meant to replace just to cover the robots when they inevitably messed up, and hire extra people to take care of the robots. Their great automation initiative cost them about a million dollars upfront and only managed to increase their overhead. It really is just a bunch of hype so useless middle management types can make themselves seem useful, since without them the workers would just continue to come in and get the job done, and no useless middle manager wants to accept that reality.
Everyone I know that owned one of those put it in the closet and used a broom.
Imagine, a human and broom is cheaper, faster and more accurate. AI..robot expensive garbage.
Give it 10 years. Even then. Why spend 30,000 dollars to sweep?
@@randymulder9105 why spend $30,000 recurringly? If the floor cleaner is good enough, and mark my words they will get good enough, you will save a lot of money paying 30,000 for a floor cleaner with the right application of course
Yet it’s still going to replace some of the jobs at your company.
Thank you! I knew intuitively that the ChatGPT and OpenAI stuff are hypes, but I had too few arguments, just a gut feeling.
You clearly don't know anything if you think that.
@@OnigoroshiZero I certainly do know something about neural nets and language models. (And using AI)
@@OnigoroshiZero can it be a new Mozart or DaVinci without copying those works before ?
I wrote an AI-centric novel where I actually predicted much of what's happening. Of course, no one cared since the story was neither dystopian nor utopian.
I asked
@@ricardogarciarevilla6922 If you don't mind my asking, what did you ask?
That's still great, and a big accomplishment that you wrote an entire novel! good for you!
@@vis4083 Thanx😁
I worked very hard and intensly in all my tech jobs. Often in Saturdays and Sundays. I'm software engineer. I'm exhausted.
the only way to make real money is investing savings overtime or having a successful business. Labor helps for survival, but thats it.
Dr. Emmanuel makes a great observation that neural nets can only solve for instances that are in it's data set of training. The edge cases for self driving cars are such an example. Us humans with much smaller training data sets can solve these kinds of problems in short order thanks to our abilities of abstraction, general world comprehension and reasoning.
WOW, I have been working for the wrong companies. I have been in IT for 30 years and a developer for 20 plus. At all the companies I worked for I put in at least 50 plus avg of hard code developing a week. Many times over 60 hours and many many all nighters on tight deadlines. I guess those companies need a 'real' tech manager or director. I do agree about scrum, it can easily slow the process down unless you have a strong scrum master. I am currently an IT director and ensure my team stays busy in 'meaningful' work. Help get me work at one of these companies and I will set things straight lol.
How can you possibly sling code for even 8 hrs straight a day, let alone 10 hrs and output quality code? In my 25yr career I've never met anyone who is capable of this. None of my reports can come close to that sort of marathon approach without burning out.
@@bloopbleepnothinghere well I did but never stayed at those companies long.
@@bloopbleepnothinghere Come over to Mainframe HLASM, mate, you'll see how it's done..
@bloopbleepnothinghere how many hours a day or per week is average or ideal? And for someone more exceptional? Thanks!
Yeah... I certainly did many 60 hour+ weeks from 1993 to 2018ish, but SINCE then I've seen a massive drop-off. In fact I was recently hired by a small company because of my expertise and work ethic, saying to me "I can't tell you how hard it is to find an actual engineer, let alone a senior one."
This was fantastic, great perspective from a very smart person.
We don't even understand human consciousness yet we are talking about giving it to machines
Human consciousness formed from natural selection without intent. When the first digital agent gains consciousness, it probably won't be deliberate. It will be a "happy" accident. We likely won't need to understand consciousness to create it. Nature didn't even need to be conscious itself to create consciousness in humans.
Consciousness is irrelevant. As soon as AI get a general problem solving ability it's an AGI, and nobody cares if it's "self-aware" or not.
@@vitalyl1327that doesn't mean that when it gets general problem solving it will recursively exponentially get smarter. Also, without consciousness and intention, it cant have its own goals in order to solve them. It eill be blank and just a ML for many fields. So unconscious agi cannot intentionally get progressively smarter by its own even if we wanted it to do so... You are another victim of the fools propaganda about doomsday fictional scenarios... Impudent
We are not ‘giving it to machines’ so much as it’s an emergent capability of neural nets at scale, resulting from the fact that we ‘borrowed’ so much architecturally from nature when we created and have refined neural nets. Also, although “consciousness” is a commonly thrown-around term, I would call it “self-reflective capabilities” and the beginnings of metacognition.
Consciousness does not come from matter
We can only create clever simulacra with these techniques
Man, what universe are these guys living in? I had 20 years of hard death marches and 60-hour weeks. Nothing but working my butt off.
@@relly793 Mostly e-commerce, but thankfully retired now.
i am working in software since 1993, the jobs that require actual work are those they pay the least, but the upper stages of the food pyramid do fit his description. the entire industry is mostly useless and where it is not useless it is harmful
@@sillysad3198I'm sorry....you just described every industry... software, ai etc are only different in the fast hiring/lay-off cycle. Every industry the people at the top work less. Even middle management with endless meetings with no seemingly productive outcomes (I've wasted so much time in those meetings).
Ai agents don't sleep, or.need coffee breaks.
that seems a familiar pattern in a lot of professions @@sillysad3198
@@sillysad3198
I've been in the industry since 92 and I've only heard this narrative recently. The narrative used to be that we all have to leave the field in our 30s due to burnout.
Best Podcast I heard in a while. I admit to having been sucked into the hype and thinking about "post labor economics", when the reality is so much more evident. Looking forward to Dr. Maggiori's next book!
Great interview - thank you very much. Finally someone rational and proficient on this topic.
Such a refreshing talk hearing someone with sensible views
No, its not a rational or proficient interview
Pareidolia hallucination in artificial intelligence can be caused by the complexity of visual data, limited or biased training data, overfitting of AI models, and human biases in development and evaluation. To address these causes, diverse training data, robust AI architectures, and human oversight are important.
Majority of the tasks people are trying to resolve with AI can be done with a where clause.
when
Example?
IF Else can still do so much
MIT students found out that they could beat Alpha Go etc by using "stupid but effective" strategies. They just did things that a "good" go player would never do, since a person would see it immediately.
Basically, the training data was all from good players. So using "impossible" strategies went outside the training.
Interesting conversation. I didn't have time to watch the whole thing but didn't watch part of the section on ChatGPT. I do think you are mistaken on the idea that ChatGPT doesn't have an understanding of the world and that hallucinations can't be understood. The architecture of ChatGPT (particularly transformers and vectors) do create an empirically derived view of the world. It is a pity we don't get the probability distributions generated by each individual prediction, but OpenAI can investigate this. Also, the embeddings do seem to extract a semantic map of human language across that 12K dimensional space (in the case of GPT 3, it's likely much more for GPT4).
I do agree that it is overhyped but the scaling laws are yet to be broken and we might see more emergent capabilities from larger models and will likely see smarter ways to apply them (i.e. multi-agent approaches) that lead to improvements.
That said, the idea that we'll be able to generate literary works of art with a prompt is clearly misguided, as are similarly fanciful notions based purely on AI hype.
I'm in the same position as the invited guest. There was a short period of luxury time, when there was no hostility towards AI, when it was all magic, but there were no user interfaces yet, so only the coders could produce said magic. And now, that all the tech goes mainstream, the barrier of entry goes lower. Almost everything AI related is instantly replicated and diluted to the point of being worthless. You get no time to recoup your investment.
I've done IT since the 80's. Seen the whole tech evolve from 2MHz 8 bit machines all the way to today. It's all been inevitable. To some, it was clear from day one, where the tech was headed. People like Alexander Bard did lectures on these matters back when the audience had no clue what the lectures were about. To some, everything was obvious. You just had to follow the right set of thinkers and futurists. People like Daniel Suarez, who foresaw the influence of social media, before there was any social media. Many visionary fiction books already described the world we are entering. And most of them weren't as optimistic as our corporate CEO's and politicians. Most of such books were dystopian.
This is the best video I've seen for ages. Thanks so much.
Very interesting to find out that AI is not as advanced as the sales pitch
His core point was litterly that "oh look there are billions in the industry and we havent hit AGI/ASI yet, it wont ever ever happen"
Thats his quintecense of it all, like with self driving cars he said. Oh no, the newest study released from waymo recorded that their self driving cars were safer than average human driver by a large margine.
Certain projects take a long time, I mean imagine how long it took to get from punchcard machines to computers. we should have given up at the vacuum tube stage. all the money that has flown into computers and nothing! besides a living room sized calculator!
@@foxt9151, AGI/ASI is a metaphysical impossibility. It is not within the realm of actual science or physics.
@@seriouscat2231As a fellow physicist I would be curious if you could elaborate on this as I do not really see how it AGI is physically impossible or would violate laws of physics.
@@tybaltmercutio, you need to reread what I wrote. Unless you are willfully misunderstanding, in which case never mind.
@@seriouscat2231 No need to feel attacked. I actually was genuinely curious about your take.
But after reading it again carefully - and combining it with your reply - I realize it is just a bunch of non-sense put together to sound smart.
This video is the 100% honest truth as I've been in the technology industry for 14 years and I am currently responsible for software and services solution scope, design and pricing for our entire business unit for Automation, Data & AI.
We have had a lot of business success and expanded from a team of 3 to 30 in about 2.5 years most of our growth has been demand driven not just growth at all cost and I am responsible for profitability.
We mostly sell solutions to handle mundane tedious business information tasks, most of what we sold that drives real value to a business is automation and data engineering with some AI solutions for very, very specific use caes and everything sold has been analysed for return on investment and future strategic goals.
All our solutions solve the most boring tasks that people and businesses strongly dislike.
Fundamentally no one cares about technology, no one cares about how much public image status your company has, everyone wants problems solved to drive value through efficiency at a price point.
Edge caes and having a really good understanding of them is the difference between a successful project/business and total failure.
I needed that first 5 minutes. I've been in the dumps. I studied CS and programmed like a mad man to get good enough to do really fun things. My first job, I really enjoyed it. I was writing tons of code, but it was nothing new. It wasn't exciting or innovative. I thought joining one of the big tech companies would give me that itch. I somehow made it to one of the big tech companies and I feel so empty, oddly enough my ambition is actually draining. I'm making great money, but I honestly didn't do this for the money. I did it to make an impact. I'm actually getting depressed because I don't feel like I'm making an impact. I'm actually taking a role in tech that is considered less "prestigious" then software engineering, but I actually find it to be more satisfying and it requires you to actually work constantly. I mean, where is my head at - that I'm willing to actually take a role that requires me to work when I can literally cash in on stock, continue making well above six figures and coast? I am really on a dry spell right now and I need to build something, but idk how I lost my edge.
You are working for the wrong companies. Do your research and make sure to invest in learning what you find interesting, and then research and find companies where you can do that sort of work that you find interesting. YOU are the solution yo that problem, but you have to take that initiative.
I completely agree. I have a one or two domains that I'm very interested in, but I've been pursuing roles that are outside of those domains and I work on things that are, sadly, uninteresting to me. @@amdenis
You say you're making good money. Why not save a money cushion and launch your own business doing what you find fulfilling?
You and the guest speaker are both working for the wrong companies. You can find jobs where you are a high paid lump if that's really what you want, but I've only ever been in one job where that was even a possibility. Most places that I've worked, if you tried to work 3hrs in a month or more like the guest speaker says he does, you'd be tossed out on your butt, as you should.
DOuble dip and work 2 jobs at once. Just don't tell anyone.
Best discussion I've heard so far.
AI "Art" is overhyped. I still think digital art is safe. AI is limited to it's database and it can't make anything new or original. It might look impressive at first glance but it has a lot problems and doesn't understand the fundamentals of art or color theory. It only understands patterns. Honestly do what you love and keep drawing.
Im not even an artist and I feel like I can always tell AI art because it lacks any sense of composition at all levels.
For example, when a person decides they want to make a picture of a hyper-detailed scifi cyborg woman flying through space, there'd be intentionality in the woman's pose, how it shows off different mechanical details, they might choose to give her an open cybernetic ribcage to add an element of body horror, or instead make the robot body parts look sleek and smooth, like an Apple product. Then the background and other scene elements come together in a cohesive way that takes lighting and perspective into account. Maybe there's a ringed planet, or an asteroid belt, or something causing conflict or intrigue like a spaceship flying after her. Whenever I see AI art try to make something like this, it always seems like it combines the elements at random, because they can technically make sense, but don't cohere into a complete vision. It feels like it was created through cold iteration on forms (because it basically was).
Not to mention, there isn't nearly as much control over these tools as people like to think. You can't really make minute adjustments with the level of precision and control that an actual artist has, and those details are what separate art that's great, from art that's "good enough".
Worst part is, AI dummies don't care. The fact that they can write a prompt, makes them look intelligent in their own eyes and they have started to belittle other humans with actual skills and calling his shit 'art' better because it takes less time to produce (newsflash, it's shit, no matter how much they tweak it, only anime art looks barely decent, but it's AI shit it could look good but it's the same vaseline crap!)... then they start to cope that they can fix it, it's all so tiresome, the technology of mediocre people
Isn't that how humans work. What is original? When I see some drawing of a Sci-Fi alien creature, its usually put together of parts from existing creatures. Everything you know and create is based on a database of experiences and consequences in your life. This is the same for all of us. None of us go deep into minds and create anything new. At least my thoughts, based on the lectures I've listened to, and my experiences.
@@joshualossner2328 you are not an artist and don't understand the difference between "AI" art and actual art done by humans... my god, the fallacy of everything has been done is the most ridiculous one. Then AI art is shit by definition, and most people defend it to death given they lack any artistic merit without it.
@@joshualossner2328 it's "part" of how humans work, but humans do a LOT more.
Humans can take inspiration from different contexts and re-shape them into new ones. Humans can form more global compositional intentions and make far more cohesive art. We are not simply statistical collage machines. Think very hard about how would go about making something creatively. Yes, you may be able to relate every element to some inspirational source, but you did much more than just place those elements together and make them roughly fit. That's more or less all an AI is doing.
Very interesting. Thank you.
It is indeed refreshing to hear a more measured and nuanced point of view. 100% agree on the waste generated by tech teams especially in investment banks.
This dude's book completely blew me away. BRILLIANT book.
Yo glad to see that my personalised profile on youtube picked up this vid, may you reach opulence!
It seems like you mean well and are trying to make sense out of things. However, it amazes me all of the people talking about AI who have never done any real AI research or development. As someone who has done deep learning for a decade, and machine learning, ANFI and expert systems AI dev for over 3 decades, I will say that what is about to happen will amaze many and stupify most, as few see what's coming, which is a double exponential growth rate already over 1,000% per year prior to Quantum-AI training, which will enable a spectrum of AGI and beyond beginning in 12 months. ASI will also come as a spectrum of evolving standards, and in some respects has begun in narrow fashion ahead of AGI. In any case, say goodbye to more than half of all jobs within 7 years and the creation of new medical, technology and social systems beyond what most could have even imagined in 100 years.
AI might have some crazy hype going on right now, like claims that we will see AGI in 2 years. But in the mid to longterm, its a no brainer where we are heading with AI and ALL of big tech is jumping on the AI train. There have been multiple big discoveries in the last 20 years in AI and computantional power per dollar is increasing on an expontionital rate and that is not slowing down at all. We are heading into a very interesting future.
It's far worse than this. Reddit and UA-cam are filled with people who believe AGI is already here because of the stochastic parrot effect, which was described in a paper warning this would become a problem over a year ago. 99% of these people have no computer science knowledge and couldn't even tell you what a context window is, but somehow have convinced themselves that 'chat GPT' as they refer to LLMs is fully capable of human reasoning. They are no words.
It's a computer program, not AI
@@Astro2024A computer program that runs AI. Why brother with naming, it's doing very impressive inteligent work as we speak.
@@Astro2024
A computer program that can explore a problem space and produce solutions better than programers working alone on that same problem without ML.
The AI part is how the data has been modeled as the result tasking machines to learn from that data.
@@Astro2024So what AI is supposed to be if not a computer program?
Damn. I was hard on myself for procrastinating some days after a month of 10+ h/d coding. But I still suffer from bad choices later in development. I found that not forcing myself actually just let me avoid getting too deep into a bad choices consequences. I just don't really see all the problems I will have to handle, and no one actually ready to pay/wait for me to finish the feature/project. It feels like the only way to make software responsible is to consume your personal budget on your self as employee. Everything else may lead you to f up scenario.
When most people say AI, they mean what the machine has actually learned. Or at least how that knowledge is being applied.
That is the intelligent part, the part that the engineers could not do independently of ML given the same data and seeking to produce the same results.
The model of the data that produces a given output relative to a given input was arrived at by the AI, not the engineers, and this is not trivial bc that gap in understanding is what creates alignment problems so it is a misconception to suggest that gap in knowledge is not an issue.
And the reason why AI is smart until it is dumb.
I think Scrum has very little if anything to do with Agile. The two main points of Scrum are to bring the business and developers together in a single team and have a 'sense-check' every so often, usually every 2 weeks and that's the only part of it that's Agile and you don't necessarily need to do a meaningful sense-check with the Product Owner. I have found that Project Managers work around the Scrum and try to make the whole process more waterfall as they need an overall plan and feel this is essentially more predictable. Teams aren't awarded for being efficient or mitigating the various risks inherent in software development, they are rewarded for sticking to a plan.
worked for 2 years, wrote a book, became a superstar.
this corraborates his story perfectly.
He is completely ignorant about the subject, and even his intelligence is questionable.
He is the one riding the AI hype train with negative views that are easier to get people's attention (as everything negative).
@@OnigoroshiZero AI is fake though.
don't get me wrong, i have no questions to the person, he is an example not an agent in this story.
the issue is in the societal perception of the "intelligent" a noob writing a guru-book is the OK.
Read the book Bull shit jobs by the late David Graber, it hits this phenomenon right on the nail
Do ❤ love so much both of you so scintillating, and thanks for accompanying me on a hand-wash (definitely non-AI) laundry overdue for 4 weeks...🎉
I didnt expect much but it actually turned out to be one of the best podcast on AI I heard so far, very informative and without all the fluff and clichés. Thank you.
Thank you for your kind words
Its not informative, its the opposite
As someone in the aviation industry, and what you described in gatewick with the A-SMGCS system. I can give you some insights why that is the case. First of all such systems exist for very very long time. But they are expensive and the certification for aviation safety of such system is very very complex task. The GPS/GNSS used to have too large of an error for ground movement operations and there are way way too many vehicles so you can't really have them all equipped with Squitters because you will just just block the frequency. Parked vehicles will always have squitters off. GPS is also not secure enough and can be easily jammed, thus in order to use for operational purposes ASMGCS system you will also have SMRs(Surface Movement Radars), From concept to operations of a technology in the sector takes more than 20 years. It is not cost or investment that is making it so long but the whole Safety First culture and the extreme regulation.
About some of the points of AI I think your view is way too balanced and you are downplaying some facts. Yes it is basically machine learning but how is human learning diffferent? The chat GPT Kenya RLHF example is not different than a human going to school being tought and shown how to solve tasks,write essays etc. The fact is that LLMs and some other AI models have shown to develop emergent properties very similar to how humans do. Even tho current models are narrow they tend to scale a lot with more compute and even tho Moore's law is dead in the sense of transistors scaling compute is actually increasing, and the fact that Mixture of Experts or multiple interacting agents that are very narrow and specific can work together and show synergy means that even tho we may hit a ceiling that it might be so high that the world can change very very fast.
While Neural Net AIs learn in a way that's... similar... to humans, they lack the logical association that humans use to learn. Knowledge, to a human, is interconnected in a way that today's LLMs could only dream of.
When person A says "I want an apple", there's A LOT of meaning/processing behind that statement. This person has recognised a state of hunger/craving, they can visualise the presence of an apple alleviating it, based on past experience. They understand that the apple is food, they have an understanding of what it takes not only to acquire an apple, but how the apple comes to be in the first place (and most steps in-between). So in addition to intent, a statement like this typically communicates an understanding of what fulfilling that intent will cost as well as why the intent is there in the first place (among many other things).
When person B replies "How about an orange?" it holds a similarly ridiculous amount of meaning underneath.
Both AI and humans decide through likelihood, but the likelihood estimations are happening on completely different levels. When person B replies "How about an orange?", many layers of meaning have been exchanged, whereas, when ChatGPT replies "How about an Orange?", it's because it calculated that those are the most likely words to follow the statement "I want an apple".
So yeah, when a human goes to school, hopefully they're extracting a vast amount of meaning from every lesson. When an AI reads a book, it's (mostly) skipping over the meaning and saying, "Ah, so this word is more likely to occur when preceded by these words". Completely different ballpark of intelligence.
(Attention and embedding are cool, but they're a single step on the thousand mile journey to human level intelligence)
I work in tech for almost 15 years and what this guy says is 100% correct in 90% of cases. The rest 10% is the extreme opposite. In reality there is a lot of work being done on software and even more work that needs to be done. But allocation of resources and skills required to do them is poorly distributed. Most tech companies are less effective in managing engineering projects than USSR was in managing their economy. That's also why USSR lasted much longer than average software project does.
When I worked in retail, it used to piss me off knowing that more educated, or folks in higher-paying jobs with more complicated tasks were often getting paid more to slack off. I knew people who worked for universities and CPS; in both cases the more educated employees were paid more to do less. Of course sometimes that highly paid individual has more responsibility, so to be clear my observations may not match up with everyone else's experience.
My sincere admiration for your deep insights and perspective of things, Emmanuel! Thanks you El Podcast ❤
I've worked as a software-developer for 12 years in 4 different companies. I have NEVER run out of things to do. Not even close. In one company we did not get anything new to do from the outside for half a year, and we were still no where close to running out of meaningful work to do. So much goes in to stuff the customers don't see that still adds value (maybe to the customer and maybe to the company). If nothing else because much of this work makes it easier to maintain the codebase and get new work out to the customer faster and in a safer manner.
Though, I do believe that some juggernaut companies have resources to waste and are willing/ignorant to do so.
That last part I can confirm it is definitively true. Picture wasting weeks if not months of man hours because the tasks they create take months to be created and when they finally arrive, it takes waaaay too little time to complete (two sprints tops to complete everything they took months to bring to us) We can't tackle tech debt because of bureaucraziness and most of the time is spent in self-training (you have to go through more bureaucraziness to get aproval for good training courses and material)
I thought it normal to look for tasks and opportunities in the absence of top level direction? I've always done so, unless was prevented by permissions or obstinate management (exception.) There's always efficiencies that can be made, e.g. streamlining and automation. Writing documentation or making the code cleaner in the absence of other tasks.
@@brytankak9598 I agree with what you said. Blessed the ones who are on teams that allow us to do that. Not all of us are that lucky tho.
Ultimately, we are part of a team and unless the team is onboard too, we may just waste our time with things that won't be used and not noted by leadership at all. The only thing left for us, is the learning experience.
It won't do any good, team wise, to do documentation no one else is going to read or keep up to date, to tackle tech debt if no one is going to approve PRs and let them get outdated with the main branch, to write tests to automate them if the team keeps doing them manually and refuses to learn how to automate tests... Without Top Level direction / leadership approving the work we do, it will remain as things we do for ourselves, unfortunately 😔
I find it helpful to recall these two principles.
1) We ae extremely conditioned by fantasies (most stories for kids, most "education", entertainment, boss speak, tribal "Knowledge", etc.)
2) The Gold Rule.
In short: feelings.
Good talk. Can relate. Thanks
i remember bill balmer saying people would not use phones without buttons.
And I remember people saying planes will replace cars (it was 50 years ago) or that by 2010 we will have a moon colony. P.S. It was Steve and people give him too little credit for what he actually did.
But they do have buttons.
@@arcomarco7131
I remember people making the OP's argument when I said ipads won't replace laptops, especially for software engineers.
😂 thanks a lot... since about more than 10 years ago when AI became a little disturbing, I decided that I can always convert my law firm into a huge SPA/ massage Palace, like what they have in Bali and all these super resorts❤
Yes cult of agile is way overblown 6:45
What are the alternatives though? I've seen a lot of all of it and agile is as good as any.
The problem is dogma. You can't be dogmatic about agile, that defeats the whole purpose. Agility accommodates the nature of tech. Engineering is ambiguous at times, and an engineering team needs to be able to accommodate spikes, injections, outages, change in business demand, etc. Other industries can't do that, manufacturing requires rigid planning because once a die is set, it is expensive to change. Software enjoys the ability to pivot at a moment's notice but to be able to take advantage of that you need a process that embraces that. That is where agile comes in. It's tried and tested, but often abused, and seen as the end, rather than simply a means to an end.
On my team we work in whatever way makes sense for the work we are doing. We change processes whenever we feel like it. We can adapt to a significant roadmap change without too much fuss because there really isn't anyone stopping us. We are asked to deliver, and no one cares how we do it. To me, that is an agile team.
I'm building an AI app that looks for tells and inconsistencies that occur when someone is lying or misrepresenting .. so this interview was very useful in that respect.
Really enjoyed the book. Great interview.
Well it's always "just a matter of time" so I don't think you have an argument there. If this was the year 1890, you'd be saying the same thing about cars replacing the horses. You'd say "Oh yeah suuure, we'll see a Model-T Ford... I'll believe it when I see it!" But it was just a matter of time wasn't it? Sure it was 20 years later. But 20 years is a matter of time. And the way Ai works is that once a certain point is reached (which we haven't reached yet) it is capable of improving itself. So that "matter of time" is smaller than waiting on the Model T Ford. It's 5 or 10 years from GPT 3. Or you could say 20 or 30 years from OpenAi's inception. So while you do make a lot of other good arguments, that isn't one of them.
Finally, an intelligent conversation about AI. Thank you for the podcast, it's really interesting and as a programmer, I find it very sobering.
You guys are excited about Anti AI 🤖 propaganda! All AI scientists and experts are worried about exponential growth of AI and one man in this video enough to tilt the balance 😅😅😅😅😅
Enjoy the ride
AGI coming my dears
He is wrong on several points, but I noted this one, when he says that LLM hallucinations are part of the system.
It's true, but it's also true for us, the more finesse the model gains, the less it will hallucinate until it becomes completely indexable to human intelligence.
I remember the Dilbert character Wally proposing to management that his next task should be testing 'system volume stresses.' And when he got refused he said 'Damn! I almost made it my job to watch [pron] all day."
There are times when it is hard to work. I quit my prior company because for 18 months we were in the middle of an infrastructure reorg that was supposed to take ~3 weeks and 18 months later we were still down. I could write code (without requirements), but there was no path to getting it built, tested, or deployed. Though that was an unusual situation. The prior 12 years had been full time good solid work - real work. The place I'm at now, it took a few months to really find a spot for me so there was a lot of thumb twiddling there as well at first. I'm about a month away from retirement, can't say I'm sorry about that.
Literally, "We need to create a team for AI but we don't know what we want to do with it." is happening at my company right now.
30:27 This is exactly what I do most excellently. I’m one of those hobbiests Dr Maggiori talks about making the great discoveries. I have stuff coming out on biblical literature that’s going to blow the curve. I’ll also have a lot to say about AI. :-)
Anyone who has worked in a distribution center knows how inhumane the process is, unloading containers, putting away, picking orders and stacking pallets to be later shipped, it's time consuming, low paying and very dangerous. So these are areas where robots and machine learning will thrive, meanwhile things will get cheaper because of efficiency and people can focus on more complex asks.
Oh, and Einstein, really bad example. He never considered rotating the Michelson experiment vertically, so the task is incomplete. Ubiquitous, but anisotropic in the presence of a mass body, the Aether shall set us free.
very interesting, thank you man
So what you are saying is that big tech companies are absorbing so many programming engineers just to "have them" around, that they are more like "Patents" used in Patent denial practices? ... They buy up all the patents to prevent someone else to access it? So, same process, hire all the programmers, just to prevent other businesses from having them?
As far as the claim of "without human knowledge" for many AI advances, yes... this is a common theme, tho I personally would not include "rules" for games as in that scope specifically because we humans created the game, and we can't have the AI solve the problem of winning the game if we don't actually tell it "the rules"... so, that seems like a bit of stretch, and clearly that should be an exclusion from the "without human knowledge" issue.
super interesting. I've felt overworked overt the years as an engineer.
I work in IT, - never had a minute of free time at work 🤯 Yeah, but everything is legit what is said here.
This sounds right. Well done management! Maybe IT audits are in order
Wow, that would be my dream job, doing nothing but always have time to do things in the meantime that i love but gives no money, gamedev for example :)
8:46 - The answer is obvious, really.. because you're stuck either online and/or in an office doing a job already.. You can TRY to remotely make more money with the rest of your working hours, if it's a desk job at least. But you'd get fired if caught. The fundamental issue therefore lives in the fact that we're working only an hour per day and satisfying our superiors enough to not be terminated, and yet we're still expected to sit there and ROT for the remainder of the time, just so everything looks good on paper; waste of life.
Yesterday ChatGpt gave me two answers posted right next to each other, in two columns, and asked me which one I prefer. Still a long way to go, yes.
Idleness in the states, while in China they are required to work 9-9 6 days and are all too exhausted to actually work.
This is by far the most accurate analysis of the current AI situation I have seen.
Chat bots also get context wrong a lot, especially if you write a lot and ask it to edit your writing.
None of the AI disaster people have explained how a Super AI would wipe out all people and then somehow live on. It needs people to run all the chipfabs that make the cpus that runs the servers it needs. Without people, all the power plants would break down and run out of power. So do they think AI would just be vindictive and wipe out people just because it can even if it means sealing its own doom?
A lot of people saying that are uneducated. They think Terminator is non-fiction.
Once you have robots to replace humans for the physical jobs, they can run the chip fab plants too. They are already replacing humans in warehouses, and remember that car manufacturers eg. Tesla don't need much human intervention to build mass produced cars. Companies will automate anything if it becomes cheap enough to replace human labour, that's how capitalism works in practice, so we humans will be the ones create the automated factories ... there's the irony.
Have you read Hubert Dreyfus who wrote on AI, in the 1960s?
Right now, all I see is Siri/Alexa with real tangible benefits and features (extracting valuable info from your e-mails, condensing wall of texts, putting timestamps with references on videos, making lists and howtos with real ease and correct grammar). I see programming tools for teaching and aiding programmers. Besides that, great hype and sure as hell no "Intelligence".
Jeeze im in tech for a mid sized luxury business, i work my rear off and never have idle time. Crazy that exists out there.
very interesting! thanks dude, greetings from Spain
I was expecting a solid argument, got nothing new
This is Gold!
I'm just wondering how to get these jobs where I do nothing and get paid a lot
We all haha
Nepotism
59:45
This is a HUGE problem in the gaming industry. That, combined with people learning improper techniques from UA-cam University like destroying game objects instead of caching and reusing them.
That's why all your favorite videogames lag out and crash constantly these days. People are just using copy pasta code.
so basically, people who hype up AI think like this:
AI = complex
real world = simple
AI = supernatural magic, we are all doomed, oh no, oh no
Every AI image generator should be required to record every output on a blockchain, so that image can later be traced back to that AI.
AI engineer for over 20 years here.. He looks like that guy from the "transformers" ? 😂
45:41 it’s going to be more about curation not creation
That requires taste and culture
I largely, largely agree that AI hype is overblown, and it's a bubble.
What I disagree with is the point of view that whatever it is that is there today, will benefit society.
These are decently powerful tools just waiting to be abused by companies, and streamlining production, cutting unecessary people, saving on production costs
these are not things that benefit society. These are things that benefit people that have like, workers, y'know?
This interview is great
What does AI have in common with a blow up doll? a) looks, b) nothing, c) everything, d) all of these. Find the correct answer in Cgbt.
I'm an independent translator. I was fired from my gig at a tech company because they have already sunk so many millions into AI which hasn't translated into any profitable product particularly, but they can at least use this AI thingie to cut internal costs. As a translator, luckily I am already near the top of the value chain , but that work is quite hard to snag and doesn't pay particularly well. Meanwhile in my days as a corporate slavey my hero was Wally in the Dilbert cartoon, the archetypal work avoider who had every excuse to not work, and he was such an entrenched apparatchik that he could fend off all challenges. Many of his strategies work surprisingly well in the real workplace.
Unfortunately, current AI can translate pretty good if promoted correct.
@@cristianandrei5462 It's "good enough" for most users, particularly if the text is something routine.
I did Support Vector Machines (SVMs) in grad school. I laugh my butt of when people think NNs and Machine Learning will be sentient. It's nothing more than a really complex spell check.
It's like most advancements. A fraction of a fraction of a percent of people will stumble across a brilliant idea using the new technology and make absolute bank. The rest of us will continue to spin our wheels and fail to see what the big deal is.
ChatGPT was completely lost while I wanted to find a specific electronic circuit design for RF application (that i cou;dn't find on the internet BUT is very well described in some vintage books about Radio ciecuits etc. from 80')... it kept repeating the same information over and over again with some minor changes into it. It can't do a specific job which only a human understand. But it can remove some stupid and repetive roles like copy and paste milins of time the same thing. Anyway I had to read the whole book and designed it myself which works very well. I am not naive and I think I understand the limits of this kind of technology.
For a start, Scrum != Agile. Scrum is a practice, Agile is a methodology.
Secondly not every Tech worker is working just microseconds a day. Some of us actually work.
Thirdly AI is a super set, ML is a subset of AI. While as a pragmatist, I do agree that there is a lot of hype in the industry and in general to AI. I also believe that there is a limitation to current systems, even as impressive as they currently are, there are still major gaps.
There is no concrete path currently which says LLMs are the future of AI, though their NLP capabilities are a major advancement, being able to communicate with a computer with natural language is very important. Chat GPT and others are not the AI we will end up with, it is a step along to the path of AI. The human mind is a predictive machine like LLMs, but that is not the only thing it is.
Sadly you are wrong..we teachers are alread discussing emergency strategies because we are already becoming mostly obsolete in Schools besides supervising that the students truly learn with their AI step by step App instead.of.going to social media
AGI will certainly be able to solve all computationally decidable problems. However, non computationally decidable problems like another Mozart, Evariste Galois, DaVinci will not be solved by AI in our lifetime… it doesn’t count to study Mozart to become one. Authentic Coolness doesn’t study, it just creates. The AI would need to be able to create tear dropping phenomenal classical music WITHOUT training !
Not at all. Teachers are being replaced by good teachers. Have you heard of UA-cam? You are like the movie rental store being replaced by Red Box . Get on UA-cam and run your skills by the people already learning there. See how you do against the competition. Oh and no union support/emergency strategy suppression of the talent already in place. What I'm saying is that you need to look a little ahead because this is the direction learning is going in.
@@deker0954 Im of the belief that a lot of "bad teaching" you see is more of a disrupted classroom management by weak and gentle admin policies. Teachers are tied to what admin and district says, so if they say you need to tolerate the disruptive student, you must tolerate the disruptive student that is making the class quality go down.
Im also of the belief that a lot of "school/teacher is bad" hate is because of an effect of spatial association, a lot of people immediately associate "school as a place" as a bad thing and therefore will act unwilling in such a space, affecting the quality of their learning and reinforcing the belief. Whereas UA-cam is associated with positive things and fun and dopamine bursts. It's far more complicated and boiling it down to "youtube being better" is naive of the current state of education.
This doesn't rule out an actual bad teacher case existing, but you'll be surprised how minuscule that factor is affecting the outcome, it's probably less than 10% by just observation.
Hard to care to individual needs over UA-cam. Guess an AGI or individual mentor/tutor might be better here than someone creating content for a huge diverse audience of million potential listeners
interesting interview. what will ai bubble burst look like. im finding some very stupid stufff being promoted by hi level vc like plaiday launching text to video. the video is 3 seconds of silence. like a half az meme maker. a 3 second ai picture is not a video. like calling a motion meme a video.
I get the impression most are interested in how much money they can make off of AI. It tends to make me think there's a lot of undue hype. And probably a lot of misrepresentation. I have no doubt people will make money off of it. But they use it the same way scammers tack on "hyper," "Quantum," or any number of other words onto something that add nothing but mystique.