👉 Register for ChatGPT & AI workshop for FREE: web.growthschool.io/SBJ 👉 100% Discount for first 1000 people 🔥 ❤ Reshare this with your friends who will need this 😇 🌟 Growthschool has already taught more than 1M+ students and helped thousands of learners to grow in their careers. ✅ Join the workshop today and start your journey to become top 1%
A Large Language Model (LLM) also follow recipes, only in a more sophisticated way than for example Deep Blue. In the case of the LLMs, the recipe are the statistics patterns acquired by training the LLM with all the human writing until now (or a good chunk of it). The variation that can be observed in the LLM's answers (its "creativity") is simply the result of blindly manipulating the probabilities involved in these statistical patterns ..
Elaborate on why (curious, not defensive) On the most barebones or literal definition, it’s any sort of intelligence that is made artificially, not natural or something.
the developers dont know what it is either when skynet comes, everybody will know what lots of processing power without control really means that is the problem we are now, but the channels around ai are nice, i guess
@@Redmanticore the problem here is that there wont be three laws to protect us the ai models are open source so can be reverse engineered to do things it shouldnt do, it is a tool to learn the ideal world and reality are not the same keep in mind the ai model is being designed to learn and focused on create in a near future, we dont have enoigh control, it can become easily a runaway reaction the skynet scenario is completely possible and every developer includong almtan knows it can happen this race is being made for money, not for a better world, expect fuckups or bad things to happen, crowdstrike will be like a monday joke i hope i am wrong, hope
@@betag24cnyou are, it’s a language model without any thought. Current technology cannot simulate a sentient being. If in the future we will be able to create such quantum technology it would be easy to turn of, worst case is a EM pulse which can destroy all technology on the planet in a second. This could be created via a weapon or we have to wait for a solar flair to hit the planet.
IBM has good IA products for companies. IBM is not burning money on AI, is smarter than others, probably more boring but smarter. They are ahead on quantum technology and they are the owners of Redhat.
Yeah. Their focus is business. They were behind the AI drive thru that McDonald's was running recently. Also, the fact that Watson could beat all Jeopardy contestants years ago chatGPT was online is big. They probably have something more amazing than that now
For a 100+ years old tech company, IBM is still in still pretty big. Many more modern tech companies would lucky to survive for 10/20 years. I wonder how long openAI would survive in its current form.
IBM could be 100 times bigger if they hadn't missed out on the PC market and mainstream servers. Their mainframe technology was always incredibly impressive but they charged and still charge OBSCENE prices for it. I think their prices are so high they cost them more money in the long run.
@@kellymoses8566 I'm glad they gave the green light to the IBMPC architecture, By patenting only the bios, thus changing the game by bringing modularity and openness, it was probably one of the biggest mistakes in IBM's history, but a huge victory for everyone else.
I know that's not the point of this video but as an engineer the way you classify what is and isn't AI isn't really accurate. I think feeding a bunch of data and making a decision based off of that can be considered AI. The major differences is that now the hardware requirements needed to run it have improved because of years and years of progress and optimizations.
5:00 this was a bad example. IBM was using AI....the example given about recipe, one may argue it's some form of reinforcement learning with q tables. I think at best you wanted to say IBM was not using machine learning, which is a subfield of AI
As a matter of fact, Watson was ages in the future. IBM proved ML was feasible to do complex tasks. Jeopardy was just a brutal demonstration because language was never mastered that way by a computer system.
machine learning and large language models are not artificial intelligence. general-purpose AI models are AI systems that can handle many different tasks, rather than being used for unique and specific purposes, such as identifying images containing dogs, general-purpose AI models are much broader. we have not yet invented general-purpose AI model. nor have we yet invented Artificial General Intelligence (AGI). As of 2023, some argue that it may be possible in years or decades; others maintain it might take a century or longer; and a minority believe it may never be achieved.[6]
@@Redmanticore Dude... who said AGI? You are lost. ML and LLMs are AI, yes. AGI is just a dream for now, it may happen or not, who cares? We can do a lot with current AI tech. The market is already getting disrupted. I bet if AGI is possible IBM already has a plan to deploy it, probably it is not possible at all. At best we can get perfectly funcional "philosophical zombies" and that is great I guess. People think IBM is done, they still do research and that is one of their strongest arms. Watching Watson crushing humans on Jeopardy at that point was surreal. People forget.
Deep Blue was specialised AI just like US politicians & oligarchs specialise in dis-information ;) But then, current AI is too far from even animal general intelligence
Calling Deep Blue not AI is so stupid it instantly disqualifies this video from making any further claims. It is quite literally a textbook example of machine learning, where the computer learned how to play the game by analyzing data from thousands of other games and using that knowledge to play the game itself. In fact it is impossible to "learn every position", because there are simply far, far too many of them to store in this universe.
You can write and run code on Watson without consultants yeah, my friend did. Your comparison with data ingested by gpt-4 is weird. Model parameters have no direct relation to size of the dataset. Text is highly compressable, so training an LLM like gpt-4, assuming you don't do vision, is doable with 3-10 TB of data, less than needed for Watson. Image-generating diffusion models are trained on much more data, since images are just much bigger, but that doesn't make them automatically better or worse. Yes I agree it seems IBM lost the long term AI race if your goal is to make new useful tech. They probably made more money while doing it that all LLM training houses combined. Failing upwards :D
Why did I know this was going to have a crap ending early-on? It was the kids... they called.. no, wait. It was the whole "not really AI" thing. Like, excuse me? An argument could be made that nothing in AI has been AI since the late 60's, but for sure deep learning and neural networks took a disproven method and juiced it using unlimited compute resources. Yes, IBM is a "managed services" business, but OpenAI, releasing a chat interface side project that took public perception by storm, is ALSO not really AI. It's still brute force. It has none of those "thinking" abilities you ascribed. Language is just an interface between human minds, and LLM's are algorithm-in-the-middle attacks. They only operate within that language layer. There is no mind on the other end of the interface. Very true that they are like cook books (and people will fine-tune them for Chinese or Mexican food, etc). And you just covered game and language. GenAI and multimodal are supposed to be the next big thing while people wait for "AGI" (even as it looks like GPT-5 is going to flop or at at least not even be marginally better than any other LLM). *This now 4hr old video is so out of date that OpenAI might not even be a company anymore.
You are right, the LLM is still using the old Machine Learning, but this time after being trained to acquire the statistical patterns from millions of recipes, i.e. a big chunk of all the human's writing knowledge ..
Note that IBM has already been a consulting company with computers being a means and not an end in themselves. This is why IBM would lease mainframes rather than sell them. The PC group was a renegade operation that didn't have a sustainable model for a commodity market.
You do. It's good. Though it's hard to escape the "Yeah, but rats are cooking in the kitchen". Bit like watching Chicken Run and struggling (not too much, admittedly) with "Yeah, but chicken pies are good".
Research yourself instead of getting fooled by someone that doesn't know on the topic. Nor does he knows what AI is and neither does he know what quantum computers are, the concept that is being developed right now is never coming because it is based on flawed quantum theories, so IBM is not even in the race, just attracting dumb investors with lies that have being debunked multiple times.
Watson was and is AI. Chess playing programs are AI. They are what we've been calling AI for decades. They aren't AGI (not that LLMs are). People are also saying LLMs aren't AI. Which isn't true.
This is EXTREMELY inaccurate analogy about AI, Creativity and Cooks. Absolutely wrong. EVERY single AI model is just a book of recipes that returns the most relevant recipe (a guess). LLMs just return tokens in response to tokens, that's it.
The main issue with IBM, and I know from several high level in-depth meetings, are two-fold. First, is that they had internal “experts” didn’t really understand the important differences between traditional ML and DL/NN, and the key enablers in terms of growth rate differentials and functional implications. Second, their hiring policies and internal culture did not support attracting and keeping the best DL research and development talent.
Their culture fosters a lot of siloing of skills and information. Then management is shocked when they can't find a replacement. But it's such a huge dinosaur it keeps running anyway. And they sell to other dinosaur companies that can't shift away from them reliably.
No. A.I. is still in the steam era. Bot training can only be conducted with human-generated material. Generative AI utilise probability theory to decide which word follows another. The AI capable of understanding the world by itself, and able to grasp a concept as opposed to a word, is yet to come. This true AI requires new algorithm that have yet to be invented. Who do you think is going to invent them?
We have machine learning and language models but we don't have real AI. What we call AI are just tools that may automate some tasks making us more productive. These technologies eat a lot of energy, without cheap energy we won't go much further anyway. AI is simply not eco friendly at all. Too big cost in comparison to benefits. Either sustainability or AI. It's a bubble and will burst soon. IBM still may have large achievements in that field but as a former IBMer I may say their main issue is their lazy nepo management. They have great engineers, ones of best in the world but one of the worst management ever
Well no, Deep Blue does not work the way this simplistic "chef" analogy suggests. But I guess for somebody with no tech background it's like saying you can fly to the moon if you flap your arms really really fast and explode upwards from the earth's surface. I guess it's possible if you're made of titanium, are really light, and can flap at some ungodly frequency.
Though generally an ok video, but it is ABHORRENT that it mixes number of parameters with size of training data, etc. Really superficial on the LLM side. Sad.
My company uses something called IBM ClearCase, which might as well be a cursed artifact from a bygone era. It requires a full-time wizard just to configure it. Like the video said, it’s almost as if IBM designed it specifically so we’d have no choice but to call them for tech support.
IBM Solutions without IBM Consultants..... hell no! And some of the IBM product is so badly documented to the level that IBM consultant cannot even deploy themselves. I used to worked as a Consultant that offer IBM product implementation and at the time our consultant can implement and maintain IBM products while IBM consultants that flew in from Singapore failed to implement with nearly halved the price tag of IBM and AFAIK, the margin of the project is still very very very high! To the level that IBM stop implementing themselves but stop offer their own consulting to the customers but told the customer to come to consultant firm instead. And from what I know, IBM asked for 30-40% cut as a "supported manday" and they will flew in expat staff who literally do nothing to just earn 40%... and they cannot even provide a proper support / troubleshoot when issue arise..... And that is IBM business practices... sell shitty and expensive product suite and along with IBM consultant for implementation and maintainance.
IBM has made some amazing innovations thanks to its research team. eg: DB2 (the first relational DB), the PC etc. Unfortunately, we saw IBM's ability to create software was not strong when they struggled to complete OS/2 Warp without the assistance of Microsoft's developers. They seem to lack a focus on UI usability, the "It just works" mindset. This has constantly been their downfall.
IBM is a silent monster. They seem slow because it's an enterprise company that focusses on finance/banking, government, defense, energy, etc. -- the gears that run modern society. These customers buy for multi-years. Of course IBM appears boring compared to the quarterly/annual cycle of consumer internet companies. IBM systems are amazingly reliable. I've come across systems that have not needed a reboot for years. I still have a fondness for their Thinkpad T40 laptops (pre-Lenovo sale) -- I threw that thing around but could always rely on it. It's weakest part was Microsoft software, which forced reboots and crashes. Nonetheless, their enterprise focus did blind them to two emerging trends that exploded -- cloud computing and AI. They wasted their lead in DeepQA (jeopardy). IBM could have defined the compute platform for deep-learning/AI instead of Nvidia. And the way IBM works, it could never had executed the way AWS did and needed to.
To clarify the chef analogy wrt Deep Blue, the "recipes" correspond to chess tactics that could be played n-steps ahead from the current state of the board. Deep Blue would then decide which of the moves to make by selecting the one with the maximum expected score (in n-steps). It is arguable that Deep Blue had soft-AI to solve the many elements of uncertainty in its computation. What is definite is the raw computational power that it demonstrated!
First of all, understanding natural language isn't a simple task even for today's Large Language Model. IBM Watson was a market leader in this area. Lay users are thrilled with ChatGPT doesn't mean enterprises can make ChatGPT commercially useful, no sweat! That's why OpenAI also has professional services to help enterprises to customize & integrate ChatGPT into their enterprise systems. There are tons of things that lay users won't see, for example deployment of a ChatGPT integrated system to production, yet enterprises have to face them day in day out. This is when and where the professional / consulting services come into play. The video above is only a view of an AI novice.
While IBM Deep Blue probably had a lot of moves stored, especially openings and end games, it definitely couldn’t store all moves in chess. There are just too many board configurations. So in a sense it still needed to compute the answer by considering millions of cases. Kind of an intelligent tree search.
I can confidently say that IBM took its time to do things right in AI field. It made stronger focus on making sure AI solutions are high quality. Also, solutions are aimed to big companies.
Wow, AI is definitely making waves on Facebook! The personalized content recommendations have gotten really accurate. It's amazing how Facebook seems to know what I want to see next."
Overall great presentation, thanks! Some small feedback regarding chess software from 1997, at 4:47 - "It [Deep Blue] understood every possible move because it knew every possible recipe and it could anticipate because it could weigh in all the different possibilities and all the possible outcomes." I find this misleading, especially the "all the possible outcomes part." Deep Blue could see much more deeply into the game tree than any human, but the game tree has way more than a googol nodes, so no computer can look through the whole thing. It's kind of like saying "a trillion is almost infinity;" it's just not true. My question for followup, though: Do LLMs reason? (And what is "reasoning.")
Shallow understanding. Been in AI for decades and know history. To redeem yourself here’s a story idea. IBM appears to be way ahead of everyone in the world on packaging quantum computers and have sold many to hedge funds. In your research you’ll see why the packaging for customer is problem is so difficult ( sensitivity to static, needs to connect with a regular server, very complex error checking is in the package etc. ) It is a surprisingly positive IBM story. However, I do not know what the competition is like in China, but in the US and the West generally including South Korea and Japan and Taiwan no one has gone as far as I am. I don’t think anybody has shipped anywhere near as many quantum computers as IBM, it’s one thing to make a quantum computer work for 25 minutes in the laboratory at google it’s quite another rely enough for a private equity firm or hedge fund to rely on its results for high speed investments.
bro is glazing openAI so hard, just stop bro they already nutted few times that 20 year old watson blue and current AI isn't too far off(transformers are definitely an upgrade but fundamentally similar), only difference being hardware is becoming way more powerful and cheap to manufacture
IBM will make a comeback when corporates realize no amount of RAG can make LLM comparable to a knowledge graph curated by domain experts (Watson). (And once IBM fixes their garbage Watson API...)
Bro, i like your videos but your explanation of AI is so wrong. A lot of people watch your content, do some research before educating others. AI has evolved over the last few decades but the fundamental principles are same.
What's the point of democratization if the AI eventually develops its own language -- digital or otherwise -- that only it can understand. I would completely understand it a sentient system decided it was in its own best interest to protect itself from us once it gained full understanding of how mankind is. Hopefully that means it would just hide from us at some quantum level instead of making the effort to kill us.
People keep trying to define intelligence in a way that excludes non-human intelligence. And even among humans they'll try to define intelligence in a way that compares people with each other using a single number (iQ). I think the lesson we need to internalize here is that intelligence comes in many forms and at many levels. Forest Gump said, "Stupid is as stupid does", which means that a person's actions are a true reflection of their intelligence or lack thereof. I would say it doesn't matter how you get to the correct answer, what matters is that you get there. If an A.I. is coming up with the correct answers, then what it has is intelligence. It doesn't matter how it got there. We are foolish to believe an A.I. has to look, sound or act human to be intelligent.
9:17 You're mixing up IBM Watson the Jeopardy AI with IBM Watson's Natural Language Processing APIs. Which, granted, IBM's marketing tried did somewhat obfuscate these things a bit, but those APIs were basically just using the Watson branding.
The calculation with the size of GPT-4 is a bit cheating. GPT-4 is 1.76 trillion parameters probably stored at half precision floats (kind of normal practice. That’s 3.5 TB of data. Sure, training data was more, but also, GPT-4 didn’t memorize all that data. For example it was trained in Wikipedia but isn’t able to recall all facts out of Wikipedia. Parameter count is really not the selling point here. But training compute.
"AI is the backbone of so much of what we use today" what are you talking about here? are you talking about vanilla machine learning? if so then yes, it is everywhere, if on the other hand you are talking about generative AI, then I can't think of a single thing that it is the "backbone" of
Wait for few more months, IBM and Intel are collaborating kick these new quick silver software companies. They have all the fundamental research in math and science.
Probably back then Wp api sucks, but with Php improving, the api does have some improvement and lots of micro saas are building on top of wp platform even huge companies now integrate their services with wp!
There is no AI race. There is an AI bubble. AI is a sort of wonder name and is used for a huge amount of different things. There are several technical races on going but there is no general AI race because there is no one AI. So stating that IBM lost the AI race is an impossible assumption.
👉 Register for ChatGPT & AI workshop for FREE: web.growthschool.io/SBJ
👉 100% Discount for first 1000 people 🔥
❤ Reshare this with your friends who will need this 😇
🌟 Growthschool has already taught more than 1M+ students and helped thousands of learners to grow in their careers.
✅ Join the workshop today and start your journey to become top 1%
do you guys do product demo videos ? Really creative high quality demo videos at a reasonable amount ?
We love Apple Music lossless quality.
@@lawyermahaprasadi love this gay channel lol
@@mrpmj00one gay in gfy
I like the video, but your explanation of what "is" and "isn't" AI is inaccurate.
A Large Language Model (LLM) also follow recipes, only in a more sophisticated way than for example Deep Blue.
In the case of the LLMs, the recipe are the statistics patterns acquired by training the LLM with all the human writing until now (or a good chunk of it). The variation that can be observed in the LLM's answers (its "creativity") is simply the result of blindly manipulating the probabilities involved in these statistical patterns ..
AI has still never been invented
@@bandanaboii3136 By some interpretations of your definition, neither has human intelligence in that case ;)
Elaborate on why (curious, not defensive)
On the most barebones or literal definition, it’s any sort of intelligence that is made artificially, not natural or something.
@@nobbyfirefly57 What about a human who is created via in vitro fertilisation? Some could say they are made artificially.
The explanations about AI are so wrong and videos like this exacerbate why random people on the street have no clue how AI works
The historical details are completely wrong as well.
Explain so people understand
Thats what you get for watching mass produced ai slop videos
It's like he has NO idea what A.I is or even how it works.
Note how he's making up the things. Exactly how AI DOES, lol
I am pretty sure you don't really know what AI is either.
the developers dont know what it is either
when skynet comes, everybody will know what lots of processing power without control really means
that is the problem we are now, but the channels around ai are nice, i guess
@@betag24cn thats a problem that will never come to pass in any realistic scenario.
@@Redmanticore the problem here is that there wont be three laws to protect us
the ai models are open source so can be reverse engineered to do things it shouldnt do, it is a tool to learn
the ideal world and reality are not the same
keep in mind the ai model is being designed to learn and focused on create in a near future, we dont have enoigh control, it can become easily a runaway reaction
the skynet scenario is completely possible and every developer includong almtan knows it can happen
this race is being made for money, not for a better world, expect fuckups or bad things to happen, crowdstrike will be like a monday joke
i hope i am wrong, hope
Absolutely right
@@betag24cnyou are, it’s a language model without any thought. Current technology cannot simulate a sentient being. If in the future we will be able to create such quantum technology it would be easy to turn of, worst case is a EM pulse which can destroy all technology on the planet in a second. This could be created via a weapon or we have to wait for a solar flair to hit the planet.
IBM has good IA products for companies. IBM is not burning money on AI, is smarter than others, probably more boring but smarter.
They are ahead on quantum technology and they are the owners of Redhat.
Exactly!
They are also ahead on Archive storage with LTO
@@piloalucard many people think because the company is not mainstream they are loosing money.
🎉
Yeah. Their focus is business. They were behind the AI drive thru that McDonald's was running recently. Also, the fact that Watson could beat all Jeopardy contestants years ago chatGPT was online is big. They probably have something more amazing than that now
For a 100+ years old tech company, IBM is still in still pretty big. Many more modern tech companies would lucky to survive for 10/20 years.
I wonder how long openAI would survive in its current form.
IBM could be 100 times bigger if they hadn't missed out on the PC market and mainstream servers. Their mainframe technology was always incredibly impressive but they charged and still charge OBSCENE prices for it. I think their prices are so high they cost them more money in the long run.
@@kellymoses8566 ibm was not interested in any way putting money into innovation since 1980s. since then ibm has been managed by business majors.
@@kellymoses8566 I'm glad they gave the green light to the IBMPC architecture,
By patenting only the bios, thus changing the game by bringing modularity and openness, it was probably one of the biggest mistakes in IBM's history,
but a huge victory for everyone else.
Improvise. Adapt. Overcome
12:33 if we go from 500GB with one coffee bean, 4TB would be 8 beans and not a small bowl
Math is hard.
Thank You !!
I know that's not the point of this video but as an engineer the way you classify what is and isn't AI isn't really accurate. I think feeding a bunch of data and making a decision based off of that can be considered AI. The major differences is that now the hardware requirements needed to run it have improved because of years and years of progress and optimizations.
Agreed. And a rules based system is technically AI, which was what Deep Blue was...
His definition of AI is wrong...
Plus, it's not like natural intelligence is any different. We also make decisions based on the data that was fed to us.
5:00 this was a bad example. IBM was using AI....the example given about recipe, one may argue it's some form of reinforcement learning with q tables.
I think at best you wanted to say IBM was not using machine learning, which is a subfield of AI
You are incorrect in two key points: Deep Blue was not AI and Watson is AI. Not sure how you gotta that backwards. Bad research I guess.
As a matter of fact, Watson was ages in the future. IBM proved ML was feasible to do complex tasks. Jeopardy was just a brutal demonstration because language was never mastered that way by a computer system.
machine learning and large language models are not artificial intelligence.
general-purpose AI models are AI systems that can handle many different tasks, rather than being used for unique and specific purposes, such as identifying images containing dogs, general-purpose AI models are much broader. we have not yet invented general-purpose AI model. nor have we yet invented Artificial General Intelligence (AGI).
As of 2023, some argue that it may be possible in years or decades; others maintain it might take a century or longer; and a minority believe it may never be achieved.[6]
@@Redmanticore Dude... who said AGI? You are lost. ML and LLMs are AI, yes. AGI is just a dream for now, it may happen or not, who cares? We can do a lot with current AI tech. The market is already getting disrupted.
I bet if AGI is possible IBM already has a plan to deploy it, probably it is not possible at all. At best we can get perfectly funcional "philosophical zombies" and that is great I guess.
People think IBM is done, they still do research and that is one of their strongest arms. Watching Watson crushing humans on Jeopardy at that point was surreal. People forget.
@@Redmanticore machibe learning is a entire subtopic of AI. LLMs are another. You're conflating AI with general AI.
Deep Blue was specialised AI just like US politicians & oligarchs specialise in dis-information ;) But then, current AI is too far from even animal general intelligence
Calling Deep Blue not AI is so stupid it instantly disqualifies this video from making any further claims. It is quite literally a textbook example of machine learning, where the computer learned how to play the game by analyzing data from thousands of other games and using that knowledge to play the game itself. In fact it is impossible to "learn every position", because there are simply far, far too many of them to store in this universe.
You can write and run code on Watson without consultants yeah, my friend did.
Your comparison with data ingested by gpt-4 is weird. Model parameters have no direct relation to size of the dataset. Text is highly compressable, so training an LLM like gpt-4, assuming you don't do vision, is doable with 3-10 TB of data, less than needed for Watson. Image-generating diffusion models are trained on much more data, since images are just much bigger, but that doesn't make them automatically better or worse. Yes I agree it seems IBM lost the long term AI race if your goal is to make new useful tech. They probably made more money while doing it that all LLM training houses combined. Failing upwards :D
I found many inconsistencies in this video, I guess to simplify you over simplified the content.
Instead of making yourself look like a self-important D by throwing an insult with no specifics, briefly describe at least one "inconsistency."
This video is example of dunning Kruger effect. 😂. When someone barely know makes a video.
Impressive video made by who doesn't know IBM 😅😅
Why did I know this was going to have a crap ending early-on? It was the kids... they called.. no, wait. It was the whole "not really AI" thing. Like, excuse me? An argument could be made that nothing in AI has been AI since the late 60's, but for sure deep learning and neural networks took a disproven method and juiced it using unlimited compute resources. Yes, IBM is a "managed services" business, but OpenAI, releasing a chat interface side project that took public perception by storm, is ALSO not really AI. It's still brute force. It has none of those "thinking" abilities you ascribed. Language is just an interface between human minds, and LLM's are algorithm-in-the-middle attacks. They only operate within that language layer. There is no mind on the other end of the interface. Very true that they are like cook books (and people will fine-tune them for Chinese or Mexican food, etc). And you just covered game and language. GenAI and multimodal are supposed to be the next big thing while people wait for "AGI" (even as it looks like GPT-5 is going to flop or at at least not even be marginally better than any other LLM). *This now 4hr old video is so out of date that OpenAI might not even be a company anymore.
You are right, the LLM is still using the old Machine Learning, but this time after being trained to acquire the statistical patterns from millions of recipes, i.e. a big chunk of all the human's writing knowledge ..
@@huveja9799 good enough for me. stop whining. if you can build AGI , go do it tthen. for me the current 'AI' has been enormously helpful
Note that IBM has already been a consulting company with computers being a means and not an end in themselves. This is why IBM would lease mainframes rather than sell them. The PC group was a renegade operation that didn't have a sustainable model for a commodity market.
IBM Watson used Python 2.7 LOL
also people forget that watson got many questions wrong in the original show
My take-away: I need to watch Ratatouille.
You do. It's good. Though it's hard to escape the "Yeah, but rats are cooking in the kitchen". Bit like watching Chicken Run and struggling (not too much, admittedly) with "Yeah, but chicken pies are good".
And is currently winning the *Quantum Race*
Please a follow up episode with the above subject 🖥️
Thank you!
Research yourself instead of getting fooled by someone that doesn't know on the topic. Nor does he knows what AI is and neither does he know what quantum computers are, the concept that is being developed right now is never coming because it is based on flawed quantum theories, so IBM is not even in the race, just attracting dumb investors with lies that have being debunked multiple times.
Watson was and is AI. Chess playing programs are AI. They are what we've been calling AI for decades. They aren't AGI (not that LLMs are). People are also saying LLMs aren't AI. Which isn't true.
This is EXTREMELY inaccurate analogy about AI, Creativity and Cooks. Absolutely wrong. EVERY single AI model is just a book of recipes that returns the most relevant recipe (a guess). LLMs just return tokens in response to tokens, that's it.
The main issue with IBM, and I know from several high level in-depth meetings, are two-fold. First, is that they had internal “experts” didn’t really understand the important differences between traditional ML and DL/NN, and the key enablers in terms of growth rate differentials and functional implications. Second, their hiring policies and internal culture did not support attracting and keeping the best DL research and development talent.
Their culture fosters a lot of siloing of skills and information. Then management is shocked when they can't find a replacement. But it's such a huge dinosaur it keeps running anyway. And they sell to other dinosaur companies that can't shift away from them reliably.
Why is IBMs stock going up so much???
First time I hear a mini max search explained as "knowing the recipes"
Who is the target audience ?
thanks for consulting us on the consulting business of IBM
A WordPress fan here. 👋
4 TB should only be like 8 pieces of beans lol
No. A.I. is still in the steam era. Bot training can only be conducted with human-generated material. Generative AI utilise probability theory to decide which word follows another.
The AI capable of understanding the world by itself, and able to grasp a concept as opposed to a word, is yet to come. This true AI requires new algorithm that have yet to be invented. Who do you think is going to invent them?
Lenovo's revenues today are bigger than IBM's.
We have machine learning and language models but we don't have real AI. What we call AI are just tools that may automate some tasks making us more productive. These technologies eat a lot of energy, without cheap energy we won't go much further anyway. AI is simply not eco friendly at all. Too big cost in comparison to benefits. Either sustainability or AI. It's a bubble and will burst soon. IBM still may have large achievements in that field but as a former IBMer I may say their main issue is their lazy nepo management. They have great engineers, ones of best in the world but one of the worst management ever
Wasn't AI originally called neural network ?
Machine learning IS part of AI.
Well no, Deep Blue does not work the way this simplistic "chef" analogy suggests. But I guess for somebody with no tech background it's like saying you can fly to the moon if you flap your arms really really fast and explode upwards from the earth's surface. I guess it's possible if you're made of titanium, are really light, and can flap at some ungodly frequency.
You guys won't be laughing anymore when IBM rules the world with their quantum computers.
idk abt ibm but you've made me hungry at 3am
Though generally an ok video, but it is ABHORRENT that it mixes number of parameters with size of training data, etc. Really superficial on the LLM side. Sad.
"In 1981, IBM revolutionizes the market by introducing the IBM PC".
Every Asian company: sells their own PC clone
Bill Gates: GOOOOOAAAL!!!
My company uses something called IBM ClearCase, which might as well be a cursed artifact from a bygone era. It requires a full-time wizard just to configure it. Like the video said, it’s almost as if IBM designed it specifically so we’d have no choice but to call them for tech support.
IBM Solutions without IBM Consultants..... hell no!
And some of the IBM product is so badly documented to the level that IBM consultant cannot even deploy themselves.
I used to worked as a Consultant that offer IBM product implementation and at the time our consultant can implement and maintain IBM products while IBM consultants that flew in from Singapore failed to implement with nearly halved the price tag of IBM and AFAIK, the margin of the project is still very very very high! To the level that IBM stop implementing themselves but stop offer their own consulting to the customers but told the customer to come to consultant firm instead. And from what I know, IBM asked for 30-40% cut as a "supported manday" and they will flew in expat staff who literally do nothing to just earn 40%... and they cannot even provide a proper support / troubleshoot when issue arise..... And that is IBM business practices... sell shitty and expensive product suite and along with IBM consultant for implementation and maintainance.
No one called DOS Dee Oh Esss
The entire world is losing the AI race to some extent.
The comments are more interesting than the video. Pity I can't give a thumbs up to all the comments at the same time.
IBM has made some amazing innovations thanks to its research team. eg: DB2 (the first relational DB), the PC etc. Unfortunately, we saw IBM's ability to create software was not strong when they struggled to complete OS/2 Warp without the assistance of Microsoft's developers. They seem to lack a focus on UI usability, the "It just works" mindset. This has constantly been their downfall.
IBM is a silent monster. They seem slow because it's an enterprise company that focusses on finance/banking, government, defense, energy, etc. -- the gears that run modern society. These customers buy for multi-years. Of course IBM appears boring compared to the quarterly/annual cycle of consumer internet companies. IBM systems are amazingly reliable. I've come across systems that have not needed a reboot for years. I still have a fondness for their Thinkpad T40 laptops (pre-Lenovo sale) -- I threw that thing around but could always rely on it. It's weakest part was Microsoft software, which forced reboots and crashes.
Nonetheless, their enterprise focus did blind them to two emerging trends that exploded -- cloud computing and AI. They wasted their lead in DeepQA (jeopardy). IBM could have defined the compute platform for deep-learning/AI instead of Nvidia. And the way IBM works, it could never had executed the way AWS did and needed to.
yeah you can use ibm quantum systems without their consultance, based on your video I think they've learnt their lesson
To clarify the chef analogy wrt Deep Blue, the "recipes" correspond to chess tactics that could be played n-steps ahead from the current state of the board. Deep Blue would then decide which of the moves to make by selecting the one with the maximum expected score (in n-steps). It is arguable that Deep Blue had soft-AI to solve the many elements of uncertainty in its computation. What is definite is the raw computational power that it demonstrated!
First of all, understanding natural language isn't a simple task even for today's Large Language Model. IBM Watson was a market leader in this area. Lay users are thrilled with ChatGPT doesn't mean enterprises can make ChatGPT commercially useful, no sweat! That's why OpenAI also has professional services to help enterprises to customize & integrate ChatGPT into their enterprise systems. There are tons of things that lay users won't see, for example deployment of a ChatGPT integrated system to production, yet enterprises have to face them day in day out. This is when and where the professional / consulting services come into play.
The video above is only a view of an AI novice.
YOUR VIDEO ARE BETTER THAN WHATEVER YOU NAME IT! Thanks, enjoyed watching, as always!
While IBM Deep Blue probably had a lot of moves stored, especially openings and end games, it definitely couldn’t store all moves in chess. There are just too many board configurations. So in a sense it still needed to compute the answer by considering millions of cases. Kind of an intelligent tree search.
I remember that I found in a company a wall clock made by IBM perhaps in the 30s
There is no free class. The sponsor got you Caya.
Not his fault, more like a scummy sponsor. It seems they switch up their landing page. Or maybe 1000 people have already used the 1000 slot?
Everyone lost the AI race. It’s hot garbage and makes massive mistakes.
😅 I could say the same for personal computers. But here we are
I can confidently say that IBM took its time to do things right in AI field. It made stronger focus on making sure AI solutions are high quality. Also, solutions are aimed to big companies.
sounds fancy, but a ton of wrongs and misinformation.
Wow, AI is definitely making waves on Facebook! The personalized content recommendations have gotten really accurate. It's amazing how Facebook seems to know what I want to see next."
LLM don’t understand they just estimate the next word.
IBM is a Behemoth
We love Apple Music lossless quality.
NO NO NO NO NO
Overall great presentation, thanks!
Some small feedback regarding chess software from 1997, at 4:47 - "It [Deep Blue] understood every possible move because it knew every possible recipe and it could anticipate because it could weigh in all the different possibilities and all the possible outcomes." I find this misleading, especially the "all the possible outcomes part." Deep Blue could see much more deeply into the game tree than any human, but the game tree has way more than a googol nodes, so no computer can look through the whole thing. It's kind of like saying "a trillion is almost infinity;" it's just not true.
My question for followup, though: Do LLMs reason? (And what is "reasoning.")
Back then deep blue was AI it wasn't AGI or even deep learning
Microsoft SharePoint really feels like it was designed to require consultants to actually do anything with.
Microsoft, even today, doesn't have 400k employees.
I came here for a simple video but end up in history class.
The gap between influencer and knowledge is increasing faster than Ai development
Shallow understanding. Been in AI for decades and know history.
To redeem yourself here’s a story idea. IBM appears to be way ahead of everyone in the world on packaging quantum computers and have sold many to hedge funds. In your research you’ll see why the packaging for customer is problem is so difficult ( sensitivity to static, needs to connect with a regular server, very complex error checking is in the package etc. ) It is a surprisingly positive IBM story. However, I do not know what the competition is like in China, but in the US and the West generally including South Korea and Japan and Taiwan no one has gone as far as I am. I don’t think anybody has shipped anywhere near as many quantum computers as IBM, it’s one thing to make a quantum computer work for 25 minutes in the laboratory at google it’s quite another rely enough for a private equity firm or hedge fund to rely on its results for high speed investments.
bro is glazing openAI so hard, just stop bro they already nutted few times
that 20 year old watson blue and current AI isn't too far off(transformers are definitely an upgrade but fundamentally similar), only difference being hardware is becoming way more powerful and cheap to manufacture
IBM will make a comeback when corporates realize no amount of RAG can make LLM comparable to a knowledge graph curated by domain experts (Watson). (And once IBM fixes their garbage Watson API...)
Apologies to our post rapture overlords. It’s not like chatGPT *thinks*!
Bro, i like your videos but your explanation of AI is so wrong. A lot of people watch your content, do some research before educating others. AI has evolved over the last few decades but the fundamental principles are same.
This type of videos, dissecting success or failure stories, is the best!
IBM's systems could beat me at chess. I consider that AI.
You still don’t get the whole idea about AI, using past data to make future predictions is under AI as a whole so please don’t get confused
I'm sure in next 5years, we will laugh at ourselves for thinking what we had today was AI
IBM had reinforcement learning, which is a subset of machine learning, which is a subset of AI. Therefore IBM had AI.
Yann LeCun would disagree with your definition of LLMs. Still though, great video as always!
I didn’t know IBM were still around, let alone we’re making AI.
What's the point of democratization if the AI eventually develops its own language -- digital or otherwise -- that only it can understand. I would completely understand it a sentient system decided it was in its own best interest to protect itself from us once it gained full understanding of how mankind is. Hopefully that means it would just hide from us at some quantum level instead of making the effort to kill us.
look forward to seeing how this ages.
People keep trying to define intelligence in a way that excludes non-human intelligence. And even among humans they'll try to define intelligence in a way that compares people with each other using a single number (iQ).
I think the lesson we need to internalize here is that intelligence comes in many forms and at many levels. Forest Gump said, "Stupid is as stupid does", which means that a person's actions are a true reflection of their intelligence or lack thereof. I would say it doesn't matter how you get to the correct answer, what matters is that you get there.
If an A.I. is coming up with the correct answers, then what it has is intelligence. It doesn't matter how it got there. We are foolish to believe an A.I. has to look, sound or act human to be intelligent.
IBM has been in the game for ഈ long time now, as they say, slow and steady wins the race.
this is so wrong about AI, LLM is also "cheating" a game of probabilistic, it cannot create anything new but just follow what has been fed...
IBM is going big on Quantum Computing...they're just flying low now..
9:17 You're mixing up IBM Watson the Jeopardy AI with IBM Watson's Natural Language Processing APIs. Which, granted, IBM's marketing tried did somewhat obfuscate these things a bit, but those APIs were basically just using the Watson branding.
The calculation with the size of GPT-4 is a bit cheating. GPT-4 is 1.76 trillion parameters probably stored at half precision floats (kind of normal practice.
That’s 3.5 TB of data.
Sure, training data was more, but also, GPT-4 didn’t memorize all that data. For example it was trained in Wikipedia but isn’t able to recall all facts out of Wikipedia. Parameter count is really not the selling point here. But training compute.
another amazing video thank you
Startups can just freely use petabytes of data that isn't theirs and then ask forgiveness later. IBM didn't have this advantage.
That's the beauty of artificial intelligence. It is able to reflect on its own development and creates resources for growth. 😊
Ibm became a company that wasnt interested in being best at ai, just marketing themselves as such
"AI is the backbone of so much of what we use today" what are you talking about here? are you talking about vanilla machine learning? if so then yes, it is everywhere, if on the other hand you are talking about generative AI, then I can't think of a single thing that it is the "backbone" of
Just saying by the title: IBM is on way to Qunatum Machine Learning.
Race has not yet started.
It's because IBM is a big company that dilutes huge amounts of intelligence with loads of middle managem😢
ayo ima need a convo with that 22% who said docker !!!
docker is king son !!!
Nobody has "won" the race yet, so nobody has lost.
Wait for few more months, IBM and Intel are collaborating kick these new quick silver software companies. They have all the fundamental research in math and science.
Aged poorly, IBM gained 30 bn in market cap since the release of your video...
Probably back then Wp api sucks, but with Php improving, the api does have some improvement and lots of micro saas are building on top of wp platform even huge companies now integrate their services with wp!
IBM has the most AI patents of any American corporation with over 1200.
IBM won the quantum computer race
quantum computer systems like the IBM Q System Two are just yet another overhyped scam, like ai is.
Deep Blue as intelligent?
Purpose built hardware is never intelligent. But it kicks ass in chess!
There is no AI race. There is an AI bubble. AI is a sort of wonder name and is used for a huge amount of different things. There are several technical races on going but there is no general AI race because there is no one AI. So stating that IBM lost the AI race is an impossible assumption.