AMAZON JUST WALK OUT:
Ever since the launch of ChatGPT in late 2022, the AI scene has exploded. We’ve seen dozens of startups cash in on the gold rush and big tech has been slapping AI onto everything they do. One of these recent innovations was just walk-out technology from Amazon. The idea was simple. You can walk into any Amazon grocery store you want, pick up the items you need, and just walk out. AI was supposed to do all the work of keeping track of which items you picked up, how much your total bill should be, and what account should be charged. Admittedly, this is some pretty cool tech and it makes shopping way more convenient, except the tech didn’t really work.
While there was some level of AI and machine learning at play, there were also more than 1,000 workers in India watching you shop and labeling videos to make sure that this technology worked correctly. Now, some level of human verification especially in the early stages makes perfect sense, but with the case of just walk out, the tech was much more human than AI. In fact, Amazon’s goal was for just 5% of purchases to be validated by humans. But in reality, a whopping 70% were being validated by humans. In fact, things were so bad that Amazon has decided to roll back the technology and implement something much simpler called Dash Carts. These are basically just smart shopping carts that keep track of what you put inside your cart. Aka, they’re just more advanced self-checkout machines that can be moved around the store as opposed to revolutionary AI that could just magically analyze video clips of people shopping and take care of the rest.
This is by no means a dig on Amazon. The truth is that developing this technology is extraordinarily hard. It’s the same thing with Tesla's self-driving technology as well. While current Tesla's self-driving is no doubt a wonder in its own right, it’s also not the self-driving robotaxis that was supposed to be out 4 years ago. So, is AI actually advancing at an unprecedented pace or is AI still just more hype than reality? Well, let’s find out.
THE CASE AGAINST AI:
If you're interested in deeper dives, interviews with insiders, and exclusive tech analysis, consider subscribing to our free weekly newsletter. But anyway, likely the biggest case you can make against AI progression is all of the severe missteps and miscalculations we’ve seen from even the biggest players. We already touched on Amazon and Tesla, but they’re just 2 examples. Many other big players have had similar challenges in bringing revolutionary tech to market like Google. Google has been investing over $10 billion into AI every quarter and they have longer-term plans of investing over $100 billion. Not to mention, they’ve been working on AI since 2010 or even earlier. Yet, they’re still not even able to bring an advanced chatbot to market. The first time they tried, their chatbot made a factual error on their own promo.
Google Bard falsely claimed that the James Webb Space Telescope took the very first pictures of a planet outside of our own solar system. And this fluke cost the company $100 billion in market cap. After this misstep, Google would take a step back, rework their AI, and launch a new chatbot called Gemini in late 2023. But, Gemini was arguably even worse than Bard. It wasn’t even able to answer simple questions about people of significance. According to Gemini, this is George Washington, these are popes, and these are British Royalty. Yeah, Gemini would cost the company another $96 billion in market cap. Again, this is not a dig on Google. Bard and Gemini are extremely impressive in their own right, but at the same time, they’re not even accurate enough to reliably replace Google search yet.
It’s the same thing with Siri. Apple has been working on Siri since 2010 and they have some of the highest R&D expenditure on the planet. Yet, Siri is quite far behind even the most basic generative AI models.
Apparently, this is supposed to change with the next version of iOS, but it hasn’t happened yet. And the point is that even the biggest companies in the world with unlimited R&D budgets and massive focuses on AI are struggling to make their visions a reality. And that’s just the big tech side of things. If we shift over to the startup side of AI, things are way messier. Investors have been in a frenzy to invest in anything that has the buzzwords of AI and machine learning. Over just the past 3 years, investors have poured $330 billion into 26,000 AI and machine learning startups. Can you believe that? Like, what in the world are these 26,000 AI startups even doing? Well, 9 times out of 10, they’ve got a regular boring business like compliance or banking but what they’ll add is an AI twist. We’re revolutionizing compliance using AI or we’re revolutionizing banking using AI. In reality, all they’re doing is plugging the business into ChatGPT APIs and calling it an AI-powered business. Usually, such integrations offer very little additional value but they’re a great selling point for investors to raise funds at eye-watering valuations.
In terms of actual AI innovation though, these guys aren’t really doing much. As you can see, the big guys are spending hundreds of billions and getting lackluster results while the little guys are just using AI as a gold rush to raise as much capital as possible. So, where is the real AI innovation actually happening?
THE CASE FOR AI:
While some projects like Bard and Gemini haven’t quite lived up to their hype or pedigree, there are sectors of AI that are already way better than humans ever will be. For example, highly complex cognitive tasks like playing chess. AI was able to beat the world’s greatest chess player way back in 1997, and chess engines have only gotten more and more advanced since then. It’s the same thing with any task that has defined constructs but a mind-boggling number of possibilities like Rubik’s cubes. Using 35 CPU years of idle computer time donated by Google, researchers were able to prove that any one of the 43 quintillion Rubik’s cube possibilities could be solved within just 20 moves. But even for champion cubers, it would be impossible to find each and every one of these solutions. Aside from high cognition and strategy tasks, AI has also excelled in mimicking natural human language. Now, when it comes to accuracy and reliability, that’s a whole different issue.
But, when it comes to carrying out regular conversations, AI has become nearly indistinguishable from humans over chat. In fact, many would argue that ChatGPT completely broke the Turing test and that we now need new ways to assess AI advancement. AI models have also made substantial progress when it comes to creative outlets like art and speech. Now, technically, AI doesn’t have any innate creativity, but it is able to draw upon the creativity of the world’s best artists to create something that appears to be new.
When it comes to anything that doesn't have clearly defined constructs or comes with a lot of variability like driving, AI has a rather tough time keeping up with humans. There’s actually a name for this observation: Moravec’s paradox. The paradox basically describes how it’s super easy for computers to do things that are extremely hard for humans mass data analysis and computations. Conversely, it’s super hard for computers to do things that are natural for humans like sensing depth, fine motor skills, and innate feedback control systems. And this paradox seems to be the biggest obstacle when it comes to most consumer AI applications. But, not everything needs to serve consumers directly. One of the biggest applications of AI is in the world of research and development.
Take chip design for example. AI has proven to be quite powerful when it comes to analyzing billions of variations in chip design to maximize performance and minimize power. In fact, Jensen Huang was recently talking about how Nvidia is no longer focused on designing chips. Rather, they’re focused on building the AI that builds chips. And chips are just one industry that have benefitted from AI-based R&D. Another industry that is being propelled by AI research is medicine and more specifically nanotechnology. In fact, many scientists suggest that super-long lifespans of 200 and beyond are much closer than we think thanks to AI. As you can see, AI is progressing faster than you think but not in the sci-fi sense of self-driving cars and Amazon’s just walk-out technology. Rather, most of the relevant AI progress is actually behind the scenes improving efficiencies and accelerating the pace of innovation. That’s where AI is really progressing faster than you think. When it comes to sci-fi applications though, those are likely further out than you think.
THE TRUE WINNERS OF AI:
Now, I don’t want this article to be overly negative because we have definitely made strides in regards to AI. But, with that being said, it seems that the vast majority of AI achievements are behind the scenes with little sci-fi applications. For AI to really live up to its promised scie fi potential, we still need a lot of time and basically infinite funding. For example, even OpenAI is only able to drive $2 billion worth of annualized revenue, yet they’re boasting nearly a $100 billion valuation. And Sam Altman is looking to raise as much as $7 trillion to make OpenAI’s vision a reality. Imagine what sort of valuation the company would need to raise $7 trillion. We’re probably talking about a valuation of tens of trillions if not even $100 trillion. If anyone can make that happen, it’s probably OpenAI, but that gives you a glimpse of just how much work, capital, and time is left to make this sort of AI a reality.
This isn’t to say that there are no AI winners though. The winners aren’t the big tech companies who are pouring hundreds of billions into AI models or the hype startups who cashing in on the gold rush or even the AI startups that are working day and night to push forward the industry. The real winners are the companies selling the shovels starting with of course Nvidia. Within just the past 12 months, Nvidia’s annual revenue has rocketed from $4.8 billion to $30 billion. That’s an increase of $25 billion. At OpenAI’s current revenue rate, it would take them over 12 years to pull in that much revenue much less net profit after all expenses and taxes. Nvidia is just one winner though. Another winner is Microsoft. By now, I’m sure you’ve all heard about their massive deal to invest $10 billion into OpenAI, but when you read the fine print, you’ll see that this isn’t actually an investment, it’s a loan. Microsoft is entitled to 75% of OpenAI’s profits until they make all their money back after which they get a 49% stake in the company. I should also mention that OpenAI has only received a fraction of the promised $10 billion so far. So, it’s not even an upfront loan but more of a “We’ll fund you as you need it sort of thing”.
If you ask me, this sounds just like a Mr. Wonderful royalty deal. Also, something else to note is that with Microsoft being such a big investor, OpenAI is somewhat forced to use Azure data centers and supercomputers as opposed to AWS or Google Cloud which brings us into the last set of winners of the current AI boom:
Infrastructure providers:
Nvidia GPUs aren’t useful on their own. You have to assemble tens of thousands of these GPUs into supercomputers to make them useful. And can you guess who these supercomputer providers are? It’s of course Google, Microsoft, and Amazon. So, the real money in AI is not in just walk out or Bard or Gemini or even ChatGPT. The real money in AI is in selling the dream to investors and startups. Startups make money by selling the dream to investors. And big tech makes money selling the dream to startups as they profit big by selling them GPUs and infrastructure. And that is the truth about AI. As usual, the money is in selling the shovels, not finding the gold itself.
If you love reading such high quality business case studies then consider subscribing to Arktrek it's free. You can scroll down to find subscription button. We promise to not fill your inbox with spam.
Comments