Way back in 1999 I was a consultant with Broadvision. We were one of the main tool providers for building online shops. I personally worked on sites for Mothercare, Dixons and Barclays. Everyone realised that online shops would radically alter the commercial landscape. So everyone piled cash into firms like Broadvision.
And then it all collapsed, almost overnight. The reason was simple. The underlying concept of online trading was solid. But the market quickly realised that through the introduction of standardised and commidified tools, site building didn’t have to be so expensive. So specialised and propriatory firms such as Broadvision were sidelined.
The same will happen with the AI software boom. China will flood the market with cheap and easy to use tools (and you wont have to give them your data) Models will get more standardised and cheaper to host and improve. The vast majority of applications of ‘AI’ will just be copies of others (just like all e-shops looked the same) with minor fringe differences.
It is ‘just another wave’. It’s a good one, I’m enjoying it. But it will hit the sandy beach of scaled commidification soon enough. Exactly like the e-commerce boom, which imho gave us Cloud computing, the trick is to figure out what will be left after the crash. Data ‘storage’ is a safe bet. Personal avatars would be my outside bet.
Here’s some history. That peak share price was an unbelievable 18,500usd
Interesting premise. If AI is useful for X application, a low cost AI trained in similar environment and application should be good enough. Optimize what you wanted to optimize > profits.
Computers and software allow to standardize processes. Why are we falling for the narrative of a “secret sauce”?
Uff! If you create it so that one can “Rule 36” ones personal avatar you will not know where to put the monies.
Imagine being told to go and fornicate yourself and you go: “Ok!”
Exactly the same with designing and building websites. I did this with my own small enterprise, including e-commerce, from 1999 until around 2015.
Then the likes of Wix, Weebly and a dozen others had replaced the designer/coder/builder working from home. Anyone who could use Word, could build and maintain a website and I retired!
In short the Internet is still young and developing. I expect it will be unrecognisable in another 20 years - a long way from the wild west we knew just 30 years ago with dial-up and free music streaming…
I agree.
Some time ago I posted I didn’t understand the business case for AI or more simply what products and services will be offered and at what prices.
Today there are multiple AIs available free of charge. If they start charging then there’s always someone who will offer free services.
I think the software is so far quite commoditized, but this relies on the goodwill of Chinese AI shops continuing to release models for free. The major western AI shops seem to have second thoughts on giving away the output of billions of investments for free.
So in the end, I think we will have some free Chinese models together with the best proprietary models, which are substantially better than the free models. If it continues this way, we could imagine sufficient competition that the software side is commoditized.
The question is whether and how quickly competition will come to hardware as currently Nvidia is running away with the bulk of the hardware market.
While I agree that things look precarious, some have been predicting a crash for years and with governments spending like crazy, who knows for how long the party could keep going on!
The biggest obstacle is the context window. There’s no intelligence in today’s AI, because it’s like a mind of a 4 year old, who remembers only the last 2 or 3 sentences, so you can never relay on saying thing once and calling it “done”. The context window won’t scale anytime soon, maybe never due to energy constraint. It’s like with CPU speeds, in the 90" people believed that it’ll always double every year…
On a side note, I bet we’re doomed to live with this dumb, crappy AI. I can see that it’s an inevitable move in software development. It’s really bad, however the PMs don’t care about “perfect” solution, are always happy to have something mediocre shipped fast. I’ve tried to get used to this, tried many solutions, you know what, I quit I am not forced do use this annoying AI nonsense at the moment, but I bet it’s over, the next job will require proficient use of AI or you’re not qualified. Using AI to help with code development is like trying to do any work with an enemy sabotaging your every step. Years ago, some smart-asses conceived that they can put 2 developers to one keyboard and they’ll be more efficient than both doing some work on their own. It was a nightmare, but hopefully the world soon forgot about this stupid idea. The AI agents are much worse than that, they have instant access to your code, you’re thinking, typing but whatever was a second ago already had changed, you don’t know where you are you have to take the hammer (ESC, Ctrl+Z) to get back to a sensible state, but wait where it was, what was I thinking a second ago, where I am aaaaarrrh !
How does one reconcile those? It seems for an average person today, a clunky AI is fine. Websites are already complaining about a lack of click-through traffic because Google’s AI summary is good enough to get the quick hit of information they want.
Clunky AI is not fine. I repeatedly have to rephrase google searches because the AI is “interpreting” my question.
AI summaries are often banal or sometimes total gobbledygook, internally inconsistent and using phrases that the AI obviously does not “understand”. AI is only going to increase the number and severity of “sub-postmaster” type problems.
There are going to be big advantages but the hype seems to be very dangerous.
Absolutely, now I feel happier hearing that’s it’s not only me. AI summaries are garbage. They are only good to glace over garbage, that is emails or documents which I have no interest reading and shouldn’t be receiving them in the first place. When summarizing an actual content I need to digest it’s only good to tell you “this recipe is about making some food or about fixing a car”. Any details, conclusions, are 50/50, either ok either totally wrong. Indeed people are using it like a new media, but yea… and no surprise, have you seen Apple intelligence, a phone which throws it’s own interpretation of everything on you?
You have to use the right models. I had AI summarize a several hundred page document that was written in several languages to give me the overall message and determine the main points written by different parties in the document and group the main positions advanced by each group of interests.
Job was done in seconds that would have taken hours to do.
This is the dilemma. A good spreadsheet saves hours of calculations. But even numbers are problematical if one does not have systems to double-check the output. With words the problems are magnified.
AI is going to be very useful if applied well and we will have to get used to using it at work and play. It is a tool.
At least Pinterest has recognised that AI is not its friend. Apparently my ranting had some effect (just joking–apparently there was widespread anger from users), because Pinterest has now made it possible to exclude almost all AI-generated content by fixing Settings. Bingo!
It’s going to be very useful if they finally solve one thing, no hallucination. The knowledge base of an AI model is very lossy compressed data, so the “engine” is picking some similar fragments but doesn’t really know what it’s picking up. AI would be useful if it could have access to the original data to validate source of the statistical pick it made before spitting out the rubbish answer.
Even when formulating answers about a user provided data, it doesn’t reason about it as a whole, but is taking chunks of it (think of paragraphs or pages of text as example) to reason about totally independently, so no surprise you’ve got wrong analysis, conclusions out of context
I am not forced do use this annoying AI nonsense at the moment, but I bet it’s over, the next job will require proficient use of AI or you’re not qualified. Using AI to help with code development is like trying to do any work with an enemy sabotaging your every step. Years ago, some smart-asses conceived that they can put 2 developers to one keyboard and they’ll be more efficient than both doing some work on their own. It was a nightmare, but hopefully the world soon forgot about this stupid idea. The AI agents are much worse than that, they have instant access to your code, you’re thinking, typing but whatever was a second ago already had changed, you don’t know where you are you have to take the hammer (ESC, Ctrl+Z) to get back to a sensible state, but wait where it was, what was I thinking a second ago, where I am aaaaarrrh !