
Seen this before
Remember a few years ago when all sorts of companies were investigating blockchain, but there were very few production applications using it? I recall reading articles like this one, differing only by s/blockchain/generative AI/g
Despite a huge surge in awareness over the last year, just 10 percent of organizations have adopted generative AI technology in production environments, according to a survey by Intel. Overall, 45 percent have made some steps in adopting generative AI, either through pushing solutions to prod, developing the models but not …
Pump and dump.
Create a start up.
Get investors all hyped up (usually old people with generational wealth and no clue about technology, but strong with gambling addiction)
Make sure hiring goes through your mates agency to get some kickbacks
Get people to build carp
Get designers to wrap these turds in a nice shiny wrappers
Get more investors on board
Get a massive bonus
Quit
Rinse and repeat
Nearly every new version of old IT products nowadays presumably contains "AI" capabilities in some form.
SAP, Excel, Fortinet, Cisco, etc.
However, if you look closer at the so-called AI capabilities, they usually do resemble closely what was advertised in the last years as simply automation of some kind, "integrated analytics", "big data", "data fabric" or any of the "smart"-somethings of the late 2010s.
The seemlingly unstoppable trend of AI-in-everything does often seem to result from a marketing department in full overdrive mode, while credibility of claims is built by some interesting and often impressive simulation of text-comprehension by publicly available LLMs - which are still just applied statistics on steroids...
Contrary to what the companies above (and many others) claim - I still need to see an "AI" implementation in older products that does not just result in a more or less useful text generator, that manages to summarize things it read "on the internet" sometimes correctly.
Well said. Especially "a marketing department in full overdrive mode" - which well described Sam Altman himself, and all his TED-like waffles.
I still need to see an "AI" implementation ... that manages to summarize things it read "on the internet" sometimes correctly
What I hope to see are summaries embedded with real, not hallucinated, references. I did download and am trying an app "Perplexity" which is focused on doing just that, which seems to index up to date news so that it can be referenced by AI queries. It's actually a layer built on top of a multiplicity of other AI models. The results are hope-inspiring - the summaries do indeed include lots of real references. The question is whether such a model can be economically self sustaining. It doesn't have ads, telling the truth is not as profitable as telling lies, and paid subscriptions (with the exception of Amazon) have never really taken off.
As I read in this rather good article just a couple of hours ago
Q. What’s the difference between statistics, machine learning, and AI?
A. The size of your marketing budget.
When something is being pushed and hyped as much as GenAI is, your being sold a pup. There is no way GenAI can ever live up to the hype. Even if it solved global warming, delivered world peace while giving you a fucking blowjob.
The reason only 10% of orgs have dipped their toes in GenAI is because orgs don’t know what to do with it. Then again neither do the tech companies and venders otherwise they would have a GenAI product and not have had to resort to sprinkling AI across their products.
But it has been a success in the 10% that have dipped their toes in GenAI. ” More than half (51%?) have improved their customer experiences, upped efficiency, and enhanced product capabilities, the report claimed”. That’s probably because the people surveyed are probably the very people who made the decision to adopt GenAI and are probably not going to admit they have been sold a pup.
Interestingly ”47 percent believed they'd saved money in the process”, so 53% didn’t save money. Also, the 47% only believed, they don’t know, they may not have saved money either. So, 53% - 100% have not saved money in the process.
BTW I recommend reading the linked McKinsey article it's a masterclass in the art of saying a lot while at the same time saying nothing.
Generative AI can be described as Content In Garbage Out.
It's where the value is lost somewhere in the activation functions.
It is also a great way to keep over hired staff busy.
That said, Generative AI can be compared to a shiny object. Business sees the shiny object and is in awe. Then the whole magic disappears and shiny object goes on the shelf.
I know of a few companies that swallowed the KoolAid on GenAI and have thrown lots of resource at it to 'improve the end-user experience'.
This has resulted in them jettisoning other departments to free up cash to pay for GenAI.
What they have all failed to realise is that the data feeding this monster is what makes GenAI good - or bad.
Guess which department specialists they let go to finance GenAI.
Now want to guess how useful the GenAI answers are?
> What they have all failed to realise is that the data feeding this monster is what makes GenAI good - or bad.
True. I think that is one of the more common misunderstandings in the current hype.
If your business problem can not be described in the form of of detecting weak statistical dependencies (or lack thereof) in very large data sets, then the current GenAI approach will probably not lead to a useful solutions for your business. Which is true for I guess 99% of companies...
It's utterly useless!
I was volunteered to take part in a Copilot, er, pilot (employer's pun, not mine). So far, I've been less than unimpressed. I cannot think of one single decent usage case for it! The examples provided are for people who write a lot of fluffy, low-to-no-content emails, Word documents and presentations, or are too busy to read their emails, or don't know how to do what I'd consider basic tasks in Excel (simple formulas and conditional formatting). The best use I've found is as a glorified "find" feature in large documents... which is slower than hitting Ctrl-F and typing in what I want to find.
As for the folks who are worried about "AI" (really ML) taking their jobs - they should be. If you can be replaced with a computer, you're not adding much value to the company.
(I'll go back to programming industrial control systems and reviewing highly technical documents for accuracy now.)