And here is another thank you. It frustrates me when people say quantum computers will make their video games respond very fast.
I worked in testing. Testing quantum computers is a scary notion. Glad I’m retired and won’t have to deal with it.
Speaking of hype, I am editing a book in Word, and the latest version seems to have their AI grammar checker. It is two steps back - maybe ten steps - from the old one. Example: a character is wearing a sleeveless blouse, and the narrator comments on her bare arms. The suggestion was to change it to bear arms. I think the AI learned from a bad comedian or something. Other suggestions are not much better.
OF COURSE AI is over hyped. It seems like a useful tool, but for all the thought pieces and marketing materials, I have yet to see how AI makes my job easier.
And in many cases, it makes my job harder as my dumb-ass Gen Z analysts use it for stuff like creating reports or responding to RFP questions and I need to spend twice as much time checking it for hallucinations or responses that are so generic as to be meaningless.
I think AI has the potential to be useful, but I don’t see generating Powerpoint decks or emails as particularly transformative.
To me the real value of AI is stuff that happens behind the scenes that no one sees or really wants to do anyway. Like reconciling millions of records of data across multiple systems or crunching more data than humans ever could to glean insights around cost savings or new markets to pursue or inventing drugs or whatever (with verifiable rationales).
AI as it currently exists is not a cost effective tool to do anything.
While the techbro’s are dumping eyewatering amounts of money in the idea and the copyright lawyers are asleep it might look as if it is awesome. But actually you are using enough power to power a small city and are using millions of stolen copyrighted works.
When the user actually would have to pay enough to turn a profit (think 100$/query) you know there won’t be much demand.
When someone builds a piece of code to pay 1/10 of library rates for every time a copyrighted work is used to answer a query that number could easily double.
AI runs on ridiculously expensive hardware using ridiculous amounts of energy, stealing works from ridiculous amounts of people.
AI is nothing but hype.
I think that’s a very good summary of the situation. And here’s another thing: Big Tech is stalled out. They ran out of new “killer apps” for the PC about 20 years ago, and they are now failing to create new ones online. Capitalism means grow or die, and these companies are in a panic to find the Next Big Thing.
Meta (lol) went caca in the futon with its Metaverse (lol), but when one is so desperate to find a transformational idea, the path of least resistance is to imagine that whatever you have is what will do the trick. (By the way, it’s not just Big Tech that is having these AI-style hallucinations; it’s happening across the economy.)
AI is all they really have right now. That’s why they are throwing an insane amount of money at it.
It really is different than in the past, in a way I think young people today would have a difficult time imagining. Case in point: the Sony Walkman, introduced in 1979. It was just a compact transistor radio (and later a tape player, then CD player) that you could wear while jogging, etc., and the public went apeshit over this! The media gave it a ton of free time on TV. There were jokes about it in sitcoms, etc.
And that was just one thing. People went crazy over Calvin Klein jeans. The latest cars, TVs, appliances, home computers, cookware—everything!
In B2B, same thing. Copiers, computers, software, robotics—you name it. And this kind of thing regularly made the news as well.
Why did people get excited? Because there was a tremendous amount of latent demand.
Today, no one gives a shit about any of this. I worked in the advertising industry from 2004 to 2023, writing mostly for major Japanese companies, doing video script translations, press release translation and writing, original branding (name/feature/service names, etc.), and so on. So I saw things from the inside out to a considerable extent. And these huge companies cannot get anyone to give a rat’s ass about their products, their “cocreation” initiatives, their something something labs, or their amazing histories and heartfelt intentions (which last two things Japanese companies endlessly blab about). Hell, Honda and Nissan are in merger talks—what the hell does that tell you?! (Since when is it a good idea for a decent company to merge with garbage?—but I guess these are desperate times!)
So yeah, when AI is the only thing exciting out there, you’re damn straight it’s going to be overhyped.
Yes, and the improvement in weather forecasting was mentioned above. AI is going to be useful in a thousand different niches.
I have my doubts that it is going to replace all that many humans in the near term, not because I think there is some law of history that technology always creates new jobs (I mostly agree with Der_Trihs on this, without the Grand Guignol aspects 
), but because I think it is just hard to squeeze more productivity out of our current economic system and industrial paradigm.
Let me explain what I mean. I’m a very good translator of Japanese into English. In the decades in which I did most of my translation work (I do more on-site industrial interpretation now), I could have made a tremendously greater amount of money had Japanese companies given more of an actual fuck about the quality of the translation they used. But my big competitor wasn’t AI or someone using Google translate; it was non-natives (i.e., Japanese people) doing shitty translation.
Hell, I once had a cheap guy at a Japanese company in the States for whom I had done some work ask me to evaluate eight or so samples of translation to see whom he should hire. This was back in 2015 or so. Two were complete gibberish, which I took to be machine translation. One was too short to judge either way. And five were intelligible but had objective errors that were easy to point out. None of the translators (except, potentially, the one who had translated too short a passage) was anywhere close to acceptable, in my professional opinion.
But how about today?! Surely the machine can do it all! Well, I have a funny story. Back in September 2024, I interpreted for two days at an audit of a local company by two Japanese gentleman. Went great. On the second day, one of them had put their final summary, written in Japanese, into ChatGPT and asked me how it was—and maybe I could just read it?
Lol! It was unusable trash! It had a few objective errors in it, but the real problem was the tone. One phrase sticks in my memory: “It is commendable that…” That is awesome if you want to sound like a pretentious school principal giving an address to the sixth grade graduating class. Reading it, I really felt that technology that could really render the Japanese text (which was perfectly fine in tone, etc., for a Japanese audience) into something appropriate for Americans was a hundred years away.
Oh, and they said that they had had another interpreter at the same company before who was terrible, but I suspect that she would have done a marginally better job than ChatGPT.
So that’s the thing. AI can in theory replace humans, but we’ve had wage stagnation for 30+ years, and all those Japanese products are commodities that no one cares about, and American laborers offer commoditized labor that no one cares about, and the AI output is already a cheap commodity that increasingly no one cares about. Sounds like a winning business model!
This. The cost is huge, and the consumer (whether B2B or B2C) is not yet taking on the burden.
People speculate when AI will be able to make a whole photorealistic movie with good continuity, etc. It may be technologically possible in the future, but will it actually be cheaper than filming a movie using the current method?
Slight hijack: Which is decent and which garbage, IYHO?
I’ve owned Hondas, Nissans, Toyotas, a Lexus, and a Mazda, and quality-wise I’d be hard pressed to tell any of them apart.
I’m guessing the garbage one is the one who merged with Renault.
Right, Nissan.
It’s not a respected company in Japan, both in terms of how the company has been run for the past 20 years and product quality.
Oh, and last I heard, Mitsubishi might join the other two as well. Mitsubishi! That’s not garbage–that’s sewer sludge. Smh.
Probably. But it won’t be this kind of AI, which is just too energy hungry & dumb to scale up well. It’s be some very different sort of AI that at most uses Chat-GPT style AI as some minor sub-component. If at all.
Exactly. AI research should be focusing on doing things that humans can’t do, instead of trying to replicate things that humans already can.
From the viewpoint of the people making the decision though, replicating things that humans can do - so that those humans can be fired - is the entire point. So that’ll be what gets funded.
Makes sense!
While I have agreed with you that there is no law of history that prevents a new technology from leaving a large number of people unemployed, I don’t think the development of AI is that targeted. The “entire point,” IMHO, is to make money, as that is the entire point of the incentive system we call capitalism.
The companies developing AI are indeed proceeding recklessly. They have no control over their creation and are simply assuming (praying?) there won’t be a “lab leak.” They may end up creating something that leaves most people without jobs, but I don’t think they are trying to do that on purpose.
AFAICT, at least one of them is doing it to that purpose. I’m not going to mention his name (to avoid this becoming political), but he seems to be developing AI/robots with the intent to replace his factory workers. Also to replace government workers.
Broad scale LLMs accessible online and running off big farms use a bunch of energy. But most people I know who are really into AI tend to run it on local hardware. You can easily run a competent model of Deepseek on consumer hardware and the same goes for a ton of other LLMs which may not be as overall universally capable as ChatGPT but they still find very useful for coding, chat bot and other tasks.
The initial training of such models consumes a great deal of resources, to be sure. Actually running an LLM or image gen models like SDXL on my desktop consumes less power than many of my recent video games (Stalker 2 cooks my room far more than any AI software I have).
I have no doubt that many of the plutocrats would love to do such a thing, but my point, somewhat poorly stated, was that the developers themselves are not customizing the AI for that and probably aren’t capable of fine-tuning it to that degree. Yet.
The developers aren’t the ones who make the decisions, they’re employees who do what they are told. The plutocrats are the ones in charge, and they do want to eliminate all employees.
Now why do I see this talk of “AI being so costly that it can’t do anything” as not even wrong?
It could be because, if one takes into account What AI is doing in Biology alone, declaring AI to be that is really silly:
This whole Verasitasium video reports on what was done already and what we can expect in the future.
It took British biochemist John Kendrew 12 years to get the first protein structure. His target was an oxygen-storing protein called myoglobin, an important protein in our hearts. He first tried a horse heart, but this produced rather small crystals because it didn’t have enough myoglobin.
He knew diving mammals would have lots of myoglobin in their muscles since they’re the best at conserving oxygen. So he obtained a huge chunk of whale meat from Peru. This finally gave Kendrew large enough crystals to create an X-ray diffraction image.
When it came out, it looked really weird. People expected something logical, mathematical, and understandable, but it almost looked intricate and complex—kind of like if you see a rocket motor with all the parts hanging off.
This structure, which has been called “Turd of the century,” won Kendrew the 1962 Nobel Prize in Chemistry.
…
In October 2024, Demis Hassabis and John Jumper of Google DeepMind were awarded the Nobel Prize in Chemistry for their groundbreaking work on AlphaFold, an AI system that predicts the 3D structure of proteins. They shared the prize with David Baker, who was recognized for his work on computational protein design.
AlphaFold 2, introduced in 2020, achieved unprecedented accuracy in protein structure prediction during the CASP14 competition. The system’s predictions were often indistinguishable from experimentally determined structures, with a median score of 92.4 GDT (Global Distance Test) across all targets. This level of accuracy is considered competitive with experimental methods.
The impact of AlphaFold has been significant:
It predicted the structures of over 200 million proteins, vastly expanding our knowledge beyond the 150,000 structures determined over six decades of traditional research.
The technology has been applied to various fields, including vaccine development for malaria, addressing antibiotic resistance, and understanding protein mutations related to diseases like schizophrenia and cancer.
**> **
> The AlphaFold 2 paper has been cited over 30,000 times, indicating its profound influence on the scientific community.The Nobel Committee described AlphaFold 2 as a “stunning breakthrough” that could accelerate scientific discovery and potentially speed up the development of medical treatments1. This achievement represents a significant advancement in our understanding of life at the molecular level and demonstrates the potential of AI to revolutionize scientific research.
I don’t think that and didn’t mean to imply that.
I mean that the development companies are spending (and losing) an inordinate amount of money and not passing that cost on to the consumer (B2B or B2C) yet. It’s like Uber losing billions of dollars a year for years while charging cheap fares.
Some AI is already very cheap and 100% effective. The greatest grandmaster in the world cannot beat a basic-good chess engine these days, and that requires very little processing power. Such use cases will continue to grow in number.