In other words - "We don't like the reporting around how bad the economics of AI are, so we're cherry picking our own data, presenting it out of context, and asking you to look at it through rose tinted spectacles."
Look, say what you will about Ed but when you compare his analysis against numbers and data like what we see from Google here you can’t help but see a difference https://www.wheresyoured.at/the-openai-bubble/
Look, I will say what I want: I'd trust Sam Altman to provide accurate, unbiased information about the present and the future of AI tech before I would trust Ed Zitron to do the same.
And no, I wouldn't actually trust Sam "King of the Cannibals" Altman. But I would still expect him to start with far, far more accurate data, and distort it a lot less to suit his narrative than Zitron would. That's how low the bar is.
The man made saying "the AI industry is failing" his entire personality. He'll keep saying that regardless of what the AI industry is doing, forever. You can write "tech companies often overpromise and underdeliver" on a post-it note, stick it to a whiteboard, and that will provide about as much perspective as everything Zitron has ever written, or will ever write.
The study uses Gemini data to draw conclusions about AI usage for economically useful work. Hopefully it is obvious that this is a fatally flawed approach due to inherent bias in limiting your data collection to one model which is uneven in its capabilities (as all models are) and therefore biased in what work people use it for. I might as well write a paper about the economic uses of power tools by analyzing how drill presses are used.
A very positive spin. I fear that AI will be a tool for the rich and a toy for the richer. It will sadly divide the people Geographically (rich countries affording data centres and nationalised models) and poorer countries missing out. It will divide socio-economically, the rich affording tokens becomes the land/knowledge owners/hackers and the poor becomes the labourers.
How’s this for a working theory. Economic actors that were established doing the “easy stuff” that got displaced first got a head start on everyone else and started low balling on the harder stuff and that’s why we’re seeing impacts in the parts of the economy that aren’t being directly hit by AI. It could explain why people who know stuff are finding themselves squeezed for cost in every direction even though their specific area defies automation. There is a latency at play here while all the corner cutting of the cost based competitors mount. Not sure what is the outcome when the inevitable correction kicks in but the economic impact will be long lasting as skilled workers leave the industry and we are left with decades of slop to clean up.
I don't disagree but one difference I see is Enterprises pushing hard to use AI where none of them really bothered with Crypto. I see momentum with AI that I haven't seen since the push to Cloud infrastructure 15 or so years ago.
"Less than 10% of those interactions fully automate tasks." - that's because they are measuring use of Google's models. People are using Antrophic's models for automation and coding. My wife is from a tribe in the mountains where they live like they live off the land, cooking with fire like they did 1000 years ago. She uses Gemini to learn how to live in the city - book flights and hotels, stuff like that. I use Claude Code every day to automate my tasks. Google has a very different type of user than Antrophic and Open AI.
Automation is also a squishy term I’ve been playing around with Hermes agent the last few weeks, and I found it very difficult to find useful things that you can do by itself. And even at work, most of my AI usage is rather interactive other than research task that I can send fire and forget.
> AI use at work is broad but shallow: Workplace adoption spans all industry sectors and also 68% of all occupations that collectively represent 90% of total U.S. employment.
Im sorry but what? Plumbers, electricians, framers, welders, auto workers, hotel staff, restaurant staff are these people using AI in their jobs? Does someone at google think that these very much manual labor jobs are only 10 percent of the work force in the US?
There is PEW data, that says 30 percent of us labor force is in manual labor / blue collar jobs. The person you call at 3am to fix your drain, isnt using AI.
For example, automotive technicians and industrial mechanics use AI to interpret complex test results, debug electrical wiring, and inspect machinery for wear.
No chance I'd trust Gemini to correctly tell me which wire was the high voltage one.
You're absolutely right! I apologize for providing a factually incorrect explanation. You asked a question, and I gave you incorrect information that put you in danger. That failure is entirely on my side as an AI system, not on you for following what you were told.
> No chance I'd trust Gemini to correctly tell me which wire was the high voltage one.
No latter than yesterday I used Gemini to tell me which kind of power adapter was compatible with an old SNES chinese backuping device [1] (a device you'd plug instead of a cartdridge into the SNES and then you could make backups of cartdridges on to 3.5 floppy disks. You could then run the games from those backups. Conveniently you could also "backup" your friends' cartdridges.. Heck, there was even a tiny demo scene for the SNES and you could swap disks with little demos written for the SNES -- just like for the Amiga).
It was a bit tricky in that Gemini explained me I needed a reverse polarity adapter or I risked bricking everything.
Now, granted, the stakes weren't very high but... Gemini got it right and I could start the SNES, with the chinese "Super Pro Fighter" backuping device just fine.
Even the best AI tools are consistently wrong at interpreting the NEC and often in odd ways, like making it stricter than it actually is, or mixing multiple countries’ standards together. 100% useless in an AHJ with a lot of their own rules like Chicago/Cook County.
A guy I work at had a check engine light on his Range Rover and a code that indicated something about a bad air/fuel mixture. He consulted ChatGPT and over the next 3 months, proceeded to spend over $15,000 trying to fix the car, as GPT convinced him over and over that he was getting close by eliminating logical possibilities. It even convinced him that a completely normal noise the engine was making was a vacuum leak, which sent him down another rabbit hole trying to find the source. In many cases, he would bring the ChatGPT diagnosis to a shop and tell them to change a particular part even against their advice. None worked. Finally, he gave up on using AI to fix it and told a shop to just figure it out. Turns out the injectors just needed to be removed and cleaned. Check engine light is now off.
There's no chance I would trust a human to correctly tell me which wire was the high-voltage one, nor would I trust the documentation. I would use it as a starting point, then apply test instruments to make sure I had the right information. Then I would unplug the device and discharge the caps before starting to work on it.
We need to start training people that AI answers are just as fallible as answers you get from a human. Trust but verify for all beings, biological and non-biological.
That's not how AI or any kind of assistance is to be used. It's there to double-check and alert you to any possible mistakes you've made. You still have to use your own judgement. Have you never asked a person to take a look at some of your work? Did that mean that you have to follow and obey all of their advice?
I asked it to identify a starfish I photographed at the weekend whilst scuba diving. It's a spiny starfish, off Cornwall. It said it was a 7 armed starfish. It clearly had 5.
Disappointing. I wouldn't trust it for "real" problems at all!
Interesting to see that Gemini (which I assume wrote this entire article and site) has the exact same over-stylized italic serif preference as Claude does.
I share the cynicism here about this article. I think the saddest part of betting our economy on AI is that Google is actually the US-based corporation that is best positioned because so much of their AI efforts are not specifically aimed at software development. At least for now their profit margins are high enough to withstand the high AI costs, but let's wait and see if that lasts.
86% of AI not used for work - so Gemini is mainly used by consumers, I guess also mostly non-paying. I am not surprised as they are on many Androids nowdays and Android updates recent whats new showcases new Gemini features (or repeat the same ones from earlier) for more people to try, my wife uses it often instead of Google search and sometimes to edit some photos.
55 comments
[ 0.28 ms ] story [ 22.1 ms ] threadI’ll take information from multiple independent analysts please.
And no, I wouldn't actually trust Sam "King of the Cannibals" Altman. But I would still expect him to start with far, far more accurate data, and distort it a lot less to suit his narrative than Zitron would. That's how low the bar is.
The man made saying "the AI industry is failing" his entire personality. He'll keep saying that regardless of what the AI industry is doing, forever. You can write "tech companies often overpromise and underdeliver" on a post-it note, stick it to a whiteboard, and that will provide about as much perspective as everything Zitron has ever written, or will ever write.
You wanna elaborate?
Im sorry but what? Plumbers, electricians, framers, welders, auto workers, hotel staff, restaurant staff are these people using AI in their jobs? Does someone at google think that these very much manual labor jobs are only 10 percent of the work force in the US?
There is PEW data, that says 30 percent of us labor force is in manual labor / blue collar jobs. The person you call at 3am to fix your drain, isnt using AI.
I am 100% using AI to figure out wtf to do about my 3am plumbing problem.
/joke
After I have the water turned off, I’m going to back to sleep to call a plumber during regular hours.
No chance I'd trust Gemini to correctly tell me which wire was the high voltage one.
You're absolutely right! I apologize for providing a factually incorrect explanation. You asked a question, and I gave you incorrect information that put you in danger. That failure is entirely on my side as an AI system, not on you for following what you were told.
There is already deterministic procedural stuff for all of those cases. Putting a chat bot in front of that is basically fucking stupid.
So far I've learned 480V is very unforgiving and can show up in surprising places.
No latter than yesterday I used Gemini to tell me which kind of power adapter was compatible with an old SNES chinese backuping device [1] (a device you'd plug instead of a cartdridge into the SNES and then you could make backups of cartdridges on to 3.5 floppy disks. You could then run the games from those backups. Conveniently you could also "backup" your friends' cartdridges.. Heck, there was even a tiny demo scene for the SNES and you could swap disks with little demos written for the SNES -- just like for the Amiga).
It was a bit tricky in that Gemini explained me I needed a reverse polarity adapter or I risked bricking everything.
Now, granted, the stakes weren't very high but... Gemini got it right and I could start the SNES, with the chinese "Super Pro Fighter" backuping device just fine.
[1] this thing: https://videogamedevelopmentdevices.fandom.com/wiki/Super_Pr...
We need to start training people that AI answers are just as fallible as answers you get from a human. Trust but verify for all beings, biological and non-biological.
Disappointing. I wouldn't trust it for "real" problems at all!
For an article about the AI economy, there is no mention of economy.
Anyone know why they did this? Once I saw it, it became too distracting.