Linus Torvalds Dismisses AI Industry as '90% Marketing' (tomshardware.com) 103
Linux creator Linus Torvalds has blasted the AI industry as "90% marketing and 10% reality" even as he acknowledged AI's transformative potential. Speaking to TFiR, Torvalds said he would "basically ignore" AI until the hype subsides, predicting meaningful applications would emerge in five years.
The Finnish software pioneer singled out ChatGPT and graphic design as current practical use cases. His criticism follows Baidu CEO's recent warning of an impending AI bubble burst, claiming only 1% of companies would survive the fallout. "I think AI is really interesting, and I think it is going to change the world. And, at the same time, I hate the hype cycle so much that I really don't want to go there," Torvalds said.
The Finnish software pioneer singled out ChatGPT and graphic design as current practical use cases. His criticism follows Baidu CEO's recent warning of an impending AI bubble burst, claiming only 1% of companies would survive the fallout. "I think AI is really interesting, and I think it is going to change the world. And, at the same time, I hate the hype cycle so much that I really don't want to go there," Torvalds said.
funny (Score:1)
Re:funny (Score:5, Insightful)
They probably don't know themselves.
Re: (Score:1, Troll)
Indeed. I mean even obvious use cases like image generation deeply suck, because by now it is blatantly obvious something was AI generated and immediately can be seen as "low quality crap".
Re: (Score:3, Insightful)
Indeed. I mean even obvious use cases like image generation deeply suck, because by now it is blatantly obvious something was AI generated and immediately can be seen as "low quality crap".
The fact that someone on a "Tech" site can say this as if the current state of computer capability is static and not a frenetic state of change makes it clear how IBM could think the world needed no more than five computers. You, on the other hand, have access to decades of computer history and quite possibly a couple decades of using computers in you daily life, and yet you think that? (You're not the only one who thinks this btw.) Any 'low quality crap' you purchase these days is on the shelf at all only
Re: funny (Score:1, Flamebait)
He didn't say it would never improve. He accurately described the current state of the art and you got triggered.
Re: (Score:1, Flamebait)
Re: (Score:1)
I notice you always accuse other people of being "triggered" when, in fact, it seems like you are the one being triggered.
What makes you think I was triggered?
Ever consider that might be a personal problem you're projecting onto others?
Sounds like you've been triggered.
Re: (Score:2)
Re: (Score:2)
That nicely sums it up.
Re: (Score:2)
Effectively, they emulate me decades ago in AP English when I barely skimmed something and had to write a page about it.
I passed and sometimes the AI passes too, but it doesn't mean it actually knows anything.
Re: (Score:3)
What's amazing is you seem to think that the incredible rate progress we saw over the past century is going to continue indefinitely, despite the obvious fact that it has slowed dramatically. Even the recent hype around AI was mostly just hope for the future, driven by that same childlike belief in inevitable rapid technological progress. You'll notice that none of the imagined improvements or social disruptions have materialized, despite the many claims we saw from the faithful that the world would be un
Re: funny (Score:1)
despite the obvious fact that it has slowed dramatically
First, define progress.
Second, slowed how? Technology doesn't just mean computers, silicon lithography, or CPU integer performance.
You should know this better than anybody. Think how many street drugs are currently in your system that weren't even within your reach 5 years ago. You know how that happened? Technology. Your dealer and his distributors have had to come up with new ways to prevent their product from being detected as it crosses the border. The cartels have had to come to with less expensive but
Re: (Score:2)
In 1980, the personal computer didn't exist. There were still typing pools,whiteouts was ubiquitous and executives created spreadsheets with a pencil and paper. And laying out a flyer was literally required a "paste up". Mail, meant writing a letter and it would arrive a few days later. Long distance calls required an operator. By 1990 typing pools had disappeared, computers were used for publishing and spreadsheets. The web didn't really exist in 1994, people used compuserve and AOL. BY 2004 those didn't
Re: funny (Score:1)
Technology is definitely not slowing down, youâ(TM)re just looking in the wrong places. Just because your personal devices are fast enough not to warrant much progress doesnâ(TM)t mean there are a ton of other systems that impact you are not being affected - banking industry, cards and traditional banks are on their way out; healthcare and medicine now have the capacity to run entire models on computers, eliminating the need for large animal testing (dog, sheep etc are no longer found at major uni
Re: (Score:2)
healthcare and medicine now have the capacity to run entire models on computers, eliminating the need for large animal testing (dog, sheep etc are no longer found at major universities)
Universities have been using abstract models for a long time and then testing them. Now they are computer models, which you don't have to test. But that is a different issue. Your list of changes have pretty limited impact in the real world. If you don't believe me, don't "post" a reply here, send me a postcard. The issue is the speed of change, not that there isn't any.
Re: (Score:1)
Again, you don't "see them" because you don't interact with them in detail, you just see the results. In other places, like basic research and healthcare, computer use beyond the 'basic' office and simple data representation are just getting started. Places like NIH are just getting started with making things like data management, code/data sharing and long-term storage and re-analysis a hard requirement, once those rules become embedded, they will accelerate research. Sure YOU won't see it, but the reason
Re: (Score:1)
Re: (Score:2)
What a stupid reply. Unlike you, I have been following AI research for about 35 years now and I know that the current "advances" may seem big to somebody without that knowledge, but they are not. They are small, incremental advances. There are no low-hanging fruit left. Any improvements of the current state that are more than cosmetic will be glacially slow. And there are also fundamental limits. For example for LLMs, hallucinations will _never_ go away. They are part of the mathematics used.
Re: (Score:3)
I think it's the age-old "99% crap, 1% great work" that tainted Generative AI. Much like the early days of computing, it's an emergent area which requires good knowledge and a certain mental discipline, so-to-speak, to generate good results.
Since most people (Average Joes) have neither, their "outputs" using AI are mostly crap.
I mean, this took me a copy/paste and a click on "Create" - took two minutes.
https://suno.com/song/b8e9f12c... [suno.com]
At the same time, if I wanted to make a proper, nice song out of that art
Re: (Score:2)
At the same time, if I wanted to make a proper, nice song out of that article, it would have taken me hours of prompt refining, dozens of iterations and much thinking.
So you are saying that AI for your creation is nothing than a random stream of associations that might inspire you to put real effort into your final work. But for that, you don't need AI. David Bowie had a computer spew out random words for inspirations 30 years ago already. And I myself did some experiment with a program which sent me seven random words each day for me to write a short story... just as some means to flex my mental muscles.
So what again was the point of AI?
Re: (Score:2)
No, I am not saying that.
I am saying that generative AI enables people who spent a lot of time studying, well, IT, for lack of a better term, dabble into the arts without having to spend another whole lot of time learning various musical instruments or learning how to sing well. It might be that they can't sing well, or can't learn musical instruments beyond a very basic level, for various reasons, yet they can now "make" songs with help from a powerful tool. Maybe they can't paint for shit, but they can ge
Re: funny (Score:2)
Re: (Score:2)
https://www.youtube.com/watch?... [youtube.com]
Re: (Score:1)
The tools aren't very good at doing anything other than regurgitating random stuff found on the Internet.
The basic problem is that you can't trust the output and there is no mechanism to fix bad output so it just keeps repeating stuff without improvement.
Re: funny (Score:5, Insightful)
Garbage in, garbage out. Programmers of old were very anxious about this fact. Modern programmers - not so much.
Re: (Score:3)
Re: (Score:2)
Re: (Score:1)
I find it particularly funny they use all that money on marketing yet do such a poor job of communicating what their tools are actually good at
They are telling us everything they know about what their tools are good at.
The thing is AI has been used for over a decade in special applications, and those applications have clearly communicated what the tool was good at, and it was a good tool. But modern LLMs are just worthless timewasters.
Re: (Score:2)
That's because they are not good at anything, except extracting money from venture capital funds.
Re: (Score:2)
Have you ever seen a website of any company selling stuff to C-suits? Like Oracle? Nobody understands what in the world they are actually trying to sell - but it can create individual reports as PDFs and highlight actionable pivot points.
10% reality? (Score:5, Funny)
Linus is overly generous here.
Re: (Score:2)
That 10% is snake oil
Re: (Score:3)
That 10% is snake oil
That's not what he said.
The Finnish software pioneer singled out ChatGPT and graphic design as current practical use cases.
That's his opinion, but I mostly agree with it. Also, the photo editing stuff seems pretty useful, ick factor of photoshopping your ex out of old photos aside. Those tools are really, really good at what they do and are very useful.
Where I'm at, use ChatGPT, yes, pay for ChatGPT, I can afford it and get a lot from it so sure, invest in "AI"? F no, not right now. I mean I'm skeptical h
Re:10% reality? (Score:4, Interesting)
I don't... entirely think so?
The 10% that's not bullshit has made its way into the world. Image recognition is tech that comes from the AI field, and like my grocery store no longer requires a PLU for fresh vegetables because a camera can ID what I put on the scanner visually. That's a real improvement in a small way.
Facial recognition, as evil as most of the usage seems to be in practice, also seems to mostly work.
It's easy to get hung up on all the garbage produced by generative algorithms as all that has happened.
Re: (Score:1)
You got modded Funny, but this is worth pointing out. I've yet to see anything real coming from AI. I can't even get it to write a simple BaSH script that removes spaces from filenames, recursively.
Re:10% reality? (Score:5, Informative)
You got modded Funny, but this is worth pointing out. I've yet to see anything real coming from AI. I can't even get it to write a simple BaSH script that removes spaces from filenames, recursively.
Typo because I was hasty and I've learned I don't need to waste time fixing typos for it to work anyway. Oh you wanted it recursive, then just ask. Want it to use find, or some other way, just ask. That much is very easy for ChatGPT at this point.
Always review the output for corner cases it missed, I have not but I did test it on a directory with "a b", "c d" and it meets the minimum requirements.
ChatGPT 4o
Write a bash script that removes white space from all file names in a give directory. Replace the, with underscores.
#!/bin/bash
# Check if directory is provided /path/to/directory"
if [ -z "$1" ]; then
echo "Usage: $0
exit 1
fi
# Loop through all files in the provided directory /_}"
for file in "$1"/*; do
# Check if file contains spaces
if [[ "$file" == *" "* ]]; then
# Replace spaces with underscores
new_name="${file//
# Rename the file
mv "$file" "$new_name"
fi
done
echo "All file names in '$1' have been processed."
Re: 10% reality? (Score:5, Funny)
Re:10% reality? (Score:4, Insightful)
Poster said recursive. The solution you're posting is single directory. My solutions:
Single directory: rename --all " " _ *
Alternatively, use detox ( https://github.com/dharple/det... [github.com] )
Recursive: find . -type f -exec rename --all " " _ {} \;
Re: (Score:1)
root@neptune:/test# ls
dir1/ this\ here\ file1.txt this\ here\ file2.txt this\ here\ file3.txt
root@neptune:/test# rename --all " " _ *
rename: unrecognized option '--all'
Try 'rename --help' for more information.
root@neptune:/test#
root@neptune:/test# find . -type f -exec rename --all " " _ {} \;
rename: unrecognized option '--all'
Try 'rename --help' for more information.
rename: unrecognized option '--all'
Try 'rename --help' for more information.
rename: unrecognized option '--all'
Try 'rename --help' for more
Re: (Score:2)
I don't seem to get a '--all' parameter in Debian either (testing on buster / trixie).
Re: (Score:2)
I ended up using this:
find . -depth -name "* *" -execdir bash -c 'pwd; for f in "$@"; do mv -nv "$f" "${f// /_}"; done' dummy {} +
Maybe someone should train the AI bots to use bash. Maybe none of us know how to use bash.
Re: (Score:2)
Commit on July 15, 2021 https://github.com/util-linux/... [github.com]
Commit included in v2.38 released March 28, 2022 https://github.com/util-linux/... [github.com]
Commit log
rename: add --all and --last parameters
Renaming files with rename often involves multiple passes in order to, say, replace all spaces with underscores because traditionally
rename only replaces the first occurrence of the expression. The --all parameter makes this task simple.
With the addition of --last, rename becomes much safer to use when replacing file extensions, whereas before it would mangle a file which had its extension also embedded elsewhere in its name.
The implied --first, together with --all and --last, round out the common cases for renaming files.
Re: (Score:2)
Wow, thank you! Wish I had mod points.
Re: (Score:2)
AI has trouble with details as it has no reasoning ability. The "recursive" simply got ignored. I have seen some similar examples from students in coding exercises.
Remember the .COM bubble? (Score:1)
Re: (Score:1, Insightful)
meh - its not going away - bubble, sure - 10000000000%? no - not even close to 100000%.
LLMs will be useful for many things, even if they aren't 'AI' in any form. Simply improving human - machine communication seems like a large and obvious use. The machine doesn't think - but you might be able to easily speak to it to get things done without the silliness of past voice control schemes. LLMs make it much more natural FEELING - even if its not.
Another simple example is games like Fallout of Skyrim - NPCs
Re: (Score:2)
LLMs will be useful for many things, even if they aren't 'AI' in any form.
llms and their application are definitely ai, a contraption of llm with other heuristics and utilitarian tools can even loosely fit the definition of agi if sophisticated enough. you might just have a too romantic concept of what intelligence is. it has many forms and a very wide spectrum. any behavior that we humans might consider intelligent is accepted as such, and llms do that very aptly, as they pass the turing test with brilliance once and again.
ofc nobody is talking about conscience, let alone sentie
Re: (Score:1)
it really will help people just as much as most puppets reading the script they are given in first level customer support roles.
It will also take money from 3rd world country offshore callcenter operations and put it in the hands of American tech firms. I call that a good thing since the money stays in our economy and won't be building up theirs. America First. Build up your own tech industry and stop trying to get "foreign investment" to do it for you.
AI? Or OpenAI, Meta, and LLM bullshit? (Score:2, Troll)
It seems to me that virtually all people commenting on "AI" right now are focused exclusively on the useless general LLMs and image generation toys out there. AI - training networks and models has been used very successfully to great success and lived up to actual expectations for over a decade. "AI" itself is an incredibly useful and versatile tool.
On the other hand what OpenAI, CoPilot, and I assume Apple Intelligence are shitting out is utterly worthless, so if they are in fact talking about this as I su
Re: (Score:3)
It always amazes me that when ever we talk about AI, we almost never talk about Deepmind and AlphaFold, considering that it is the only company that can use AI to win a Nobel. To me it has been obvious for almost a decade who is the leader in the AI field, but every time I talk with anyone, even from the IT-field, they don't even know the company.
Re: (Score:2)
Because people are talking about generative AI. Just a subset which is way more visible.
Re: (Score:2)
Who are you talking to in IT that hasn't heard of Google?
Re: (Score:2)
No one said AI. No seriously go outside to your IT group and ask them individually who owns Deepmind and AlphaFold. I'll bet you a Marsbar a not insignificant portion of them respond with "IBM" or with "Deep what?"
Re: (Score:2)
Exactly. The fact that Linus is talking about this now shows he's just as much sucked in to media definitions of the state of the industry as any commoner. Ask anyone when AI was created and they'll probably tell you last year since they never heard of it before ChatGPT came out. But this is marketing at play here. I have to assume when Linus is talking about AI he's talking about what is being marketed currently. Outside of today's hype AI has been very useful for over a decade.
He's not wrong (Score:2)
AI is the ultimate hype generation tool (Score:5, Insightful)
Re: (Score:1)
I sincerely hope you're right about that. Rarely have I seen a slashdot post as worthy of the "Insightful" mod as this. The irony of the human marketdroids that hyped "AI" ultimately being supplanted by robotic marketdroids that become semantic analysts would be spectacular.
He's wrong (Score:1)
Yep, wrong. It's 95% marketing.
Still waiting for business use cases.
Re: (Score:1)
Except, it's not reliable. Even in a game, how can you constrain the thing enough that it doesn't start recommending maxi pads, in response to user input?
Re: He's wrong (Score:1)
Re: (Score:2)
Still waiting for business use cases.
Ask Nvidia about that. AI has made them a fortune.
Selling shovels, etc.
Web search as well (Score:2)
Web search is 90-100% marketing. Seems to have served Google pretty well. And indirectly Linux once people started making clusters of inexpensive PCs to serve all this stuff.
I suspect AI is going to alter how marketing is done in some very dramatic ways. And very soon, if it isn't happening already.
Depression (Score:4, Insightful)
Re: (Score:2)
When I was young, I too thought I was the smartest guy in the room, and it bothered me a lot when I worked with somebody who was smarter. But over time, I learned to embrace the fact that some others are better than I am, and I learned to love working with such people, because of what they could teach me. It's not necessary to be the smartest or the hardest working, to succeed. Just smart and hard-working *enough.*
Re: (Score:2)
Where I live we need to register our vehicles once a year, I registered mine early this year. A few weeks later I got an email "It's time to renew your registration". So I'm thinking, why am I getting this email? Did I forget to renew? As a developer I'm thinking, why wouldn't they check my registration status before sending me an email? But I called and sure enough, they just send out the email and don't even check wh
Re: (Score:2)
I get it, your gripe is that people with big mouths who are incompetent, often get jobs or promotions over people who quietly do their jobs. Yep, very true. And if you work for a company that is that clueless about who actually contributes what, you might want to consider changing jobs.
Re: Depression (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
Yeah (Score:1)
Re: (Score:2)
Re: Yeah (Score:1)
AI use (Score:2)
I use AI as a glorified search engine otherwise it is an annoyance.
Sturgeon's law (Score:2)
It's working (Score:3)
I am one of those who are sold on the technology. It makes my life easier, starting with giving me better search results. Even Bing is "good enough" when used with AI!
Agreed, but... (Score:2)
...it's only true in the mass market stuff. Just about all of it is crap generators.
Serious scientists use tools like Alpha Fold to do real work
Even Linus misunderstands AI (Score:2)
Linus joins the masses in defining AI as only AGI and other generative forms of AI. I'm an AI believer, but even I think practical AGI won't happen in my lifetime. However, there are many forms of AI that have already been used and monetized for years. Recognition and translation will continue to be the near-term successes that replace many expert systems and simplistic models.
Re: (Score:2)
Re: (Score:2)
Linus is wrong (Score:2)
Linus is wrong, only 90% marketing is too low (Score:2)
Linus is wrong. 90% marketing is I am pretty sure an underestimation of how much marketing it is.
I would put it at closer to 98-99% marketing.
And almost every result(possibly every last result--I don't remember one being useful) I have received from AI (google search) have been wrong. Making AI worse than useless as it puts useless crap at the top of the search. Other AI's I have seen results from are similarly useless. Give it a detailed search results and it plays the odds as asks "is it plugged in
AI (Score:1)
Linus Torvalds is 100% correct (Score:1)
All in a name (Score:1)
Just stop calling it artificial "intelligence" and half of these disagreements would be resolved.
What is “open software” (Score:2)
What is “open software” and how is it different from ‘Open Source’
--
The one use I have found for ChatGPT is as a natural language interface into the WEB. The down-side is the moderators have decided to protect me from sensitive opinions. The other downside it that ChatGPT will just make-up quotes and who said them, if it don't know the answer.
30 Years (Score:1)
Praise (Score:1)
Lot of people in the IT think so (Score:1)
Management won't listen to us though, so it is useful to have someone widely recognized to have said so, as a reference.
Re: So Meta (Score:2)