Ask HN: Your thoughts on Linus Torvalds' view on AI: 90% marketing, 10% reality?
Linus Torvalds reportedly grimaced as labelled the current state of the AI market as “90% marketing and 10% reality.”
“I think AI is really interesting, and I think it is going to change the world," he noted, "and, at the same time, I hate the hype cycle so much that I really don’t want to go there."
"So my approach to AI right now is I will basically ignore it because I think the whole tech industry around AI is in a very bad position," Torvalds added, "however, it seems like there is almost too much AI BS around for the Fin to tolerate."
https://www.techradar.com/pro/linus-torvalds-slams-ai-as-90-percent-marketing-and-10-percent-reality Linus being generous - it however is actually 99% marketing, one percent reality. - The text output generated by LLMs is immediately recognisable, is not reliable enough that you hand it over writing wikis on any legal, technical or financial aspect. - Code examples generated by LLMs CANNOT work without supervision and review of an experienced engineer. - Outputs of various image models and their generated artwork is immediately recognisable no matter if it is in YouTube thumbnails or blog posts. For any serious work, you need a human designer to design the marketing websites etc or you are guaranteed to be not taken seriously by anyone. Upside of this AI spring (after a long winter) are mostly in text to speech and speech to text which are not perfect but very useful in many domains. > Code examples generated by LLMs CANNOT work without supervision and review of an experienced engineer. My experience as someone who also hates the hype cycle is that I forced myself to try it to generate code despite being deeply skeptical. I havent had a single good experience, meaning getting the code that does the thing I want. I've also tried leetcode exercises and also failed here. And that's what surprised me the most - surely if anything it should be able to get right publicly available questions with publicly available solutions? I don't get it. Once I asked it for a name of people, and I asked it to format it as an html table and it did that right which was nice... > Code examples generated by LLMs CANNOT work without supervision and review of an experienced engineer. While the code generated by AI is generally garbage, I would note that it often has better-than-human debugging ability. If you copy-paste a function that isn't working, it can often figure out what you were trying to do and then explain what you are doing wrong and how to fix it. That is the point.. AI is a useful tool for many professionals to have on their toolkit.. But it is not even close to replacing those professionals like many have being claiming.. Pretty obvious thoughts? Modern internet society tends to oversell everything of small interest tenfold. Idk why exactly he hates the hype cycle, but for me it is the complete unsearchability of the topic. You can’t find anything because everyone shills, sells, shills again. Stupid tutorials, handwavy explanations, all sorts of obsolete copypasted recipes, “startups”, spam. We are an absolutely obnoxious kind for the internet and you can’t do anything but join the damned hype until it dies down. I hate the word and everyone who rides it with all my soul strings. A planet of apes. I think Linus is correct, and demonstrating his new abilities with diplomacy with his phrasing. Take a look at what the Metaverse bubble has left us vs what it cost, in direct dollars and in lost opportunities to pursue other goals. this cycle of AI is as bad or worse. at least the leftovers when the bubble pops are going to be more generally useful than the VR detritus. Getting the Cult of CurrentThing to stop mindlessly objecting to power generation is a big win; but i doubt they've secured that long term. Leftover nvidia vector processors may have legitimate uses after they get scrapped from the big AI places, too. My thought is simple: "sounds about right". Well at least the good news is that we aren't doing anything mad like hinging large sectors of the economy to the commercial viability of the technology, and it isn't too energy intens...ive... Oh. According to ChatGPT: "Some estimates suggest that a significant portion—perhaps 30% to 50%—of what you see in marketing claims about AI may be exaggerated or overly optimistic. " My own estimation is that is on the low side.