| Blog | https://danluu.com |
| Patreon | https://patreon.com/danluu |
It's interesting how much (some) people are adopting AI-isms in writing even when not using AI.
It makes sense that people would fall into the same patterns after seeing them everywhere, but it's still very weird to, live, watch someone who never used to write like this, write out something that's full of "It's not X, it's Y", etc.
If you actually read it, it's a little too well reasoned and doesn't have enough nonsense phrases to be AI writing but, superficially, it's very AI-coded writing.
Yossi has this post, https://yosefk.com/blog/people-can-read-their-managers-mind.html, where he uses the term "insane employee" to describe someone who fixes problems they're not incentivized to.
In his parlance, if you want to be able to buy a car designed to keep you safe or a CPU designed to not crash, you would need an "insane company".
There were more of those when competition was less cutthroat and the market was less efficient. Overall, I'd take what we can get today over what we could get then, but there were some nice bits.
I guess this shouldn't be surprising for reasons discussed in https://danluu.com/nothing-works/, https://danluu.com/car-safety/, etc.
There's *maybe* one car company that tries to do more than the bare minimum on structural safety, but their software quality is horrendous, so ADAS, etc., are untrusthworthy (https://mastodon.social/@danluu/109667147825098339) (and software controls brakes (https://www.nhtsa.gov/press-releases/volvo-recall-urgent-brake-failure-warning-select-vehicles), etc.), because why would the market provide a safe car or non-crashing CPU? Those more to make and you can't charge more.
With CPUs, the problem is that, you really need the company to go the extra mile if you don't want random crashes.
I'm not sure there's a company you can reasonably buy a standalone CPU from that will do that. AMD has always had issues (coincidentally, the last AMD machine I owned regularly crashed due to what turned out to be a CPU bug), Intel has favored velocity over quality since getting spooked by ARM. When I was at IBM (over two decades ago), they were dismantling their culture, etc.
Sure, there are the companies that were on fire as hypergrowth startups that grow into companies with ok quality, but none of those (that I know of) are in the same quality league as companies that ship really high quality products.
The worst startups that survive fix the issues that are a threat to the business, but if the company culture is to severely cut corners, it seems very difficult, bordering on impossible, to move the culture to one that goes the extra mile on quality.
I was thinking about this prediction that we were going to see a lot more CPU bugs, danluu.com/cpu-bugs/, as yet another Intel CPU bug story is on the front page of HN.
At the time, Intel threw quality under the bus in the name of velocity. I've seen this happen quite a few times now, sometimes with the stated intention of improving quality later. I'm sure it's ever happened, but I don't think I've ever seen a company do this and then ship very high quality products later.
If you read widely cited folks saying AI economics are bad today (e.g., Ed Zitron from the quote), they're not credible. Sure, Ed writes rants, swears a lot, and throws the word "scam" around, all of which gets a lot of clicks, but the reasoning seems wrong.
If I were going to be skeptical about the economics of AI, it would be some kind of reasoning about how cheap models keep improving, so the frontier labs need to stay ahead (I don't follow it closely enough to know if that's plausible).
Why are so many people so sure that the big AI providers are losing money on inference? It reminds me of the comments about how Uber can never make money. Their unit economics were fine and they were only losing money because they chose to do so on customer acquisition.
In this case, I don't think it's even obvious that inference should be money losing today. API cost for frontier models has gone way up and the people claiming some cost for inference are relying on made up assumptions.
such as adding "sanjay ghemewat" or anyone I've tried to try to seriously improve API design or code quality.
Maybe expecting that you magically get better API design by doing the moral equivalent of saying beetlejuice three times is obviously silly and shouldn't be expected to work, but this silly thing that shouldn't be expected to work seems to work in a lot of other areas.
What areas does it work in vs. not and why? Even if there were no practical implications, I'd be curious about this.
I've been experimenting with this more (not rigorous experiments). When you tell it to use a persona, it often says something like "I'll use a distributed systems lens" or w/e, making it sound like adding a 2nd persona with the same background wouldn't help, but (for example) having "marc brooker" and "kyle kingsbury" seems to help more than 2x of either or 2x generic distributed systems persona.
There are also personas that don't seem to work even though the person would be incredible,