Post

I've started to notice that many of the most "critical"/"skeptical"/"practical" takes on AI are so steeped in pro-AI framing that they immediately give away the plot. They can't say a sideways word without first rolling out platitudes as if to calm the pro-AI before the mildest take.

The problem is that this framing grants these platitudes validity. "AI is here to stay, but...", "AI is changing [x] industry...", "AI is useful in [x] context" & so on and it's so tired and contrite that we're bought into the framing. Most recently I saw "AI amplifies what's already there"...

And I can't help but just bark at that. Because what the fuck do you mean, amplified? It's a mediocrity machine at best, so what it adds is more *slop*. I suppose if you're already slinging slop it looks multiplicative instead of additive.

"AI is changing the way we work", why? How? And for who's fucking benefit? Because the work output sucks and y'all are mostly just firing people so who the fuck asked for this?

it certainly changed the number of security bugs being found and patched in recent Chrome releases:

& we're right back inside the frame. 😔

If the frame is "AI works for some things, doesn't work for others, and has a bunch of bad externalities" – and you can't find anyone objecting to the first part – I feel like that's just because it's correct / true?

It certainly doesn't work for everything claimed by AI boosters, but I personally find it hard to argue that it's totally useless; am I too frame-captured?

(and feel free to ignore me if you're just wanting to rant, I appreciated hearing your thoughts on this)

It is legit useful for some things, but we've long since left the station on that. Frontier LLMs are being pushed everywhere as a panacea to the detriment of actual humans. Did it help chrome find bugs? Yes, but I have to question whether those bugs might be found through other means.

We've still got *so* much more to explore with non-LLM based security analysis and even if we didn't I wonder how much less expensive and damaging it would've been to just have a dedicated team comb through Chrome.

That criticism is still unfortunately gnawing at the frame and not escaping it entirely, because the whole fucking argument is now that LLMs are "good enough" and we just need to redirect more resources into it. It's not just hurting individual thought,

I think it's pretty impressive how Google has been able to ship new Chrome releases for the past 18 years with absolutely no humans doing security reviews. They've clearly been quite successful in doing so, since zero people have ever suffered from vulnerabilities in Chrome!

I'm not exactly sure what you're trying to say here – you're speaking obtusely enough that it's not clear if you're trying to be sarcastic / rude, which is discourteous when everyone else is trying to speak plainly.

Has G employed 1e2-1e3 SWEs to maintain Chrome for the past 18 years? Y Is G going to rewrite Chrome in Rust to reduce its attack surface? N Is G management willing to FT hire 1e4 IOI medalists to find CVEs? N Given the quality of Chrome's codebase, who finds more CVEs/$ today: humans or Gemini? G