We need to talk about the people who still think “AI will just go away”.

Even the pump-and-dumps never went away — they just got formalized into Gambling On Everything With Insider Trading Characteristics

Also, the latter assumption is often based on the fact the crypto bubble popped. But crypto’s still around and pouring billions into political lobbying. So even that assumption is based on a kind of lack of object permanence (‘if it’s not in front of me, it doesn’t exist’)

But this idea still kicking around that a bubble will pop and everything will just go back to the way it was? More than a little ridiculous.

It may just turn out long-term to be a cost centre under engineering rather than a core revenue engine, and, in the bearmost case, something only homebuild hackers, HNWs who pay corporate rates for personal access, and people in enterprise get regular access to.

The underlying tech is useful for a range of tasks, there are no easy substitutes for ‘context-aware natural language search’ or ‘synthetic text generation’, and they will continue to be used (like earlier gens of neural nets) for as long as the need exists and no better solutions are present.

And even should such a failure happen, the idea that the IP, talent, and assets *won’t* be acquired by another business with an established revenue engine who will continue R&D and commercial application (just with greater engineering and financial discipline) is downright silly.

Google, Microsoft, and Meta, like Alibaba, Tencent, Baidu, Bytedance, and High-Flyer, are still going to do LLMs even in the very unlikely event the upstart labs completely fail.

we made being a charlatan a viable career path — TV talking heads, opinion columnists; even journalism proper has become little more or less than industrialized charlatanism

The FT’s AI optimism rests on shaky science Last weekend, the Financial Times published an article about the broader social impacts of AI. Studying, managing, and mitigating the negative impacts of the technological transition is supremely impo...

So the FT has this in-house data guy who makes a lot of shit up and vibe-stats his way to justifying his weird ideological whims

most recently he wrote a piece on how AI would make everyone centrist

I wrote about it, and for you methods nerds, you will never believe it

unherd.com/newsroom/the...

Great piece in FT on the crisis of humanitarianism.

archive.is/OBSGt

re: polls, this the most recent data from this summer

things have changed a lot recently, kevin. these were in our elite paper in the past couple days.

more importantly, there’s a deep conceit to dismissing an entire country’s growing concerns about threats to their sovereignty from an increasingly aggressive neighbour

like under what threat model is it not possible, DoD has been writing plans for decades on how they’d take every country on the planet

I think you have your head in the sand, assume there’s even remotely force parity, and assume the risk of a handful of Americans being captured in Canada is equivalent to senior Canadian staff being taken in the US.

also, the moment troops start moving around it is way too late to do anything.

I also think where you see 4D chess and deliberate distraction, we see an extended effort to shift the overton window, normalize the possibility of seizing new territory, and warm the population to the prospect over time

I don’t think you grasp the seriousness here, or how little telegraphing matters, or how joint command makes Canadian officers easy to seize and capture

for us it was never about the imminence of an invasion so much as it was about the need to respond to a very significant change in US posture and global security architecture that could easily escalate to acting on stated claims, for which we need to massively alter our economy and society

I don’t think any targeted military action would involve conscripting academics, no

I am sure as an American it is very nice to be able to just say “nah he doesn’t mean it” but that’s just not a stance you can reasonably take when you’re on the receiving side

by the time there aren’t it will be too late to prepare, which is what we’re doing

we have substantive reason to be concerned up here, kevin

truly we have an object permanence problem

the crypto bubble taught a bunch of people that if they just plug their ears and yell ‘fake’ for long enough, things they don’t like will just go away

(even if those things are still very much still around and the people behind it are funding a superpac of unprecedented size)

Context Widows or, of GPUs, LPUs, and Goal Displacement

Nature would do well to publish more content like this thoughtful piece from @kevinbaker.bsky.social and fewer Buzzfeed listicles gussied up as career advice "Five productivity hacks for using AI in your scientific workflow"

It is an amazing piece!

the most striking part is America is pursuing something whole-hog it hasn’t even clearly defined

at least Star Wars was deliberate in the 80s, but now? oof.

the fact China seems to have a much clearer idea of the state of AI and where present tech is on the innovation curve than their American counterparts (who are still wholly consumed by the hype machine) is profoundly concerning, even in the event they *are* a little behind