Thomas Johnson

[Thomas Johnson @ThomasJ02]

A new doomer meme is saying that asking how AI kills all humans is like asking how Magnus Carlsen would beat you in chess. If you've played chess competitively, even at very low levels, you have an actual answer to this:
* He'll know the theory of the openings better than you, and so he will know the best move to play without having to do any thinking because he's memorized them
* He'll very rarely miss a tactic that allows him to win material (although this still happens; even super-GMs miss mate in 1 when under extreme time pressure)
* He'll have a much bigger set of motifs in his head - he has thousands of games memorized (see the terrific examples at https://t.co/ucbSFae8xx) and he can draw from these patterns when playing
* In the endgame, he's studied and memorized the exact optimal moves to play with common piece setups.

These are just a handful of examples.

People with high p(doom) need to admit their forecasts are made on vibes. We shouldn't doubt their sincerity - Coxon and other researchers 100% believe in what they're saying, just like people did in other Doomsday cults.

We need more, better research on safety and alignment! It's super important! But the current "AI will kill us all but I don't know how" damages the credibility of the research.

[Thomas Johnson @ThomasJ02]

@allTheYud @greg16676935420 It is only due to our highly superior intelligence that we have succeeded in causing animals like mosquitos, rats, cockroaches, and all invasive species to go completely extinct.

[Thomas Johnson @ThomasJ02]

The fundamental issue is that you insist on preaching to the choir. You claim that you want to make people more aware of risks, but then you include a ton of in-jokes and obscure references that are only understandable by people who already are deeply engaged in your AGI/ASI/doomer-pilled subculture.

[Thomas Johnson @ThomasJ02]

@totlsota LTCM would like a word

[Thomas Johnson @ThomasJ02]

@andrew_n_carr Same vibes as Magnus Carlson being able to identify chess games from decades ago from looking at just one position

[Thomas Johnson @ThomasJ02]

It's pretty surprising that modern coding agents still don't use any tooling like debuggers or profilers

[Thomas Johnson @ThomasJ02]

A very strange thing: People who don't use AI much think that the AIs are less capable than they are (that's why they don't use them). But those same people are quickly convinced when the AI tells them that it can do the thing, even without a demonstration.

So they don't trust the AIs, but then one response later they do.

[Thomas Johnson @ThomasJ02]

If you're going to fail every student who 'cheat's with AI, you should fire every member of faculty and staff who uses AI as well.

[Thomas Johnson @ThomasJ02]

@Aella_Girl Improved ability to determine the right next step towards goal based on past experiences. Age makes fluid intelligence decrease but crystallized intelligence increase.

[Thomas Johnson @ThomasJ02]

I published my first ever article on arXiv. Just a little side project I've been working on using a simple transformer architecture to play like humans and predict the outcome of blitz games (including things like time pressure), plus some preliminary results on modeling human thinking time.
https://t.co/8tY5EM15X0

[Thomas Johnson @ThomasJ02]

@tszzl @tenobrus Ship the Linux app. How have we forgotten to ship for the OS that is literally designed for coders

[Thomas Johnson @ThomasJ02]

Essentially all of this can be handled by a qualified night nurse. If you plan to breastfeed, you might be awakened for a moment or two when they bring you the baby and take it back when it's done.
If you're screwed on just 15 min less of sleep you probably want to talk to a doc about that, it's pretty not normal

[Thomas Johnson @ThomasJ02]

@totlsota This is actually legitimately how an accidental shot is described whether you’re at the range or at a competition or in security. In shooting competitions (eg USPSA) a “ND” results in an automatic “DQ” (disqualification)

[Thomas Johnson @ThomasJ02]

I see what you did there

[Robert Youssef @rryssf]

researchers at Max Planck analyzed 280,000 transcripts of academic talks and presentations from YouTube

they found that humans are increasingly using ChatGPT's favorite words in their spoken language. not in writing. in speech.

"delve" usage up 48%. "adept" up 51%. and 58% of these usages showed no signs of reading from a script.

we talk about model collapse when AI trains on AI output. this is model collapse, except the model is us.

[Thomas Johnson @ThomasJ02]

@dieworkwear Agree, Gen Z does not fully appreciate fiber to the curb compared to Gen X copper twisted pair.
Oh wait you were talking about clothes, nm

[Thomas Johnson @ThomasJ02]

This AI companion from @Lethain is the first time I've seen something like this besides @tylercowen's GOAT book.
I think this will become the new norm for a lot of non-fiction.

[Thomas Johnson @ThomasJ02]

Claude has gotten sassy

[Thomas Johnson @ThomasJ02]

Hosting sleepovers makes me anti-natalist

[Thomas Johnson @ThomasJ02]

What if you wanted to run a lot of vending machines?

I just calculated Sharpe ratios on Andon Labs’ Vending-Bench-2.

Claude Opus 4.5 edges out Gemini 3 Pro — even with slightly lower final PnL.

Chart below.