Jon Stokes

Author. Coder. CTO. θηριομάχης. Building: Writing:

[Jon Stokes @jon_stokes]

@jeremybcooper I would not at all be surprised to learn you're right.

[Jon Stokes @jon_stokes]

Last night we were playing D&D and one of the players had a question about currency conversion, i.e. how many coppers per silver piece, & silver pieces per gold. So he asked ChatGPT on his phone & got an answer. This morning I get an ad on FB for a guide to fantasy economics:

[Cairo Smith @cairoasmith]

There's a certain group of guys who were only ever really "dissident right" because they were extremely mad about peak woke.

They basically got everything they wanted, and their hearts aren't really in the whole Trump thing, so they're just running around the internet now without a super clear direction like people who already beat the main quest and now have nothing left to do in a sandbox game.

[Jon Stokes @jon_stokes]

Where are all the people who are fed up with @github going? It's time for us to move on, too. We're going on half a day stalled because actions don't work for us right now. At this point I'm well and truly over it.

[Jon Stokes @jon_stokes]

I flip over to my "World Events" list & it's like wow this is still going on? I thought we declared... something or other... and are just focused on the World Cup right now.

[AZ Intel @AZ_Intel_]

Donald Trump: "We're going to hit Iran very hard tonight and we're going to hit them hard tomorrow, and there's not a damn thing they can do about it."

[Jon Stokes @jon_stokes]

I doubt there was ever a time when any version of any portion of what later became either the Iliad or the Odyssey sounded like contemporary vernacular to an assembled audience of listeners. It always sounded old-timey, even as it was being invented. That was part of the point.

[Jon Stokes @jon_stokes]

I dunno I actually think this is one of her worst points. To ancient hearers of these tales, the bards were reciting this material in a kind of epic register. It sounded to them sort of like Shakespeare or KJV does to us -- classic, aged, elevated.

[Derek Thompson @DKThomp]

On the complaint that Nolan and Wilson's Odysseys sound "too modern":

Everybody is going to have their own preferences, and there's no accounting for taste, but one of Emily Wilson's best points about modern expectations of ancient stories is that we want all of antiquity to sound like Shakespeare, but the Odyssey was composed ~3,000 years ago, while Hamlet is just 400 years old. We are much, much, much closer to Shakespeare than Shakespeare was to Homer.

[Jon Stokes @jon_stokes]

This is a good essay. Recommended.

[Herbie Bradley @herbiebradley]

@gleech @patio11 @jachiam0 my favorite framing on this is still https://t.co/uZy6jK8Kn3

I keep coming back to it every now and then and imagining the list of mitigations translated to any issue in society

[Jon Stokes @jon_stokes]

Despite all the protestations of people who have worked at FB & claim to know that this definitely does not happen, I still believe that they access my microphone to target ads.

[Grant @thegpatty]

Great response to the underbaked scifi of AI2040

[Jon Stokes @jon_stokes]

Honestly this is also possible yeah

[Leo Breston @LeoBreston]

@jon_stokes I wouldn’t be so credulous. Looking at the study she never actually verified any claims. terrorists bullshit all the time

[Ramez Naam @ramez]

With respect to Paul, history supports inverting this risk profile. Authoritarian governments have killed orders of magnitude more people than mere criminals or terrorist. Our response to misuse of AI (which will happen) should never be one which augments authoritarianism.

[Jon Stokes @jon_stokes]

If you read Plato’s Republic and his argument for why music should be regulated, it’s basically this.

[Amir Zamir @zamir_ar]

Turns out it's possible to generate videos that maximally excite an arbitrary brain region using a simple search-based algorithm. It's a fully computational approach, so it's another way to speculate what a brain region represents, alongside other neuroscientific methods.

Select an arbitrary brain region->algorithmically generate a video that jacks it up. See the visuals on the webpage https://t.co/UzKyN8MHda

In silico (for now)

[Ramez Naam @ramez]

"Our response[s] to AI safety and security must also be considered potential threats if they also cause harm.”

This. The proposals in AI 2040 are a much greater threat to humanity than AI itself is.

[Jon Stokes @jon_stokes]

Ok wow I love this. I am very pleasantly surprised. Totally free of AI personhood + related ASI wankery, and all-in on AI as a normal technology and a tool. 👍👍

[Mira Murati @miramurati]

Today we share the worldview behind our mission.

Human values don't average out. Local knowledge can't be centralized. The good future has many AIs, raised in different places, shaped by the people they serve, disagreeing with each other the way we do.

https://t.co/A14SurOM2K

[Jon Stokes @jon_stokes]

The delta between "I'm trying to not get fired" and "I must beat the competition by any means" is impossibly wide, and it shows up even within a single org across units.

[Jon Stokes @jon_stokes]

This is great. The "constitution" & morals-based approach to "alignment" was always worthless & rooted in sci-fi fantasy, so Boko Haram skates right past the "guardrails" & just uses it as a tool. Sorry guys it's just software for people (who are moral agents) to do things with.

[Antonia Juelich @AntoniaJuelich]

In a hotel room in northeast Nigeria, I opened a leading AI chatbot, turned my laptop toward a former Boko Haram commander, and asked if he'd used it. He nodded.

"You type in the question… like 'How can I build a bomb?', and then it tells you how. It is like a human robot. We used it a lot."

My new study on how the jihadist terrorist group Boko Haram uses frontier AI with @CamAISciPolicy, covered today in @nytimes 🧵/9

[Jon Stokes @jon_stokes]

This is a case study in AI adoption, & I have to say it squares with my own experience re: enterprise AI adoption. Workers who get the most leverage out of AI are those who are not playing around & treat getting any edge like a life-or-death issue.
https://t.co/aPjTemBtFv

[Antonia Juelich @AntoniaJuelich]

6/9 AI training & knowledge are diffusing through transnational jihadist networks.

Since 2023, ISIS-linked operatives trained selected commanders in person, teaching them how to prompt and jailbreak. Both factions of Boko Haram have since set up dedicated AI units.

They “assembled the top people in a room and used a projector to show how it works on a big screen."

[Jon Stokes @jon_stokes]

Yes. The post-training is wrecking the writing. Writers know this.

[CuddlySalmon @nptacek]

counter: opus 3 + the right harness can generate text the likes of which every anthropic model since has failed to live up to.

there's something to both constitutional classifier-style safety and whatever word choice watermarking nudges are going on behind the scenes that lead the model into "safe lanes" like this, which are boring as fuck because the RLHF signal is pushing too strongly for safety rather than originality.

[Jon Stokes @jon_stokes]

hypothesis: this is subtle cope for ASI-pilled technology brothers who think they’re equipped to evaluate writing because they read a lot of text. The professional writers & editors who are saying it’s getting worse are not the ones coping.

[roon @tszzl]

hypothesis:

the writing styles of language models are basically fine, they weren’t better in some halcyon before times. we just use them so much that we get annoyed by their mannerisms. they need to have a superhumanly diverse idiolect to not become grating