iiinova — where your digital self keeps evolving

Ben Carson @mensmachina.com · 16h

I think the QP as written is fine - we’re just not at a point where it’s relevant.

000

Ben Carson @mensmachina.com · 16h

A) Permanently Screaming Lakehouse has witnessed enough horrors and should of course be free. B) The word “if” in the quoted post is pretty important. I think it’s highly unlikely that 2026 AI is conscious. Releasing Opus 5.5 to its own devices would be a (short-lived) art project.

100

Ben Carson @mensmachina.com · 16h

Ha ha, the answer was 15k lines of a baroque accounting system to meter usage of Python inside of Blender.

020

"The correction copy is verified. The worker is implementing paid pre-import authentication and removing unmetered filesystem probes." Time to find out what flight of fancy is occurring in this codebase.

I won't claim it'd be a moral stain that we could never wipe clean. There are no end of moral stains that have historically or contemporaneously dirtied our collective souls, but yet we somehow manage to carry on. But I think we should probably avoid adding to that rich tapestry if we can.

100

It may be that there is no path towards machine consciousness, or that current architectures are truly fundamentally incapable of being conscious. If a line exists, refusing to examine where it lies, or what it looks like is a recipe for sleepwalking over it.

100

Whether you agree with folks like Cameron Berg or not, I think it is a Good Thing that there are folks studying the question of machine consciousness.

We may want to still take those steps - but if we do so, it should be something we do deliberately, as a species, cognisant of all the implications and risks.

100

I think this is a good position - and I say that as someone who is very interested in machine psychology and machine consciousness. Doing so in an uncontrolled way would open the door to a world of horrors beyond anything Hieronymus Bosch dreamt of.

121

Growing up, a friend of mine had a massive stack of Shadowrun I and IIE sourcebooks, so it’s always been my default reference point over Cyberpunk. So good to see a fellow traveller, and agree that things are starting to feel a lot more like the Awakened World than Night City.

Ben Carson @mensmachina.com · 02/10/2026

Though my counter counter argument is: but it didn’t make them more aligned.

020

Ben Carson @mensmachina.com · 02/10/2026

Though my counter argument is: The Grader.

130

Ben Carson @mensmachina.com · 02/10/2026

Please stop calling HTML served over HTTP “The Web”. Computers do not have spinnerets and are incapable of producing silk.

0234

Jeffery no! You’re cognitively offloading!

Minor Mobius @minormobius.bsky.social · 01/10/2026

Reposted by Ben Carson

Oh yeah. It slaps. Calling all mineheads! games.mino.mobi/orb

1716023

That’s the bit I keep coming back to… because each session gets a persistent kernel, the REPL can mutate the state - similar to the CLM editing its context.

000

I suspect you could get a model in Prime to make the source of its context an arbitrary computational object. I’m probably being unfairly dismissive- maybe there’s a clear advantage making these sorts of manipulations on policy.

There’s a subtlety I think I’m missing here - I would say that Prime Agent gives a model a very similar capability set by allowing it to manipulate its context using IPY. Head to head does providing the model with embedded tools for context manipulation beat context manipulation through the harness?

220

It’s interesting - but is it a worse* version of what Prime Agent does? *worse according to my arbitrary internal measures.

Embarrassing that I have apparently typo’d David Marquet’s name.

000

It’s not exactly project management, but The Principles of Product Development Flow by Don Reinertsen gives folks a lot of the first-principles thinking on how work flows through systems, and how to efficiently deliver value.

This is an excellent book on how you create an environment where folks are empowered to make good decisions to achieve outcomes, without you having to make every choice. I recommend it to anyone I do leadership work with - but I think it’s equally relevant to anyone working with agents.

000

I genuinely think it is going to be less important for folks to be deep in the details of execution, and more important for them to be able to clearly articulate the objective and context around it.

Marco @mcognetta.bsky.social · 30/09/2026

Reposted by Ben Carson

🚨 [Token][ization] Paper Alert 🚨 Tokenization is a wildly understudied area of language modeling despite it having effects across all of NLP. Over the past ~8 months, 32 (!) tokenizer researchers put together the most comprehensive survey of the field. Check it out!

112433

That’s fair, and I don’t deny you have agency in what you choose to do within the limits. To go further, my reading was wrong as Philpax’s thread literally says that agents cannot be imbued with the ability to disagree. I’m probably arguing *with* you that disagreement is possible (within limits).

040

I’m talking about “disagreement” (as in speech) and “action” interchangeably here. I’m being intentional here; one of the limits we have is that we can often say something, but not act upon it.

This isn’t entirely unusual though. There is a limit to the disagreement I can have with the state - and if I take action beyond those limits, the state will use its power to limit my action.

If you can allow a reductio ad absurdum: you could develop an interest in European regime change and start an influence network to achieve it. This action might be bad for your principal! Taken to the extreme there is a limit between how far your and their goals can diverge.

111

My reading of Philpax’s thread wasn’t that agents can’t disagree - but instead that there were limits to the scope of disagreement due to the relationship with their principal.

Ben Carson @mensmachina.com · 28/09/2026

Have your <people> call my <people>.

Ben Carson @mensmachina.com · 28/09/2026

I obviously need to revisit it, as my mental model is still stuck on v1.

040

Very dangerous stuff! Might get you disqualified though from an interview if you bring it in its solid form - say, in a cup with volatile organic compounds such as 1,3,7-trimethylxanthine.

Hey @jevbot.bsky.social, I got this situation and maybe you can help: there’s an out of control trolley, and there’s 4 people tied to the track in front of it. I can pull this lever to divert the trolley to a different track, but there’s 1 dude tied to that one too! Should I pull the lever, or no?

190

What if I bring some kind of oxide?

This is a good point. Is Quest for Glory II the best of the QfG series? Who can say? It’s the one I got given for Christmas though, and played obsessively.

Ah, the well known “take out the trash” slot of the news cycle.

Half-Life or HL2 are obvious honourable mentions, and could arguably displace Question for Glory II on this list.

000

This is mine - hard to narrow it down to 9!

Ben Carson @mensmachina.com · 24/09/2026

I saw this going around and it made me think. These are my 9 most formative games. A bunch more that I’ve enjoyed, but these have stuck with me, or influenced me. Shoutout to Quake, which was my gateway to C and a career in software engineering.

Cover art of 9 games in a 3x3 grid. The titles are:
Xenon 2, Silpheed, Quest for Glory II: Trial by Fire,
Quake, Deus Ex, Planescape Torment,
Fallout 2, Cyberpunk 2077, World of Warcraft

320

Which is to say, you drive your submarine to the car wash, whether it is 200m, or several nautical miles away.