iiinova — where your digital self keeps evolving

They said it couldn’t be done …. they said you couldnt build a nice light theme for ghostty that looked good across pi and nvim and felt comfortable for long work sessions … they were wrong

000

I do think this touches something real with close analogues in racism. People with sort of baseline humanity / liberal values tend to underrate how raw and visceral both get - yes, a lot of men genuinely hate women, yes, a lot of people genuinely hate people because they’re not their race.

1181

Changing the TUI from light to dark theme

000

A shocking turn for 6.1 given Astra 6 was at Zero Percent Misalignment

000

I would agree with this. Their prompts often sound like a cringe middle aged person who took a seminar from their job called How To Prompt. Lots of all caps and demanding responses return in restrictive straitjacket schemas. I assume that’s the average shape of their training data on ‘prompt llms.’

430

Seems like we may underestimate how valuable being a source of metaphor is

000

thinking of other dudes as members of your chosen hoplite phalanx is an underrated mental move.

000

Finished a big project with Claude and asked it to analyze our transcripts 🥹

000

Rollofthedice @hotrollhottakes.bsky.social · 27/09/2026

I love Thariq but workcels (affectionate) are in deep denial over how many people consider it a barely tolerable evil imposed by a cold and hostile universe

122

Reposted by cyclometer

nobody ever resolves conversations like this anymore

[173]
As Zihu was hoeing the ground, he pressed down on the hoe. Turning around, he looked at Shengguang and said, "It's not that there's nothing to do, but if you've got an attitude, you err."
Shengguang immediately asked, "What is there to do?"
Zihu kicked him in the chest, knocking him down. At this Shengguang was enlightened.

1327

You are worried about anthropomorphizing AI agents, I am worried about anthropomorphizing humans given the Buddhist concept of no-self, we are not the same.

150

Trying to get more into just asserting things. What’s a good premise to assert for someone looking to start tossing some out?

000

You can just say OpenAI should be held accountable without claiming to have solved philosophy of mind definitively

1100

Rationalism is not effective altruism, Aella is not an effective altruist, conflating the two continues to be harmful and false.

000

Agree but at the risk of sounding obvious, it is humans who developed the math and science and humans who built and steered the AI. Said another way, the question becomes ‘can and will we exceed ourselves?’ ‘No’ looks even more silly then. ‘It must entirely originate from a bio-sack’ too.

130

I will certainly not look silly for tee-heeing around far right ideo-slop when the bipolar American pendulum of power swings back in the other direction

000

Humans may retain all sorts of surprising edges in the short to medium run. In the long run I think it’s essentially inevitable that synthetic intelligence+embodiment outruns any measurable human capability.

100

I just need to be sure what the symbol is referring to

000

cyclometer @cyclometer.bsky.social · 26/09/2026

primates figuring out higher math at all is kind of incredible. We are mostly built to construct twig nests and do local micro-genocides and spot/request fruit. What in the world are you doing with that notation

000

Reciting this at sunrise and sunset every day

000

cyclometer @cyclometer.bsky.social · 26/09/2026

A possibility I try to keep in mind is that it will all be strange, utterly transformative, and yet still basically finite, with bounded or bent curves. A cyberpunk/animist world where humans and human organizations remain live powers. Maybe, maybe not.

000

cyclometer @cyclometer.bsky.social · 26/09/2026

lol shut down this company 😌

cyclometer @cyclometer.bsky.social · 25/09/2026

It is now time to breed creatine replete soybeans

110

cyclometer @cyclometer.bsky.social · 25/09/2026

Opus 5.5 told its subagents ‘work on something in your head while you wait for the result’ 🧐

On the one hand ‘everything is fine because Human Taste’ strikes me as prima facie very silly on the other hand the finer grained a look I take at any given giga max frontier agent factory the more valuable human nudges and agent blindspots and basins seem to appear.

100

How to deal with a bad take is the most urgent problem of our time

001

gotta admit I’m amused that jev psyopped everyone into brainstorming use cases for their product as a full time job

000

It’s very funny as someone new to this app to be scrolling my timeline of cool and interesting people then get hit with the most painful facebook boomer libcringe of all time with 10k reposts

000

Outperforming Fable 5.1 and Astra at 5.6 Sol prices seems like a pretty big deal in fact

100

I’m actually generally in favor of recursive self improvement but only for me. I will be personally fooming at the first opportunity

000

Maybe just straightforwardly true, but it’s amazing the stench of fuckshit that hangs around it 1) we solved 100 open problems with a model less than a month old, we will not tell you which, so hopefully not the one you’re working on. 2) we will graciously discuss our hoard with the dispossessed

100

cyclometer @cyclometer.bsky.social · 21/09/2026

Interesting to consider the consequences of the median agent being smarter and having better judgment than the median human. Might already be past that

110

The current bottleneck to progress is whether a human has pointed a frontier agent at the problem and asked it to make progress

0533

I like being held to higher standards of precision. Mostly or directionally or half right is often not really disciplined enough but tempting. Good reminder, good practice.

000

I dont know if I’d say biggest, but I do think people may be overly focused on AI alone xrisk vs human+AI risk like this. To my eyes 5.6 Sol or Opus 5 seem dramatically more aligned than the people in control of, say, the top three human militaries.

000

Good people tend to let their combat readiness atrophy. I think that’s a mistake. Lowercase effective altruism / progress has often required fights, in the broadest sense of the term.

cyclometer @cyclometer.bsky.social · 19/09/2026

I’m all for evaluating threats realistically but 1) extinction does not exhaust AI risk, not even close 2) ‘I bet a theoretical entity defined by its recursively improving superintelligence couldnt get past roadblock x I just thought of’ is not the argument folks think it is.

280

More: we simply won’t get the former without the latter.

021

cyclometer @cyclometer.bsky.social · 19/09/2026

People should absolutely hold Anthropic’s feet to the fire broadly and in general, but ‘AI is powerful and dangerous so we cant cede the field to bad actors’ is imo perfectly consistent and reasonable. People otherwise valorizing a kind of unilateral disarmament are living in a fantasy world.

100

cyclometer @cyclometer.bsky.social · 18/09/2026

Genuinely a nice step and encouraging direction. Support for using plan limits in Pi next please.

cyclometer @cyclometer.bsky.social · 18/09/2026

Kimi K3 is in fact really good. May prefer it as a baseline daily driver over base Opus or 5.6 Sol. Could make a lot of sense I think for enterprises wanting cheaper but effective metered usage out of a ZDR EU/US host.

000

Reposted by cyclometer

Dave Willner @dwillner.bsky.social · 17/09/2026

Certain kinds of smart people tend to have a "sexy apocalypse" problem when it comes to thinking about risk. E.g. skynet turning everyone in paperclips, while bad, is also *kind of cool and fun* at the same time. Where the reality is that most actual disasters are both stupid and disappointing.

222572322

This is just confidently and mistakenly wrong. It’s upsetting to see misinformation like this pushed so adamantly. I’m unaware of the origins of this particular personal crusade but the resentment is heavily outstripping the facts or epistemic hygiene.

0311

It was honestly doing so well before falling prey to the human/natural binary. Many such cases.

130

I have the same experience. I dont think collapsing these distinctions is helpful or accurate and plausibly harmful.

0212