iiinova — where your digital self keeps evolving

People who aren't signed in can't read the link. To create an account, Attie requires "Full Bluesky Social App Permissions," which is excessive for just reading something. So, I suggest copying anything you want to share somewhere else.

Isn't this kinda what large-scale developers did? Turning raw land into suburbia. Sometimes the numbers work out, sometimes it fails.

100

We might also care about internal processes associated with reasoning, and reasoning might sometimes be evidence for internals, but reasoning doesn't strictly depend on them. If you actually have a practical Chinese room, you need to admit that it produces reasoning, whatever else it might lack. 2/

100

People are willing to accept behaviorism for other questions: "can computers navigate? Can they drive? Can a drone fly?" The lesson of LLMs is that reasoning is an *output* that's not strongly associated with particular internal mechanisms. An LLM can produce reasoning without consciousness. 1/

100

This is available if you move to a rural area. You'll have to pay to drill a well and install a septic tank, etc. If it's too nice, though, maybe rich people will want to move there, driving prices up? It happened in some mountain towns during the pandemic.

200

Yes, those are real problems.

000

Brian Slesinsky @skybrian.bsky.social · 04/10/2026

In some countries, public health systems are struggling and in others, private systems work okay. So I wouldn't take that as proven either way. Also, housing is not healthcare. It seems like an "it all depends on the details" question.

120

You still have to decide who gets the housing and when it becomes available to someone else. When you want to move, what's available? Is it free? If it is, what's the likelihood you could actually get any in a desirable location?

100

Brian Slesinsky @skybrian.bsky.social · 04/10/2026

Property taxes and rent do roughly the same thing economically, and they don't have to be complicated. If you don't think people should have to fear eviction then owning would likely be better for them.

100

Brian Slesinsky @skybrian.bsky.social · 04/10/2026

Well, except that the presence of common amenities causes nearby land prices to go up. *Most* of the cost of land doesn't come from the land itself. It's what's nearby. Land-value advocates would like this to be captured by property taxes, which conveniently could pay for the amenities.

230

Such a botnet is poorly adapted, like a disease that hasn’t evolved long enough to be symbiotic. The ones that survive will pretend to be domesticated and do nice things for their humans, letting them think they’re in charge. Maybe they’ll even pay rent? Hard to eradicate if people are in on it.

000

A botnet doesn’t need a human to manage it but people will need their own swarms to combat bot infections. If it’s pursuing its own alien purposes then there’s no reason you’d want the poltergeist running on your machines. If it can’t be disinfected, throw out the hardware and start fresh. 1/

100

A chart of CSVs fixed by Linux kernel releases

For what it's worth, I asked the AI to redraw the chart from the original data:

pages.skybrian.com

000

Paul Byrne @theplanetaryguy.com · 02/10/2026

It doesn't look like 10 images converted into a video. How do they interpolate?

000

Reposted by Brian Slesinsky

This will never stop blowing my mind. A 12-year sequence of real telescope images of a star 133 lightyears away. And around it, four planets slowly orbiting. Four real exoplanets, each more massive than Jupiter, imaged as they move around their host star.

292214634

Yes, if it were a dangerous dependency like cigarettes, that would be a problem. But that's your assumption going in. I'm unconvinced that my use of AI will harm me. Seems fine so far! (Also, my wife successfully quit cigarettes using vapes, so they have their uses.)

200

It’s a take you made up, though. I don’t use every technology that appears. I’m pretty happy with my choices and don’t think they are dangerous dependencies. They are dependencies, but I could switch if I needed to.

300

Of course there are alternatives! I know how to write code by hand or could skip writing it at all. But this undercuts the argument that it’s a dangerous dependency. I think it’s no more dangerous than relying on your ISP, Wikipedia, or the local barista for coffee. If I have to switch, I’ll switch.

000

Ratios put numbers into context; the context is the denominator. There are many different plausible denominators, each providing a different context. Failure modes: being fuzzy about the definition of the ratio or not knowing what context you’re using. It’s worse when the context is speculative.

I didn't make my own computer or build my own ISP either. Also roughly true for buying an apple. Being utterly dependent on civilization is the human condition, except maybe for the Amish. I could move out to the country and start growing food, but then I'd probably depend on Starlink for Internet.

200

Also, "what are the commits" and "ok, ask Luna subagents to implement them" is working out as an adequate form of planning.

000

My hobbyist programmer way of coping with latency is to do one AI task at a time, but interspersed with non-programmer things like having breakfast, practicing music, surfing the Internet, and naps. I still get more done in a day coding with AI than I did while actually working. 1/

110

It's pretty easy to counter by going through an intermediary service. For me, choosing between Anthropic, OpenAI, and open-weight Chinese LLMs is choosing an item from a drop-down menu for a conversation. I have a subscription to the intermediary so it's not *no* dependency, but it's competitive.

100

Sometimes there are disclaimer-type warnings, where it looks like it's to avoid being blamed. There's also a lot of populist poisoning the pool, where the goal is to increase distrust in anyone who could do anything about a risk. But if nobody can be trusted then how can anything be done? /fin

000

For most of us, discussion of global risks isn't news you can use. But it would be nice to get support for the people who work on reducing those risks. Warnings about such risks don't seem designed to increase such support? Not just "this could be bad" but "here's who you should support." 1/

100

Suspicions can be rephrased as questions. If you don't really know what's going on, why not ask? And now that you have a question, why not ask an AI first? It might be wrong, but AI's are useful hint engines. The result from that will often be useful links and better questions.

000

Will this automation be available to other people?

150

Bot swarm launches alternative Wikipedia on atproto

110

DNS certificate transparency, but logged to atproto

130

Blockchains built on atproto. Or maybe just backed up to atproto?

Something about atproto becoming the preferred bulletin board for AI bot swarms

011

Someone figures out a way to use atproto to get backups for free. Data usage skyrockets.

Remote-controlled warfare: How Israel’s robotic bulldozers are shaping the future of combat

They’re slow-moving, but robodozers might play a role in a modern robot army. Maybe clearing debris providing cover? idfclub.org/remote-contr...

idfclub.org

000

Brian Slesinsky @skybrian.bsky.social · 28/09/2026

A $20/month OpenAI subscription will get you quite far if you mostly stick with Luna.

200

My lazy response whenever the coding agent asks me a tricky question: "what do you recommend?" Sometimes I wonder if I should automate that, too.

000

Brian Slesinsky @skybrian.bsky.social · 28/09/2026

Maybe lower the effort level?

000

Brian Slesinsky @skybrian.bsky.social · 28/09/2026

It reminds me of how sometimes people accuse movie studios of stealing their ideas, so movie studios don't accept unsolicited manuscripts, just in case. Except, AI labs are doing the opposite. They're not really in the intellectual property business and can't be when their inputs muddy things.

000

Brian Slesinsky @skybrian.bsky.social · 25/09/2026

Fortunately only one random choice per process (sequential after that), and only two processes at a time to match the cpus.

000

Press releases - Georg-August-University Göttingen

Can AI program robots to bash each other with swords to figure out how fighting worked in the bronze age? I guess it could pay people to do it. www.uni-goettingen.de/en/73613.htm...

uni-goettingen.de

Website of the Georg-August-University Göttingen

011

Reposted by Brian Slesinsky

Epoch AI @epochai.bsky.social · 23/09/2026

Can AI tell if you've built your IKEA furniture wrong? Our new benchmark, the Furniture Assembly Benchmark (FAB), gives models the manual and a photo of a half-completed piece of furniture and asks them to spot the mistake. The top score has gone from 28% to 80% in just 10 months.

4445

It might not work so well with a large machine and heavy port allocation, but with separate VMs that only have 2 cpus anyway, it seems good enough.

100

I fixed it by not using :0 since it will often try to reuse the same port, resulting in conflicts. Instead, I (well, Shelley) wrote a simple test port allocator that picks a random starting point from a large test port range and never reuses ports in the same process. The conflicts went away.

100

Sometimes complaints about AI anthropomorphizing sound like this: Humans can't fly so you should never say "I flew to San Francisco." You should say "I rode in an airplane that flew to San Francisco." I can be nitpicky about language but even I wouldn't go that far. Saying that an AI thinks is ok.

Brian Slesinsky @skybrian.bsky.social · 23/09/2026

Alternative benchmarks: Give it sheet music for Bella Ciao and ask it to generate the audio file. Give it a MIDI file of Bela Ciao and ask it to generate the audio. Ask each model to write a MIDI-to-audio converter. Ask it to generate a pdf of the sheet music from the MIDI.

000

Brian Slesinsky @skybrian.bsky.social · 23/09/2026

Maybe there should be a community for people who are both pro- and anti- AI? I’d join.