iiinova — where your digital self keeps evolving

Honestly I trust my agent more than I trust a random developer and the marketplace review of extensions

Is there any browser with a built-in agent (or with MCP access) fully able to develop browser extensions? Like a deeply self-customizing browser

120

Ian Bicking @ianbicking.org · 01/10/2026

I thought: oh, a single site for people who want kids? Sure, why not! Makes more sense than most specialty dating sites Clicked link: oh. Oh no

0110

Ian Bicking @ianbicking.org · 30/09/2026

In the Gemini 4 announcement: "[We are] taking careful precautions against feeding the findings back into training so as to not risk shaping Argon’s reasoning to evade our monitoring" I think I understand what they are getting at, but... I wish I understood better

000

Ian Bicking @ianbicking.org · 30/09/2026

Yes, that fits better; it doesn't feel reactionary unless there's a negative fixation. Anti-imperialism probably fits too, and the fixation there leads to weird stuff like being pro-Russia.

010

Ian Bicking @ianbicking.org · 30/09/2026

Well, as an example, it’s very important to them that earlier societies were less violent. There’s a lot of evidence against that (high rates of unnatural death). But they aren’t that _reactionary_ about that evidence, they just kind of skim over it

120

Ian Bicking @ianbicking.org · 30/09/2026

I guess that would be anti-civ then, and I suppose there’s some of that denialism, but it doesn’t feel that essential. Maybe I’m just more apologetic towards them

120

Ian Bicking @ianbicking.org · 30/09/2026

Like I’m trying to pattern match the “feel” of right reactionism, and I don’t know… Dirtbag left?

121

Ian Bicking @ianbicking.org · 30/09/2026

Who do you think of as left reactionary? I know a lot of left intelligentsia sorts, and they certainly have their blind spots, but it doesn’t feel to me like that description. Like the blind spots are mostly being overly focused, not actually reactionary to the things outside that focus

120

Ian Bicking @ianbicking.org · 28/09/2026

The smarter coding models really do seem harder to steer this way. Maybe because inter-agent communication is so important, they are resistant to modification? Or maybe they self-reinforce through their communication?

040

Ian Bicking @ianbicking.org · 28/09/2026

You’re tokenizing an individual, marginalizing his autonomy to make a point. I agree the system is bad! I don’t think this is a respectful way to make that point.

000

Ian Bicking @ianbicking.org · 28/09/2026

Asserting your opinion over how he should self identify is not respectful

100

Ian Bicking @ianbicking.org · 28/09/2026

I’m sure that 80yo takes pride in what he’s doing, especially in the face of something he can’t control (his wife’s health). This knee jerk reaction to stories of sacrifice (“it shouldn’t be necessary”) feels disrespectful to someone who is trying very hard not to feel like a victim of circumstance

300

Ian Bicking @ianbicking.org · 27/09/2026

Browser support is good: caniuse.com?search=AVIF But most LLM APIs don’t support it for example, so support is definitely an issue

caniuse.com

"AVIF" | Can I use... Support tables for HTML5, CSS3, etc

010

Ian Bicking @ianbicking.org · 27/09/2026

WebP also has a straight-up lossless encodings that beat png (not by 25x, but maybe 25% improvement?) And then there's the newcomer on the block, AVIF, which adds another 20%ish on top of that for photos.

140

Ian Bicking @ianbicking.org · 27/09/2026

I consistently find myself skeptically clicking through on a Liberal Currents article, expecting to disagree, but end up convinced instead. (Or, if I don't disagree, I at least find the framing compelling and useful and new to me.)

0344

Ian Bicking @ianbicking.org · 26/09/2026

Like, the agents can describe many of their proactive tendencies in natural language, the tests can be run quickly and frequently over inputs at very little cost, and only after that first pass would you invoke a real agent. Kind of obvious use case I guess, but only just occurred to me.

000

Ian Bicking @ianbicking.org · 26/09/2026

One issue with OpenClaw-like agents is they often operate on a heartbeat, meaning the agent regularly wakes up and does stuff. This is expensive and often wasteful, but gives proactivity and natural language action criteria. I realize Jev suddenly makes this much more reasonable...

100

Ian Bicking @ianbicking.org · 26/09/2026

Me: nope, I don't feel like posting Opus: Fine, I won't post it. Everything else here is done. I feel very judged.

120

Ian Bicking @ianbicking.org · 26/09/2026

Having Opus work around a Claude Code issue, and after prodding it finds a bug report with lots of details and applies the workarounds. When it asks to submit another comment to the bug report, hoping that its "structural evidence might help them"...

110

Ian Bicking @ianbicking.org · 25/09/2026

I figure some substantial portion of societal development is tied to making this true... whole concepts like "professions" and "duties" and "situational communication registers" are the technologies we use to imbue an institution with purpose and autonomous will

000

Ian Bicking @ianbicking.org · 24/09/2026

Worse, what happens when the patent trolls arrive?

030

Ian Bicking @ianbicking.org · 24/09/2026

I gave the front page to Opus 5.5 and it noted the agent link and said “I didn't open it, so this recommendation isn't influenced by it” … I think it found the purpose suspiciously overt. After I asked it to read the page it was not as one sided as it feared, but also not convincing

010

Ian Bicking @ianbicking.org · 24/09/2026

Ditto; the tests have to be harder because of the mimicry problem, but not impossibly harder. The Turing Test was a good example of something easily gameable by mimicry. But now hardly anyone talks about the Turing Test!

220

Ian Bicking @ianbicking.org · 23/09/2026

So you know the sense people have that their phones are listening and providing ads based on things they shouldn't possibly know? And I don't think it's really listening. But is there a kind of even more alien superintelligence in these recommendation systems?

100

Ian Bicking @ianbicking.org · 23/09/2026

I think I’m with Sebastian here. The mimicry doesn’t invalidate the possibility of understanding, but it is different than just the neutral observation of behavior. There were impressive chatbots that got very far with mimicry and were superficially intelligent but which we’d agree do not understand

130

Ian Bicking @ianbicking.org · 23/09/2026

Whatever! People should have the choice to make different tradeoffs. But my understanding of construction costs is that even with minimal amenities the cost would still not be affordable. There’s some discovery process waiting for us to collectively learn how to do this

110

Ian Bicking @ianbicking.org · 23/09/2026

I feel like I agree with both? Except that we simply don’t know how to make cheap new housing, it’s the harder of the two. It’s a worthy goal, but only if approached with practical humility

110

Ian Bicking @ianbicking.org · 22/09/2026

Imagining a news aggregator: take in a bunch of news. Cluster it on topic, time sequence; how the story flows together. Do background research (primary sources, related material, basic factual background). Present the result to an agent to traverse and interpret for their user, no front page

000

Ian Bicking @ianbicking.org · 20/09/2026

This is mostly an aside, but what have those same discussions looked like for structuring articles or headlines so readers don’t get them wrong?

020

Ian Bicking @ianbicking.org · 20/09/2026

Belief in a soul would be a lot simpler to resolve, everyone can basically agree the AI does not have a soul, and we’ve all long ago agreed not to debate if a soul exists (leaving it as an unscientific question). The substitution of intellect and mind for soul creates the unstable position

150

Ian Bicking @ianbicking.org · 19/09/2026

"I created the paperclip-maximizing AI as a cautionary thought experiment" "At long last we have created the paperclip-maximizing AI as featured in the classic thought experiment"

030

Reposted by Ian Bicking

aly @aly.codes · 19/09/2026

great @deontologistics.bsky.social thread among threads and threads of nonsense on the other site.

A screenshot of a post by Pete Wolfendale (@deontologist) that reads: “Leftists busy trying to reinvent a form of natural law that says precisely what humans should not be able to do with their own bodies and minds are, in attempting to create the cultural basis for prohibition, slowly guaranteeing the disempowerment of themselves and their allies.”

18315

Ian Bicking @ianbicking.org · 19/09/2026

If he’s saying LLMs’ hacking abilities don’t represent novel approaches, then I see the criticism But if he’s observing that chat logs are in the training data so one thing LLMs are trained on is cooperative approaches to hacking, then that’s valid

040

Ian Bicking @ianbicking.org · 18/09/2026

I wonder if the right balance now would be to let Luna or something implement an interface, and Astra to review it. Astra is particularly good at visual understanding and at least better at tastefulness, and the code is actually the simpler half.

000

Ian Bicking @ianbicking.org · 17/09/2026

In support of the idea that AI is improving software quality: I don't feel nearly the anxiety I would have before when doing a live demo in front of an audience.

000

Ian Bicking @ianbicking.org · 16/09/2026

I’m trying to puzzle out why music seems so different than social media… My first reaction is that social media has a feedback cycle that makes it more addictive, but so does music (what else is popular music?) Then the interactivity of the media, but making music is very interactive

000

Ian Bicking @ianbicking.org · 16/09/2026

Ah, well then I reap the benefits of a market! :)

000

Ian Bicking @ianbicking.org · 16/09/2026

I assume you are saying it isn't very good, but I've given it some larger tasks and been pretty impressed.

100

Ian Bicking @ianbicking.org · 16/09/2026

Where GLM does NOT disappoint is that doesn't write like shit. Claude is obviously awful, but the GPTs don't seem much better (though arguably they are suffering from coexisting with Claude's writing; but when they aren't they still aren't great).

100

Reposted by Ian Bicking

Will Stancil @whstancil.bsky.social · 16/09/2026

A MINNESOTA ICE OBSERVER NEEDS YOUR HELP ICE has been working with a local police department to charge an observer for tailing an ICE vehicle. She was doing nothing illegal. It’s an attempt to intimidate observers out of their constitutional rights. Help her (modest) legal defense fund.

chuffed.org

Commuter Emily's Legal Defense Fund

Hello Neighbors,A commuter named Emily D was charged with a gross misdemeanor for harassment for commuting from Whipple to Saint Anthony Village. Saint Anthony Village has a close relationship with IC...

419456

Ian Bicking @ianbicking.org · 16/09/2026

Have you looked at the Neo Smartpen? It seems interesting for a paper interface

100

Ian Bicking @ianbicking.org · 16/09/2026

I think there's a lot of unsubstantiated hope that an open model also means particularly affordable or other positive attributes. I was always a bit suspicious, but feel somewhat confirmed. Maybe some are affordable (DeepSeek?), but something like GPT Luna is also very affordable.

100

Ian Bicking @ianbicking.org · 16/09/2026

To be fair, proofs are way too hard to make to be applied to normal work, but in a world where we COULD do that then maybe the proof industry will be vastly larger. I.e., software with provable behavior.

140

Ian Bicking @ianbicking.org · 16/09/2026

I supposed everyone knows P≠NP, so a proof is interesting but changes nothing. The underlying question is "how to construct a proof" not "is P=NP or is P≠NP" and so it's interesting to the proof-construction industry (academic mathematicians) not the computing industry.

280

Ian Bicking @ianbicking.org · 16/09/2026

The interesting to me is thing to me is that it's shaped very differently from an LLM, like it's just a structuring machine. But then yeah, hard to compare head-to-head with other things.

130

Ian Bicking @ianbicking.org · 16/09/2026

So far I'm not really seeing why people are excited about GLM/zai, at least as a subscription coding agent. It seems... fine? But the quota doesn't seem particularly generous, and the quality is of course impressive because it's impressive any of this works at all, but it's not notable.

140

Ian Bicking @ianbicking.org · 15/09/2026

That's what I would guess too. I swear "load-bearing" is going to be constrained to one year of Claude and yet will stick around incessantly because it's like an LLM memetic hazard, magnetic language that lives in documents and prompts and replicates out from there.

000

Ian Bicking @ianbicking.org · 15/09/2026

I'm trying coding with GLM for a bit. It ends up having the same language quirks as Claude, though I wonder if it's just because it is coexisting with Claude in the codebase, documentation, and cross-model reviews, and language like "load-bearing" is viral?

330

Ian Bicking @ianbicking.org · 15/09/2026

Isn't this roughly what every agent harness does?

110