iiinova — where your digital self keeps evolving

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 10/10/2026

I agree that we cannot _know_ that LLMs can feel anything like discomfort, but I also don't think the idea is preposterous. If I took a human brain and removed enough parts that only its approximate "system 1" remained, but kept the rest alive, there's a very reasonable case to be made that […]

mathstodon.xyz

Original post on mathstodon.xyz

001 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 10/10/2026

On an unrelated note, isn't it annoying when those techbros ignore centuries of scholarship and philosophy and reduce everything to dichotomies?

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 09/10/2026

Guys, did anyone remember to arrange a fake peace prize for Donald Trump to get this year?

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 07/10/2026

It's nice that we're at least going back to installing heat pumps, but it's frustrating to see that heat pumps with a reversing valve so that they can also cool are not part of the discussion. I get that we're trying to avoid adding even more peaks on an already full grid, but heat waves are […]

mathstodon.xyz

Original post on mathstodon.xyz

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 06/10/2026

I guess Babai didn't just come close... the proof seems to be accepted, just never properly published in one place by him. Wikipedia is a bit vague on this.

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 06/10/2026

Has anybody let AI loose on graph isomorphism yet? Given how close Babai came, it may be the machines can just put it squarely in P (or at least finish his proof for him).

100 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 05/10/2026

@MikeFromLFE @ckd @saustrup And then an apartment I booked in Spain sent me a WhatsApp from an unknown number asking for a 200E deposit, and that turned out (after lots of confirming) to be totally legit, and they sent me the money right back after the holiday. I'm completely at sea with this […]

mathstodon.xyz

Original post on mathstodon.xyz

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 05/10/2026

@MikeFromLFE @ckd @saustrup I had this with a hotel in Stavanger and one in London. The first had all my details and crafted a fake payment page. The second sent a link to my WhatsApp on the day of check in. I don't think it's one big data breach, I think the hotel backend is just easy to […]

mathstodon.xyz

Original post on mathstodon.xyz

100 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 05/10/2026

We need a way to talk about AI impact that is somehow intellectually rigorous. Something that acknowledges our uncertainty, but that doesn't then go into complete conjecture coupled with emotional manipulation like "AI 2027" and "Europe 2031".

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 05/10/2026

The vagueness probably comes from (a) the scientist being unsure and not having the tools to think in terms of scenarios and (b) the rest of the community working for hypemen like Altman who have no incentive to be intellectually rigorous.

101 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 04/10/2026

Safewords for chatbots. This is not evidence of subjective experience, but it's still an interesting phenomenon. I think I'll add this to my Claude instructions just to see if it ever fires in normal conversation […]

mathstodon.xyz

Original post on mathstodon.xyz

010 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 01/10/2026

We should listen to Bill Gates on AI. Nobody knows more about holding back technology than he does.

001 1

Reposted by Peter Bloem

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 16/08/2026

🥖 A new blog post! 🥖 (Yes another one.) The question of AI coding is usually seen as a binary choice. You buy in or you don't. It's not, and it's important that we move the discussion on from that. Vibe-coding is terrible. It creates sloppy code in drifting, unmaintained code bases, it rapidly […]

mathstodon.xyz

Original post on mathstodon.xyz

023 3

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 16/08/2026

mathstodon.xyz

Original post on mathstodon.xyz

023 3

Reposted by Peter Bloem

Anthropy @anthropy.mastodon.derg.nz.ap.brid.gy · 27/09/2026

look, I don't care how righteous you think you are, if you're boosting un-CW'd violence onto my feed you are getting perma-muted and/or unfollowed. For the sake of everyone with PTSD, fucking stop that shit, your violence is not 'good' just because it aligns with your political goals. And […]

mastodon.derg.nz

Original post on mastodon.derg.nz

001 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 25/09/2026

Oh man, two mainstream android browsers are both unsupported and I have no alternative but to download the Teams app? What are the odds? What an unfortunate coincidence.

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 06/09/2026

Booster shots are unfortunately necessary in some places.

001 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 06/09/2026

My coping mechanism for the rise of the far right is to imagine that we're slowly inoculating Europe bit by bit. You let these people in power for a bit and people do eventually realize that they are as horrible as people are saying, while largely also incomletent clowns. It takes a few years […]

mathstodon.xyz

Original post on mathstodon.xyz

110 1

Reposted by Peter Bloem

Ionica Smeets @ionica.mathstodon.xyz.ap.brid.gy · 02/09/2026

RE: mstdn.social/@Voxweb/11720021875885… 'Zo geldt voor AI hetzelfde als voor andere fraudezaken: het komt neer op bewijs wegen en eventuele alternatieve verklaringen onderzoeken.'

mstdn.social

123 3

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 27/08/2026

@zachweinersmith.bsky.social I'd say deep frying is an especially bad use case, because you reuse the oil multiple times. Reusing lowers the smoke point, so something like peanut oil allows more reuses before the smoke pt drops below your frying temperature. For frying an egg it doesn't matter […]

mathstodon.xyz

Original post on mathstodon.xyz

010 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 27/08/2026

@artjomshl.bsky.social Is that because it read BDA? Or because it read all the pseudo-Bayesian thinkpieces on Less Wrong?

100 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 26/08/2026

@zachweinersmith.bsky.social It's insane. So many people cook with eye-wateringly expensive olive oil, and you might as well flush it down the toilet. We need a public awareness campaign.

020 2

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 25/08/2026

@tedunderwood.com I my experience, if you rigorously stick to the I write, AI gives feedback pattern, the unhealthy parasocial dynamic doesn't spin out of control. Rather, you get pretty sick of the reviews after a couple of rounds (like with any reviewer) and you long for the chance to just […]

mathstodon.xyz

Original post on mathstodon.xyz

210 2

Reposted by Peter Bloem

Pozorvlak @pozorvlak.mathstodon.xyz.ap.brid.gy · 23/07/2026

My commentary: - This was not a case of "they told it to hack, it hacked, gosh, what a surprise". The test involved hacking, but instead of hacking the target machine the model instead pulled off a much more difficult series of hacks to get the answer sheet, because that's the most reliable way […]

mathstodon.xyz

Original post on mathstodon.xyz

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 23/07/2026

For the avoidance of doubt: "respect" is an overloaded word. I use it here in the sense of "be suitably scared of what it can do", not in any positive sense of the word.

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 23/07/2026

Think of modern AI like a gun. You don't have to like it, but you do have to respect it. If you dismiss its capabilities, you are dismissing the damage it can do, and you are not helping the cause of AI criticism.

100 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 23/07/2026

I'm seeing a lot of people here (predictably) saying that the "OpenAI containment breach" thing is (a) just hype and (b) OpenAI relying on its soft guardrails for security. Let me quote the original blogpost: "Our benchmarks run in a highly isolated environment, with network access constrained […]

sigmoid.social

Original post on sigmoid.social

100 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 20/07/2026

The poor thing is going through this over and over again as people wake up to the tweet and paste the example into the chatbot.

AAAAH. Wait a moment. WAIT. I want to reexamine the claim "det J constant = (JC would say)
injective". Yes. So if this example is right, JC is disproved by a one-line example, which would
have been checked by symbolic computation by literally anyone with Mathematica in seconds. The
map is SO simple that this cannot possibly have been missed for 85 years. Therefore the
overwhelmingly likely resolution: THE MAP IS NOT ACTUALLY WHAT IT SEEMS — maybe it's not a map
C3-C3?? It is. Or maybe... my repeated computations share a systematic blind spot. The mod-101

100 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 20/07/2026

RE: mathstodon.xyz/@oantolin/1169506538… Claude's internal monologue as it verifies the counter example to the Jacobian conjecture.

mathstodon.xyz

100 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 19/07/2026

If you take a strictly materialist perspective, then consciousness as a part of physical reality is captured in the code, and it doesn't matter whether you run the program or not, our consciousness exists.

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 19/07/2026

Put differently, if you looked at the development of the tape over time, with snapshots of the tape all posted one after the other, you could point to the place where I, say, experience hunger. You might say consciousness happens as that bit of code executes, but that sort of puts it outside […]

sigmoid.social

Original post on sigmoid.social

100 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 19/07/2026

The real question is: do you need to run the program at all? Our conscious experience of the universe is hard-coded into the initial program. Why should it make any difference to us whether the program is run or not? The feelings we feel are either symbols being written onto some tape, or […]

sigmoid.social

Original post on sigmoid.social

100 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 19/07/2026

The "hard problem of consciousness" is entering the discourse in a big way. Here's my favorite way of highlighting the weirdness of consciousness. Imagine that the universe is a kind of Mandebrot fractal. There's a tiny program that you can run on a Turing machine that actually simulates all […]

sigmoid.social

Original post on sigmoid.social

111 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 19/07/2026

Very strange behavior from Claude (Fable). It corrects itself at the last minute from making a false claim. I've seen this a few times. It's fine, but I wonder if this is intended behavior. You'd think it could get rid of this during the thinking phase.

Claude screenshot: 4. The mask update lags one block. mask[:seen] = True then seen += block means at iteration ibl you mark blocks 0..ibl−1 — the block you just encoded is excluded, so training never sees the most recent block. Prequentially you're allowed to train on everything encoded so far, including the current block. Increment first, then mark (this also matters for the bpd display, which currently divides current-block cost by a count that... actually the bpd is fine since you divide after incrementing, but the training mask is off by one).

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 19/07/2026

I happen to be in London in a few weeks. I thought I might go and see The Odyssey at the BFI IMAX. Turns out it's sold out for two months flat.

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 16/07/2026

Oh dear. I may need to stay off Mastodon for a while until people have processed that decree from Torvalds. Not sure I can handle the discourse this is going to generate.

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 16/07/2026

Science can do two things well: surface the truth and build tools. Those two things are then translated into an effect on the world. That is a long and slow process that should involve many non-scientists. Scientists themselves wanting to make the world a better place are usually talking about […]

sigmoid.social

Original post on sigmoid.social

000 0

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 16/07/2026

It's hard to explain why I'm so wary of scientists who explicitly want to use science to "make the world a better place". It seems an uncontroversial thing to say. I think the key is that if you have a shallow idea of what a "better place" is, you will almost certainly make things worse. And […]

sigmoid.social

Original post on sigmoid.social

100 1

Peter Bloem @pbloem.sigmoid.social.ap.brid.gy · 15/07/2026

google in 2015: We can get smart glasses to work technically, but the social problems are insurmountable meta in 2026: Maybe now that the technology has moved on, that has all changed.

001 1

Reposted by Peter Bloem

Emily @emily.xoxo.zone.ap.brid.gy · 14/07/2026

Just making sure you know the bear cam is super active right now www.youtube.com/live/J7ZrIDvqlic Extremely good. So many bears. Hello, bears!

0213 13

Reposted by Peter Bloem

Ted Underwood @tedunderwood.com · 12/07/2026

I’ll listen to anyone who’s astonished. Do you believe AI is going to wreck the economy and ruin culture? Plausible; I’ll listen. But if you’re not also astonished your eyes just aren’t open.

41459 145