iiinova — where your digital self keeps evolving

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

I think @andymasley.bsky.social has a valid take over on the other site:

Tweet: "Basically any belief you pick about AI has a bizarre amount of money behind it and you kinda gotta pick your poison."

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

I think a better approach is "follow the attention" and then dig into the ideas and arguments critically. Far too much engagement is about who belongs to what group, rather than what they actually say. Sometimes people try to say things about reality... and that can matter.

100 1

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

"Follow the money" is often used instead of critical thinking about facts and ideas, as if seeing an organized interest behind something means it can now be discounted as mere propaganda.

100 1

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

The closest I got to seeing this kind of messy dynamics was when I did an OPSEC seminar at FHI in 2017, noticing that as AI was becoming strategic we should be a bit concerned... but most focus was elsewhere.

100 1

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

I think the correct ideas do tend to win out in the long run. But the long run can be very long indeed, and for some risks we may be all dead before convergence.

100 1

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

I am a believer in the marketplace of ideas, but everything sells better with marketing! So unfortunately, absent an enlightened disinterested debate about the topic aiming at generally agreeable compromises, we get this mess...

100 1

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

One of my big kicking-myself-for-not-seeing-it-despite-claiming-to-be-a-futurist, is that of course once AI is big business or geopolitics, AI safety would be buffeted by interest groups. modelrepublic.substack.com/p/who-is-beh...

modelrepublic.substack.com

Who is behind the AI safety backlash?

Following a viral resignation, the backlash against AI safety was sudden and swift. Various signs suggest an industry push is supporting it.

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

OK, I could not resist doing a metal song. suno.com/song/7f2813d...

Hafnium

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

000 0

Suno also recognized an electron wavefunction and made appropriate lyrics. suno.com/song/f0e3e97...

Somewhere in the Turning

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

100 1

Happy to get a celebration of logistics and infrastructure from a Newark sunset photo. suno.com/song/5bf2505...

Every Crossing

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

100 1

That my home area Hagalund produced rap was however entirely expected. suno.com/song/2024d8f...

Above the Estate

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

100 1

An epitaxy machine at MIT produced some nice techno. suno.com/s/4aIh0hFPO9...

Laboratory Tick

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

100 1

Suno can make songs based on pictures. OK, how weird pictures? Mathematics memes, surely not? Oh... it recognized that this is a Gram-Schmidt process! suno.com/song/312316a...

Gram-Schmidt Lemons

Anders Sandberg @arenamontanus.wsocial.eu · 06/10/2026

100 1

Adding an AI rendering of my scalesona to Suno produced a surprisingly good longtermist metal song. It is interesting to see how much background context sneaks in here. suno.com/song/02aa52d...

Five Eyes Watch

Centre for the Study of Existential Risk @cser.bsky.social · 01/10/2026

000 0

Reposted by Anders Sandberg

Applications are now open for our MPhil in Global Risk & Resilience! From AI & biosecurity to climate change, this 1 year course will equip you with the skills to understand how global catastrophic risk can be managed &and mitigated. Applications close: 8 Dec 2026

111 1

Anders Sandberg @arenamontanus.wsocial.eu · 21/09/2026

I suspect these add-ons will be equally popular.

000 0

Anders Sandberg @arenamontanus.wsocial.eu · 20/09/2026

One can make skeptical takes, like that this is roleplaying at being in pain and acting accordingly, but even in that case we get misaligned behavior. A good check would be to do the same experiment with hunger or some other somatic experience of less moral and motivational valence.

000 0

Anders Sandberg @arenamontanus.wsocial.eu · 20/09/2026

millenniumproblems.bio OK, we should bet on when they get solved (likely by AI). My own guess is "way sooner than would have seemed reasonable a few years ago" with a few exceptions (like, sadly, cryopreservation and limb regeneration) that are not amenable to lab automation.

millenniumproblems.bio

The Millennium Problems for Biology

A working catalogue of twelve open problems in biology — each a concrete, testable target for the field.

Anders Sandberg @arenamontanus.wsocial.eu · 20/09/2026

This goes back to an older paper of ours, where we noted that lab automation may make the risk chain change character. We were concerned with low-power human actors, but now we are getting AI actors (of increasing power too!) that can aid them. journals.sagepub.com/doi/full/10....

Figure 4 of (Sandberg & Nelson 2020), showing probability curves of successful bio-attack as function of agent power as the number of steps in the risk chain are reduced. The 7 step base case is a rough normal distribution around mid-power agents; as the steps are reduced more and more probability moves over to the weakest agents, until they eventually dominate.

011 1

Anders Sandberg @arenamontanus.wsocial.eu · 20/09/2026

My main take-away was that we should watch lab automation much more. The recent news of Anthropic setting up a wetlab is both interesting and mildly concerning. While it might be more about the scientific output side, I expect we will see a boom in lab automation. www.reuters.com/world/anthro...

reuters.com

EXCLUSIVE: Anthropic quietly sets up biology lab as it ramps AI drug program

At a time when fears of AI are gripping the public, the startup has built a wet lab — or place for physical experiments — in the San Francisco Bay Area, two people familiar with the matter said.

Anders Sandberg @arenamontanus.wsocial.eu · 20/09/2026

This discussion was part of a preparatory series for a new program based in Oxford Physics on the challenges posed by rapid technological progress, particularly in AI.

Anders Sandberg @arenamontanus.wsocial.eu · 20/09/2026

Participants: Nick Bostrom, Tom Milton, Rahul Nandkishore, Anders Sandberg, Shivaji Sondhi FRS, and K. VijayRaghavan FRS. The paper by Arc Institute and Stanford researchers that prompted our discussion: www.science.org/doi/10.1126/...

science.org

Generative design of bacteriophages with genome language models

Many important biological functions arise not from single genes but from complex interactions encoded by entire genomes. We report the first generative design of complete bacteriophage genomes using g...

Anders Sandberg @arenamontanus.wsocial.eu · 20/09/2026

Under the auspices of Oxford Physics, we held a discussion on the implications of recent work using genome language models to design novel, viable bacteriophages—viruses that infect bacteria. drive.google.com/file/d/18Szy...

drive.google.com

AI_Enabled_Biological_Design_Public_Summary.pdf

Anders Sandberg @arenamontanus.wsocial.eu · 17/09/2026

Jag dyker upp på "En AI till kaffet", snackandes AI, existentiell risk, medvetande, Fermiparadoxen, kryonik, framtidsforskning, frihet... youtu.be/XwxG3DeoxGg?... Spotify: open.spotify.com/episode/5pIW...

youtu.be

Anders Sandberg: Risken att AI utplånar mänskligheten är omkring 10% | En AI till Kaffet #104

YouTube video by En AI till Kaffet

000 0

Anders Sandberg @arenamontanus.wsocial.eu · 15/09/2026

Ethically I am not worried... yet. Fruit flies are not much of moral patients, and insofar a proper brain emulation would be equal, these are likely less like flies and more like generic neural networks. Might still have inner experience; hard to tell. But eventually we will have to confront ethics.

000 0

Anders Sandberg @arenamontanus.wsocial.eu · 15/09/2026

It also shows the power of AI- and github-supported open science: amateurs can take results from "serious science" and do new things nearly instantly. Much of it is nonsense, but the openness is important.

100 1

Anders Sandberg @arenamontanus.wsocial.eu · 15/09/2026

The "train fruit fly connectomes to do task X" community is fun. In one sense it is "take this topology and train it to do X" with no biological relevance. In another sense it is testing "does this topology have a very useful, pretty general structure for sensing/acting?" gizmodo.com/google-mappe...

gizmodo.com

Google Mapped a Fruit Fly’s Brain. Now It’s Playing Doom and Super Mario 64

Researchers published a complete map of a male fruit fly’s nervous system, and programmers are already making it play video games.

100 1

Reposted by Anders Sandberg

EBRAINS @ebrains.bsky.social · 15/09/2026

🧠 Explore whole-brain modelling at the EBRAINS Users Days in Heidelberg. On 7 October, Spase Petkoski will lead a hands-on TVB tutorial on setting up and running whole-brain simulations through research use cases. Programme and registration 👇 flagship.kip.uni-heidelberg.de/jss/HBPm?m=S...

011 1

Reposted by Anders Sandberg

jenny tightpants @jtp.bsky.social · 15/03/2026

transhumanism that works is just called medicine

854241 542

Anders Sandberg @arenamontanus.wsocial.eu · 15/09/2026

And given that the actually demonstrated risks have shown the ability to scale up relatively readily, one has even more reason to be cautious. Especially since one goal is to create a chain reaction of capabilities improvement, which would scale up scaling.

Anders Sandberg @arenamontanus.wsocial.eu · 15/09/2026

This was delightful. A good demonstration of how one can use generative art together with traditional media. youtu.be/WmdZ04y0ufI?...

youtu.be

Verdelis & Mycos - award winning sci-fi animation

YouTube video by Little Martians

000 0

Anders Sandberg @arenamontanus.wsocial.eu · 15/09/2026

We are perhaps approaching the moment when Washington DC will finally realize that Silicon Valley actually means what they say, rather than doing political and commercial signaling.

000 0

Anders Sandberg @arenamontanus.wsocial.eu · 15/09/2026

This is a very thoughtful essay from Daniel Selsam (OpenAI capabilities researcher), seeing firsthand the growing problems we have in making AI safe. Also, it might help outsiders understand that uncontrolled superintelligence is something insiders take seriously. docs.google.com/document/d/e...

docs.google.com

Personal Statement on AI Risk

100 1

Anders Sandberg @arenamontanus.wsocial.eu · 13/09/2026

Well, you can participate on several and cross-post your posts. I do that, and the different responses are often illuminating.

Anders Sandberg @arenamontanus.wsocial.eu · 13/09/2026

Is this sad? A bit. But I think the point of understanding is understanding itself, not necessarily the struggle. If we can get to (1) or (2), there will be understanding. Wir müssen verstehen – wir werden verstehen.

Anders Sandberg @arenamontanus.wsocial.eu · 13/09/2026

When (2) happens human mathematics is over, in a sense. It might well be that humans retain the key taste for what makes problems interesting beyond that, but I have doubts this will remain indefinitely. Eventually it will be like chess.

Anders Sandberg @arenamontanus.wsocial.eu · 13/09/2026

But I think (1) will require a lot of hard work because we do not have a method of doing it yet. Experimental mathematics may point vaguely in the right direction. Also, the motivation problem needs to be solved.

Anders Sandberg @arenamontanus.wsocial.eu · 13/09/2026

I am optimistic here: I think (1) humans and AI may well figure out deep and interesting things from the opaque solutions, it is just a different form of science. (2) AI will likely become good at the compression/explanation game eventually.

Anders Sandberg @arenamontanus.wsocial.eu · 13/09/2026

The fear is that we now will see solutions to every named problem without any good theory about why that solution. There is nothing to learn from the solutions.