Jonas Hallgren
@jonashallgren.bsky.social
I build stuff I find interesting, mostly in Collective Intelligence and AI Safety.
I'm the director of eq-network.org where we also build stuff.
I'm the director of eq-network.org where we also build stuff.
Are there any major changes or is the general points approximately the same? (Trying to decide whether ai need to reread it or not)
February 10, 2026 at 5:19 PM
Are there any major changes or is the general points approximately the same? (Trying to decide whether ai need to reread it or not)
It's very much just in the writing down the ideas phase, I haven't decided what to do with it yet:
www.overleaf.com/read/bdjvstt...
www.overleaf.com/read/bdjvstt...
February 9, 2026 at 7:32 AM
It's very much just in the writing down the ideas phase, I haven't decided what to do with it yet:
www.overleaf.com/read/bdjvstt...
www.overleaf.com/read/bdjvstt...
understanding the vsm was a background project of mine for the last year so it might be fun to compare notes!
February 6, 2026 at 10:06 AM
understanding the vsm was a background project of mine for the last year so it might be fun to compare notes!
I wrote up an active inference interpretation of the VSM if you're into that, it kind of makes it make a bit more sense at least if you know RL Stuff?
I would uhhh need to work on the diagrams a bit before but happy to send it over if interesting.
I would uhhh need to work on the diagrams a bit before but happy to send it over if interesting.
February 6, 2026 at 10:06 AM
I wrote up an active inference interpretation of the VSM if you're into that, it kind of makes it make a bit more sense at least if you know RL Stuff?
I would uhhh need to work on the diagrams a bit before but happy to send it over if interesting.
I would uhhh need to work on the diagrams a bit before but happy to send it over if interesting.
building the infrastructure where a biologist and a physicist can collaborate without one of them eventually writing a subtweet about "rigor"
bridges not primitives: equilibria1.substack.com/p/bridges-no...
bridges not primitives: equilibria1.substack.com/p/bridges-no...
Bridges, Not Primitives
Why DeSci should stop searching for universal verification and start building compositional translations.
equilibria1.substack.com
February 3, 2026 at 10:57 AM
building the infrastructure where a biologist and a physicist can collaborate without one of them eventually writing a subtweet about "rigor"
bridges not primitives: equilibria1.substack.com/p/bridges-no...
bridges not primitives: equilibria1.substack.com/p/bridges-no...
when two different lenses on the same phenomenon converge? that's robust. verification from multiple directions.
when they don't converge? you've found exactly where something is missing.
either way: everyone stays in the room. nobody cries.
when they don't converge? you've found exactly where something is missing.
either way: everyone stays in the room. nobody cries.
February 3, 2026 at 10:57 AM
when two different lenses on the same phenomenon converge? that's robust. verification from multiple directions.
when they don't converge? you've found exactly where something is missing.
either way: everyone stays in the room. nobody cries.
when they don't converge? you've found exactly where something is missing.
either way: everyone stays in the room. nobody cries.
the solution isn't flattening everything into shared primitives
it's building BRIDGES between local verification structures
formalize each domain on its own terms. then find structure-preserving maps between them. category theory been doing this quietly for decades.
it's building BRIDGES between local verification structures
formalize each domain on its own terms. then find structure-preserving maps between them. category theory been doing this quietly for decades.
February 3, 2026 at 10:57 AM
the solution isn't flattening everything into shared primitives
it's building BRIDGES between local verification structures
formalize each domain on its own terms. then find structure-preserving maps between them. category theory been doing this quietly for decades.
it's building BRIDGES between local verification structures
formalize each domain on its own terms. then find structure-preserving maps between them. category theory been doing this quietly for decades.
the thing DeSci keeps getting wrong is that we're trying to build universal verification primitives. one protocol to rule them all.
but physics demands five-sigma because particle physics has that signal-to-noise ratio. psychology doesn't. different fields adapted to different causal grains.
but physics demands five-sigma because particle physics has that signal-to-noise ratio. psychology doesn't. different fields adapted to different causal grains.
February 3, 2026 at 10:57 AM
the thing DeSci keeps getting wrong is that we're trying to build universal verification primitives. one protocol to rule them all.
but physics demands five-sigma because particle physics has that signal-to-noise ratio. psychology doesn't. different fields adapted to different causal grains.
but physics demands five-sigma because particle physics has that signal-to-noise ratio. psychology doesn't. different fields adapted to different causal grains.
like you could explain salt using quantum chromodynamics, work through the schrödinger equations, trace the electron probability clouds...
or you could say "sodium has an extra electron, chlorine wants one, they kiss, now you have fries"
both are true. one gets you invited to parties.
or you could say "sodium has an extra electron, chlorine wants one, they kiss, now you have fries"
both are true. one gets you invited to parties.
February 3, 2026 at 10:57 AM
like you could explain salt using quantum chromodynamics, work through the schrödinger equations, trace the electron probability clouds...
or you could say "sodium has an extra electron, chlorine wants one, they kiss, now you have fries"
both are true. one gets you invited to parties.
or you could say "sodium has an extra electron, chlorine wants one, they kiss, now you have fries"
both are true. one gets you invited to parties.
Amazing! I was thinking of bringing some friends along onto semble but I felt I wanted to be able to subscribe to collections before I did that.
January 29, 2026 at 8:44 AM
Amazing! I was thinking of bringing some friends along onto semble but I felt I wanted to be able to subscribe to collections before I did that.
Why should I care about it? I don't know but I clearly should. It seems I've been outwizarded, I'm becoming the businessman I set out to destroy, oh why world, oh why. Spells should run free.
January 26, 2026 at 11:57 AM
Why should I care about it? I don't know but I clearly should. It seems I've been outwizarded, I'm becoming the businessman I set out to destroy, oh why world, oh why. Spells should run free.
Posts the following 50 page paper about non equilibrium physics and refuses to elaborate arxiv.org/pdf/2311.13749
arxiv.org
January 26, 2026 at 11:56 AM
Posts the following 50 page paper about non equilibrium physics and refuses to elaborate arxiv.org/pdf/2311.13749
The hypothesis is that once people write something they get over the threshold a bit and realise that it isn't as scary to write, basically CBT for getting cross the psych safety barrier of posting online
January 13, 2026 at 8:59 AM
The hypothesis is that once people write something they get over the threshold a bit and realise that it isn't as scary to write, basically CBT for getting cross the psych safety barrier of posting online
I think it is a function of average psychological safety and that it might just be about providing opportunities to answer something that is easy to answer? Experiment: Run a low-stakes post or similar asking something like "What is a weird cool paper you've read recently" or things like it
January 13, 2026 at 8:58 AM
I think it is a function of average psychological safety and that it might just be about providing opportunities to answer something that is easy to answer? Experiment: Run a low-stakes post or similar asking something like "What is a weird cool paper you've read recently" or things like it
Is this really optimal for your attention management? Just take a 10-15 min nap and then a cold shower, you can do it like 2-3 times per day without endocrine issues. The problem is energy not not knowing?
January 12, 2026 at 11:10 PM
Is this really optimal for your attention management? Just take a 10-15 min nap and then a cold shower, you can do it like 2-3 times per day without endocrine issues. The problem is energy not not knowing?