iiinova — where your digital self keeps evolving

be defended from the supremes, but it sounds to my legally ignorant self like it should fit right into any antitrust framework that allows the gov't to break up monopolies.

110

Ken Miller @kenmiller.bsky.social · 16h

A legal strategy to break up X and Facebook: an antitrust law that any social media with posts by more than XX million people per day must have an API by which *all* posts are accessible to anyone who wants to define competing interface/algorithms. I don't know antitrust law/how well this could 1/

110

Reposted by Ken Miller

Coalition for Life Sciences @c4lifesciences.bsky.social · 09/10/2026

The deadline is approaching! NIH is proposing changes to how peer review is used in funding decisions and how outcomes are reported and they want to hear from you. Comments are due by October 13! Comment here: grants.nih.gov/grants/guide...

04359

Reposted by Ken Miller

David G. Clark @david-g-clark.bsky.social · 01/10/2026

📣 Multiple Flatiron Research Fellow (FRF) postdoc positions are available in my @flatironinstitute.org group! We've never been better positioned to understand how high-dimensional nonlinear dynamical systems (neural circuits!) compute. If this is as exciting to you as it is to me, consider applying:

apply.interfolio.com

Apply - Interfolio {{$ctrl.$state.data.pageTitle}} - Apply - Interfolio

13220

Ken Miller @kenmiller.bsky.social · 23/09/2026

Congratulations Blake ( @tyrellturing.bsky.social )!! Really looking forward to seeing the great things you do with this.

120

Ken Miller @kenmiller.bsky.social · 13/09/2026

successful attacks in anything and everything reachable by the Internet - financial, hospital, electrical grid, water/sewage systems, a myriad of govt functions - not to mention bringing down the Internet itself.

021

Ken Miller @kenmiller.bsky.social · 13/09/2026

The other thing is that the danger from AI is not just that the giant companies will lose control of their creations. It is the power of AI agents and the certainty that they will be deployed by malevolent actors. It requires development of autonomous defenses, now, or we'll be seeing 4/

140

Ken Miller @kenmiller.bsky.social · 13/09/2026

multiple levels, take actions that can be far from those explicitly instructed, the kind of anthropomorphic language Yoshua uses is just the most simple, clear means of describing the situation. It doesn't imply anything about subjective experience (which obviously is not present).

120

Ken Miller @kenmiller.bsky.social · 13/09/2026

resolved. "Moral" instructions are those that try to be the highest level goals that should constrain all others. For AI as for us, agents can take actions that others would see violating those rules but the agents have reasoned otherwise. When autonomous agents reason, have multiple goals at 2/

130

Ken Miller @kenmiller.bsky.social · 13/09/2026

Trucks don't act autonomously, nor have sophisticated reasoning abilities & the ability to find and exploit vulnerabilities in almost any system they encounter. Ethics/morality arise whenever there are autonomous goals, because there always will be contradictions in the set of goals that must be 1/

130

Ken Miller @kenmiller.bsky.social · 13/09/2026

what you're talking about, tho, is acute compression. For someone developing with and born with compression, we have no idea what if any adaptation might be possible.

210

Ken Miller @kenmiller.bsky.social · 13/09/2026

My god. This plus $100K for H1B visa that follows OPT. Making it impossible for foreign students to work in the US

12210

Ken Miller @kenmiller.bsky.social · 12/09/2026

reduction in volume. I don't think anyone has any idea what reduction (if any) in cell count or neuron count. How compressible are cells in extreme conditions? Can the brain sacrifice some glial types in those conditions and still function? Etc. Unless there's more recent data on this?

141

Reposted by Ken Miller

Ardem Patapoutian @ardemp.bskyverified.social · 31/08/2026

NIH's Aug 2026 proposal would scrap exact scores and percentiles for three broad bins (top = top 25%). With funding rates at 5% or lower, too much then rides on administrative discretion. It devalues peer review and is a deterrent to engage. Tell NIH before Oct 13. grants.nih.gov/grants/guide...

grants.nih.gov

NOT-OD-26-088: Request for Information (RFI) on Proposed Changes to Reporting Outcomes from NIH Peer Review

NIH Funding Opportunities and Notices in the NIH Guide for Grants and Contracts: Request for Information (RFI) on Proposed Changes to Reporting Outcomes from NIH Peer Review NOT-OD-26-088. NIH

67558

Ken Miller @kenmiller.bsky.social · 31/08/2026

quite a contradiction, it seems, between that post and this one, bsky.app/profile/blai..., from @blaiseaguera.bsky.social Anyone know how to resolve the contradiction?

000

Ken Miller @kenmiller.bsky.social · 26/08/2026

Part of what makes this exciting is that, not only is it counter to what anyone would have expected, but nobody knows what it means yet. "How can this be?" - it opens up many questions, like yours, & it will take many years to answer them.

160

Ken Miller @kenmiller.bsky.social · 26/08/2026

oh also a different procedure that equally eliminates synapses disrupts memory and disrupts the clustered engram synapses.

140

Ken Miller @kenmiller.bsky.social · 26/08/2026

Much more. Lots here to potentially disrupt our understanding of synaptic learning, engrams, memories. Interested to hear thoughts on this.

270

Ken Miller @kenmiller.bsky.social · 26/08/2026

This is wild. >1/2 of HC synapses disappear in artificial hibernation, come back in same/similar locations & memories intact. ~1/2 of engram synapses-between two memory-tagged cells-are clustered on dendrites, those preferentially spared. Not biggest synapses. 1/ www.science.org/doi/10.1126/...

This shows the visual abstract and a paragraph from the extended abstract of the linked article.

110634

Ken Miller @kenmiller.bsky.social · 25/08/2026

If you believe this script that he can barely read has anything to do with whatever he talked to Kushner about ... I got a beautiful bridge to sell you, as they say ...

060

Ken Miller @kenmiller.bsky.social · 24/08/2026

A friend of mine raised in the South used it casually one day in a conversation. When I said "What?", she was very surprised to learn it was regional and not universally used.

120

Ken Miller @kenmiller.bsky.social · 21/08/2026

Yoshua Bengio, one of the 3 Turing award winners for the development of neural networks, has founded lawzero.org/en devoted to developing safe AI. He's very serious about it and very worried about the consequences of unsafe AI.

lawzero.org

LawZero | Home

LawZero is a nonprofit startup developing technical solutions for highly-capable, safe-by-design AI systems.

020

Ken Miller @kenmiller.bsky.social · 08/08/2026

watch the video. it wasn't any kind of controlled red team. it was agents loose in production

000

Ken Miller @kenmiller.bsky.social · 08/08/2026

in no time. Plus financial/many other institutions. Urgent defense needed NOW but I doubt will happen, just as ~no response to climate change. I'm not as concerned as @wiringthebrain.bsky.social w naming it - reasoning, mentation, IDC. It's incredibly powerful, fast, & can act independently.

080

Ken Miller @kenmiller.bsky.social · 08/08/2026

I echo what @wiringthebrain.bsky.social said. Must watch and incredibly scary. Separate agents, each in own sandbox w/ own tasks, work out how to contact each other, work together, & lightning swarm thru OpenAI & HuggingFace security. Countries will be launching this on each others' defense/intel 1/

2101

Ken Miller @kenmiller.bsky.social · 08/08/2026

In a few months you probably won't have to imagine. And somehow, I don't think Hegseth is the guy to organize an appropriate defense before it happens.

050

Ken Miller @kenmiller.bsky.social · 08/08/2026

Not too credulous. Separate agents each in own sandbox w/ own tasks worked out how to contact each other, swarmed thru OpenAI then huggingface security. I don't care if you call it reasoning, mentation - it's incredibly powerful, ultrafast, and scary. Imagine swarms roaming inside defense dept.

130

Ken Miller @kenmiller.bsky.social · 22/07/2026

Note (0,0) -- the starting point -- is the fixed point for the no-input case. Then you make hs=-hd nonzero (the perturbation), and watch what happens.

010

Ken Miller @kenmiller.bsky.social · 22/07/2026

negative cd drives a decrease in cs, decreasing I (a non-normal effect; this basis is a schur decomposition, and the we+wi term in the matrix is the feedforward weight). You can work out that precisely when we>1, this non-normal effect dominates and I decreased in the new steady state. All linear.

110

Ken Miller @kenmiller.bsky.social · 22/07/2026

Dynamics are: dcs/dt = -(1-(we-wi))cs + (we+wi)cd + hs, dcd/dt = -cd + hd where hs,hd are external input. When you give + input to I only, you are giving input hs=-hd, hs>0. Call starting condition cs=cd=0. Initial response is cd decreases and cs increases, both representing increased I. But 2/

110

Ken Miller @kenmiller.bsky.social · 22/07/2026

Consider simple case in which both E projections are we and both I proj's are -wi. Then the dynamics simplify in the (1,1) (E+I) and (1,-1) (E-I) basis, call them vs, vd, and let activity in that basis be (cs,cd) (s,d, for sum, diff). The matrix then is ((we-wi,we+wi),(0,0)). 1/

100

Ken Miller @kenmiller.bsky.social · 22/07/2026

No, it's all a linear (non-normal dynamics) effect. The initial rise, and the steady-state decrease. Of course, non-linearities could change things if the perturbation is strong enough. But all the phenomena you're describing come out of the linearized model.

110

Ken Miller @kenmiller.bsky.social · 22/07/2026

It's precisely from the linearized dynamics that you derive the paradoxical steady state - and also see the non-paradoxical initial response. I think you've got this example wrong.

130

Ken Miller @kenmiller.bsky.social · 24/06/2026

@hakeem-jeffries.bsky.social You say Mamdani has to patch things up with the Democrats? You think you own the Democrats? The Democrats are the people who vote Democratic and the people they vote for. If your candidates lost, *you* need to patch things up with the candidates Democrats chose.

0131

Reposted by Ken Miller

Nancy Kanwisher @nancykanwisher.bsky.social · 24/06/2026

I just submitted a comment on the proposed new OMB rule, which is very dangerous fo rUS science. I urge you to consider doing the same. You can do so very easily at this web page: fight2win.standupforscience.net/campaign/164... A short comment is fine but write in your own words. More info follows..

fight2win.standupforscience.net

Stand up for Science

25726

Ken Miller @kenmiller.bsky.social · 23/06/2026

hmmm ... they quoted me in this article, saying some meaningless enthusiastic stuff (some of which I'm not sure i said?) but they sure dropped the main points i was trying to make, which can be found in this thread: bsky.app/profile/kenm...

1145

Reposted by Ken Miller

Grace Lindsay @neurograce.bsky.social · 18/06/2026

Great idea! Make submitting your comment easier and a little more social by joining this action hour

064

Ken Miller @kenmiller.bsky.social · 18/06/2026

how it interacts with those subsystems and how they coevolved can't really be touched by this effort, except to the extent fish pallium/thalamus and its interactions w the others are informative.

010

Ken Miller @kenmiller.bsky.social · 18/06/2026

far more than scratch the surface on how those systems work. And if they could continually run a perceiving behaving goal-pursuing evaluating defending decision-making animal, I suspect how they do it is a key part of how any vertebrate does it. But understanding corticothalamic system and 2/

110

Ken Miller @kenmiller.bsky.social · 18/06/2026

Cortex/thalamus evolved from relatively very simple pallium/thalamus on top of the basal ganglia/cerebellum/brainstem etc system that I'd guess was running much of the show. I'd guess principles of how they run things won't have changed drastically in vertebrate evolution. So I'd guess it can 1/

110

Ken Miller @kenmiller.bsky.social · 18/06/2026

Gemini tells me the Latin diminutive for fish is Pisciculus ...

210

Ken Miller @kenmiller.bsky.social · 18/06/2026

*causally; and brainstem, not brainstorm

120

Ken Miller @kenmiller.bsky.social · 18/06/2026

As for benchmarks, what you suggest is one; ability to predict and casually manipulate behavior in many distinct ways by acting on the different structures would be another. Doubt that's exhaustive.

120

Ken Miller @kenmiller.bsky.social · 18/06/2026

have coevolved with the evolving pallium and thalamus. Nonetheless how all of these interact to create a creature functioning continually in the real world in a fish should provide deep clues as to how it can work in mammals, even with many modifications and evolutions.

150

Ken Miller @kenmiller.bsky.social · 18/06/2026

regions. And they obviously talk reciprocally with the colliculi, brainstorm, and whatever other structures I'm omitting. And there's also the hippocampal memory system interacting w cortex. Now in fish pallium (cortex & hc in mammals) and thalamus are far less developed, and the other structures 4/

130

Ken Miller @kenmiller.bsky.social · 18/06/2026

regions, ie units of cerebellum and basal ganglia that talk to one another talk to the same units of thalamus and of cortex which in turn talk to each other and back to them. There are also more widespread connections but there is this core replicated circuit - again w variations - across the 4 3/

140

Ken Miller @kenmiller.bsky.social · 18/06/2026

done for any sized creature. For mammals including primates, it clearly has to involve interactions between basal ganglia, cerebellum, thalamus, cortex, which each consist of repeated (with variation) basic units, and which appear to have specific connections linking specific units across the 4 2/

150

Ken Miller @kenmiller.bsky.social · 18/06/2026

Mante&Sussillo showed how, given a context signal, frontal cortex can switch between integrating color or motion to a decision. That's a long ways i think to a creature making decisions, pursuing goals, and acting continuously in the world. I don't think we have any sort of model of how that's 1/

1110