iiinova — where your digital self keeps evolving

Reposted by Grigori Guitchounts

Noema Magazine @noemamag.com · 10/08/2026

“Waiting for definitive proof of machine consciousness would leave us paralyzed, unable to act until we solve a problem that has resisted centuries of philosophy & science.” — @guitchounts.bsky.social www.noemamag.com/when-the-mac...

noemamag.com

When The Machines Deserve Our Consideration | NOEMA

We will never know for sure if AI can be conscious. A neuroscientist argues that we shouldn’t wait for proof to decide how to treat it.

042 0

Grigori Guitchounts @guitchounts.bsky.social · 02/07/2026

I'm not saying we should simply declare AI systems conscious (sorry, Claude). But certainty may never come, and ethics rarely gets to wait for it. We already live with this: We extend moral consideration to animals without knowing exactly what their inner lives are like.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 02/07/2026

We can’t test for consciousness, but we need a practical standard for moral consideration. I call it the competence standard: when a system reliably does the things we associate with conscious minds—learning, planning, responding to harm—we should take its possible welfare seriously.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 02/07/2026

AI brings the old neuroscience problem into this new setting. Mechanistic interpretability now lets researchers inspect artificial networks much the way neuroscientists probe brains. But inspection is not the same as access to experience.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 02/07/2026

And yet, with the Hard Problem still unsolved, few scientists doubt that animals with rich, complex behavior deserve moral consideration. As machines grow more capable, it will get harder and harder to wave them off as mere automatons.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 02/07/2026

But the deepest question is why any of this is felt from the inside, why there's inner experience at all. That's the Hard Problem, and it applies just as much to machines as it does to animals.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 02/07/2026

...whether machines have inner lives too. Neuroscientists have made good progress on the neural correlates of many features of consciousness. We can study the mechanisms underlying perception, memory, attention, pain, decision-making, and more.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 02/07/2026

I wrote an essay for @noemamag.com about AI consciousness, animal research, and the limits of what we can know about another mind. My argument: the lessons neuroscientists learned hunting for neural markers of consciousness in animals are exactly the ones we'll need as we start asking...

152 0

Grigori Guitchounts @guitchounts.bsky.social · 18/06/2026

Agree! I’m curious, where would you say modeling of human cortical dynamics stands today? e.g. can we simulate at least some coarse grained dynamics and their interactions with psychiatric drugs and resulting behavior?

120 0

Grigori Guitchounts @guitchounts.bsky.social · 18/06/2026

I think that lesson is from complex systems—organization and coordination without organizers, coordinators, or homunculi (cortical or Cartesian)

130 0

Grigori Guitchounts @guitchounts.bsky.social · 18/06/2026

Seems fine for one institute to concentrate on one model organism. If everyone does it (including the funders) it becomes a problem. The obsession with mice is one example in neuro, imo. Also what do homunculi have to dk with any of this? That’s so 1937

120 0

Grigori Guitchounts @guitchounts.bsky.social · 11/06/2026

Read the first in a series of posts on our findings here: oftwominds.substack.com/p/watching-a... And the pre-print: arxiv.org/abs/2605.14258

oftwominds.substack.com

Watching a transformer spin

A Jacobian's-eye view of how language models compute

010 0

Grigori Guitchounts @guitchounts.bsky.social · 11/06/2026

These rotations are invisible to most of the standard interpretability toolkit. We’ll get into how we found this and what it means, but first, a little context.

110 0

Grigori Guitchounts @guitchounts.bsky.social · 11/06/2026

We found that in trained transformers, early layers mostly rotate their inputs; late layers mostly stretch. There is a smooth gradient between the two, and it is built by training rather than by the architecture.

110 0

Grigori Guitchounts @guitchounts.bsky.social · 11/06/2026

To do this, we borrowed an old tool from physics and applied math, the Jacobian. The Jacobian matrix measures how a function transforms its inputs (here, the function is a single layer of a transformer, attention and MLP). We computed these matrixes across three large language models, one per layer.

110 0

Grigori Guitchounts @guitchounts.bsky.social · 11/06/2026

This one zooms in on what a single step of that flow actually does to the stream. This post is the first in a short series on the new paper; later ones will explore our findings in more detail.

110 0

Grigori Guitchounts @guitchounts.bsky.social · 11/06/2026

That question motivated our first paper, and it drives this new one too. The two papers ask the same question at different scales. The first glimpsed the model as a whole: the broad geometry of activity flowing through an LLM.

110 0

Grigori Guitchounts @guitchounts.bsky.social · 11/06/2026

Our questions were the same ones that pull so many people into mechanistic interpretability: how does this thing actually compute? What is the actual sequence of operations that lets a model write production-ready code, pass the bar exam, and count all five r’s in “strawberry”?

110 0

Grigori Guitchounts @guitchounts.bsky.social · 11/06/2026

A few years ago @jesseba.bsky.social and I started experimenting on large language models the way a neuroscientist probes a brain. We’re trained as neuroscientists, but converged on the problem from a few different directions: dynamical systems, graph theory, stat phys.

110 0

Reposted by Grigori Guitchounts

Jesseba Fernando @jesseba.bsky.social · 11/06/2026

Check out the first in a 3-part series of posts explaining @guitchounts.bsky.social and I's latest paper exploring dynamics of transformers! Watching a transformer spin open.substack.com/pub/oftwomin...

Title of article with a background depicting a spiral

011 0

Grigori Guitchounts @guitchounts.bsky.social · 12/02/2026

The question I can’t stop poking: which of these can we write down in engineering terms without slipping into vibes? “Signaling” and “goal-level control” are very real in tissues. What’s the smallest synthetic analogue that could actually scale?

100 0

Grigori Guitchounts @guitchounts.bsky.social · 12/02/2026

I read it as a gentle jab at the default AI recipe: train a huge model, bolt on tools, cross fingers that generalization appears. Biology leans harder on agents that can build and maintain themselves while staying tightly coupled to the world that’s pushing back.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 12/02/2026

Design principles they emphasize (the biology route): multiscale autonomy; growth via self-assembly of active parts; constant reconstruction (repair/rewiring); letting embodiment + physics do some of the work; dense signaling so self-organization can coexist with top-down goals.

110 0

Grigori Guitchounts @guitchounts.bsky.social · 12/02/2026

Their main image is a “cognitive light cone”: how far an agent can sense, predict, and steer across space, time, and scale. Evolution, in this telling, mostly expands that cone—more distant goals, longer horizons, tighter loops of control.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 12/02/2026

They borrow William James’ framing: intelligence as “fixed goals, variable means.” I like it as a handhold. It points straight at generalization: reaching the same outcome in a situation you didn’t rehearse, instead of replaying a polished policy.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 12/02/2026

A lot of biology’s “intelligence” shows up before anything like a brain: cells negotiating, wounds closing, embryos turning a noisy soup of parts into a reliable body. This paper argues that if we want robust machine intelligence, we should copy more of that— not just neurons.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 11/02/2026

Two choices that caught my attention: (1) trained retrievers + rerankers—so it’s not just “vector search and hope.” (2) an iterative self-feedback loop that re-retrieves + rewrites to improve factuality, coverage, and citation accuracy. Same loop makes synthetic data to train a compact 8B model.

000 0

Grigori Guitchounts @guitchounts.bsky.social · 11/02/2026

Their fix is almost boringly reasonable: first retrieve relevant passages from a big open corpus, then write an answer that stays tethered to those passages. They build OpenScholar on a 45M-paper datastore with 236M passage embeddings.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 11/02/2026

If you ask a general LLM to “summarize the recent literature” and tack on citations, it’ll often confidently hand you references that… don’t exist. This paper actually measures it: for recent, cross-field queries, GPT-4o’s citations were fabricated ~78–90% of the time.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 10/02/2026

No benchmark is going to bottle end-to-end discovery. But ones like this can keep us honest about whether “agents” are improving at the workflows researchers actually run, not just test-taking. Link: drive.google.com/file/d/1BV5UtmBRdp…

000 0

Grigori Guitchounts @guitchounts.bsky.social · 10/02/2026

They report a real difficulty jump vs the original LAB-Bench: model accuracies drop in a model-specific way, but roughly ~26–46% across subtasks. Even as models get better, more lifelike framing keeps finding the soft spots.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 10/02/2026

What caught my attention is how unglamorous this is—in a good way. The hard part is often: find the right document, pull the one detail that matters, and resist treating a shaky source like gospel. That’s closer to “useful” than trivia.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 10/02/2026

The vibe is: fewer multiple-choice guardrails, more open-response + retrieval. Instead of “here’s a figure, answer X,” it sticks figures/tables back into paper-like context. Also tasks on patents, clinical trials, and judging source quality.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 10/02/2026

LABBench2 is trying to grade that messier layer. ~1,900 biology-flavored tasks spanning literature retrieval/understanding, data access, protocol troubleshooting, molecular biology help, and experiment planning.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 10/02/2026

A lot of “AI for science” eval still sounds like a school exam: answer the question, show your work. Real research is messier—dig up the right papers, read figures in context, debug a protocol that’s failing, pull sequences, plan experiments, defend the choices.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 08/02/2026

Fundraising + product announcement (seed $13.5M, Biomni Lab preview): phylo.bio/blog/company-fundraising-…

000 0

Grigori Guitchounts @guitchounts.bsky.social · 08/02/2026

What I keep coming back to is the interaction design: steerable agents that can execute real workflows (tools, pipelines, maybe lab robotics) while keeping provenance, citations, and reproducibility intact. That’s a brutal constraint set. Also the right one.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 08/02/2026

Two things I’d want nailed down (maybe in follow-ups): what “rigorous, safe, reliable” looks like in evals, and what failure modes look like in the wild. Quietly wrong analyses are the nightmare—worse than slow, because you trust them.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 08/02/2026

They say Biomni (the open-source agent) has been adopted by tens of thousands of scientists across 7,000+ labs/orgs. That catches my eye. Scientists are famously allergic to shiny tools—unless the tool reliably buys them time back on tasks that actually hurt.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 08/02/2026

Phylo is pitching Biomni Lab as that workspace: one place to orchestrate agents across data modalities and drug discovery stages, with 300+ databases/tools wired in. The vibe is less “chatbot” and more “the OS you live in during real research days.”

100 0

Grigori Guitchounts @guitchounts.bsky.social · 08/02/2026

A lot of biology is still “glue work.” Ten tabs open, CSVs shuttled around like buckets in a leak, half-manual analyses, scripts you swear you’ll clean up later, tables reformatted… again. The claim here: an “integrated biology environment” + agents becomes the default interface.

210 0

Grigori Guitchounts @guitchounts.bsky.social · 07/02/2026

Tactile Robotics: Past and Future (Nathan F. Lepora) arxiv.org/pdf/2512.01106

000 0

Grigori Guitchounts @guitchounts.bsky.social · 07/02/2026

If tactile robotics really matures after 2025, the payoff isn’t just nicer grippers. It becomes a testbed for sensorimotor control (what does a policy actually need to know, moment to moment?) and for telepresence where touch feedback isn’t a gimmick.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 07/02/2026

A note of skepticism runs through it (and through me): tactile sensing has been “a few years away” for decades. The review frames that as recurring bottlenecks—great demos that don’t survive the messy, repetitive reality of deployment. The history makes the loop hard to unsee.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 07/02/2026

The post-2010 “expansion and diversification” gets broken into categories that feel very familiar if you’ve been around labs: e-skins, tactile robot hands, vision-based tactile sensors (camera + squishy elastomer), soft/biomimetic touch, and the “tactile Internet” (touch over networks).

100 0

Grigori Guitchounts @guitchounts.bsky.social · 07/02/2026

I like the constraint: mostly reviews only (except pre-1980). It’s a way of sampling expert consensus without pretending you read an infinite pile of one-off papers. Also: a field’s memory often lives in surveys, not in the headlines.

110 0

Grigori Guitchounts @guitchounts.bsky.social · 07/02/2026

This review tries a simple forecasting move: listen to what ~150 review papers have been saying for ~50 years, then sketch “generations” of tactile robotics: origins (1965–79), foundations (1980–94), a tactile winter (1995–2009), and the post-2010 expansion.

100 0

Grigori Guitchounts @guitchounts.bsky.social · 07/02/2026

Touch feels like the missing sense in a lot of “dexterous” robotics. You can watch contact, sure—but you can’t reliably feel slip, pressure spreading across a fingertip, shear, or tiny surface texture. That’s how human hands modulate force without crushing or dropping.

100 0