iiinova — where your digital self keeps evolving

Daniel Asarnow @dabiophysicist.bsky.social · 7h

It even prevents *recurrences* in ppl who've had (HPV related) cervical lesions removed, so yes! It's probably not just about new strains bc those studies only used Gardasil-2 and -4. It could be vaccine derived immunity is just better somehow (more focused, or stronger bc of adjuvanting).

052

Daniel Asarnow @dabiophysicist.bsky.social · 17h

It's true "heat vision" as well, via the ion channel TRPA1, which we all experience as the Wasabi (/mustard/allicin/&c) receptor. Interestingly, the human one is also heat sensitive, but not to the extent of the TRPV channels we actually use for thermosensation.

010

Daniel Asarnow @dabiophysicist.bsky.social · 17h

Before attributing anything there to malice, remember that they also built an entire terminal without ever modeling aircraft parked at or taxiing to the gates...

000

Daniel Asarnow @dabiophysicist.bsky.social · 04/10/2026

That's what authors & interlocutor mean by a "coherent pain representation" - as you say it isn't (and can't be) denotation only. IMO Ant is quite a bit more circumspect there than those preprint authors about what that means ("strange foundational notions").

100

Daniel Asarnow @dabiophysicist.bsky.social · 03/10/2026

I do unironically hate the desktop metaphor and disable it if I can though

000

Daniel Asarnow @dabiophysicist.bsky.social · 03/10/2026

A friend sent me an interpretation of one of those just this morning, which I think is probably right.

Screenshot of a text message.

Friend:
https://arxiv.org/html/2609.16247v1

Doesn't this just mean that there exists a supported vector in English that makes everything sound bad? Idk stuff like this strains my ability to engage with the actual argument they're making because it seems to have such strange foundational notions

Paper:
"Summary of our findings.
We found a direction in the activation space that correlates with pain in all 25 models we tested. The signal is nearly orthogonal to fear and negative emotion and appears to be learned cheaply during pre-training. When injected into the residual stream, it produces the same ladder of distress in every model, regardless of size or training regime. This is evidence that models have coherent pain representations, as captured by our diverse sets of examples."

Friend:
English. English has a coherent pain representation

160

Daniel Asarnow @dabiophysicist.bsky.social · 03/10/2026

Apparently they all do! I'd note it's not uncommon in general to have multiple vice-presidents (or presidents) - I don't think the lower case "looks weird" at all.

000

Daniel Asarnow @dabiophysicist.bsky.social · 02/10/2026

maybe it has to do with deploying LED arrays during that time? i did find one paper suggesting it might take ~15 years for LED life cycle costs to be lower than sodium, and another noting that costs depend a lot on who operates the lights, with cities saving more than utilities after upgrades.

180

Daniel Asarnow @dabiophysicist.bsky.social · 02/10/2026

Strong memory of one RA saying to another, "it's called a TERMINAL bleed"

120

Daniel Asarnow @dabiophysicist.bsky.social · 01/10/2026

I mean that imo "extrapolation" is more apt, fits both technical and English definitions (the post 20c examples in the OED are kinda fun)

https://www.oed.com/dictionary/extrapolation_n?tl=true

1903

The necessity for having recourse to the risky expedient of ‘extrapolation’—that is, of applying unrestrictedly to the unknown, rules gathered from observation over a comparatively narrow area.

A. M. Clerke, Problems in Astrophysics 9
1922

By extrapolation of the curve of our knowledge we can reconstruct within the range of conceivability, if not of probability, the course of process.

J. Y. Simpson, Man & Attainment of Immortality i. 16
1926

Dr. Hubble strongly combats the contention that all non-galactic nebulae are spirals, which he characterizes as a ‘daring extrapolation’, ‘not justified by our present knowledge of nebular forms’.

H. C. Macpherson, Modern Astronomy 135
1965

One of the chief concerns of all scientific theories..is to sanction extrapolation..and the expression of the distant and unknown in terms of what is familiar.

J. D. North, Measure of Universe xiv. 297

020

Daniel Asarnow @dabiophysicist.bsky.social · 01/10/2026

It may otherwise occupy something that doesn't aid your solution

010

Daniel Asarnow @dabiophysicist.bsky.social · 01/10/2026

If it's reasoning, who's to say it's not during training?

110

Daniel Asarnow @dabiophysicist.bsky.social · 01/10/2026

It would have to be extrapolation as well, no?

110

Daniel Asarnow @dabiophysicist.bsky.social · 30/09/2026

I took the hard math classes because they were objectively much easier. That's not a dumb joke about being bored or whatever, those courses were just built to convey understanding instead of "weeding out"

010

Daniel Asarnow @dabiophysicist.bsky.social · 29/09/2026

Yeah true. It did turn out there was a lot farther to fall though.

010

Daniel Asarnow @dabiophysicist.bsky.social · 29/09/2026

The tools you can just go download are like that, but a lot of people must only be using it via stuff like Vertex and O365.

020

Daniel Asarnow @dabiophysicist.bsky.social · 29/09/2026

You can select "Code" in the desktop app. No idea if the behavior is different from text mode, but imo both have top-down UX. Something like Pi really molds itself to the user but I don't think most people want that (me either, I want it to pretend to ask permission before hacking the Gibson)

000

Daniel Asarnow @dabiophysicist.bsky.social · 29/09/2026

Trivial example, "turn on the light," light turns on. You'd only read the paragraph about why it's probably the light in the room you're in, and how it makes sense to turn on lights at night, if you're curious why it didn't work.

010

Daniel Asarnow @dabiophysicist.bsky.social · 29/09/2026

When it's acting as an interface the "output" doesn't really matter, only the "result," which could be self-validating.

110

Daniel Asarnow @dabiophysicist.bsky.social · 29/09/2026

Broken Earth flashback, very spooky indeed

010

Daniel Asarnow @dabiophysicist.bsky.social · 28/09/2026

I think there are older and perhaps less fraught examples too, like the implicit workload reduction from sensor fusion in the F-35 or going way back the Kommandogerät of the FW-190, which put ~five manual controls on one throttle lever.

010

Daniel Asarnow @dabiophysicist.bsky.social · 28/09/2026

I think the DVI can be used to switch MFD pages, change radar modes, etc. Apparently in Soviet doctrine GCI would direct/talk pilots through that stuff. I agree w/ you but I think those developments *increase* demand to ease workload wrt complex physical inputs like the MFD buttons and TDC.

110

Daniel Asarnow @dabiophysicist.bsky.social · 28/09/2026

There are also less stringent applications! I wondered if it's possible to train specialized models with smaller output spaces that are faster and more robust...I think the expense of training still favors general models, but conceptually it seems like what they're trying to do with Jev.

110

Daniel Asarnow @dabiophysicist.bsky.social · 28/09/2026

I learned the EF2000 had some form of DVI a long time ago, from Wikipedia, but after LLMs brought true natural language interfaces into view I tried to dig into what pilots actually think about it. Great doc about natural language input in general imo.

...there had become a tendency for them to forget the details of commands to be given. This human deficiency was noticed mainly during flight demonstrations or during periods of high workload. On one occasion a pilot asked the author over the aircraft intercom, "What do I have to say to Go Direct to Whiskey India Tango?" The author replied, "Why don't you say, Go Direct Whiskey India Tango?" Even the pilot's question had not reminded him of what was required. In the short term, this problem can be addressed to 3ome extent by automatic prompting. A longer term solution would be to use recognisers with more intelligence that could interpret what the pilot meant.

From https://dn760101.eu.archive.org/0/items/DTIC_ADA230240/DTIC_ADA230240.pdf

120

Daniel Asarnow @dabiophysicist.bsky.social · 28/09/2026

However one will command the radar with Jev, or whatever similar thing that solves "how do I say go direct to Whiskey India Tango"

120

Daniel Asarnow @dabiophysicist.bsky.social · 28/09/2026

Their own labs like Calico or Verily?

010

Daniel Asarnow @dabiophysicist.bsky.social · 28/09/2026

They aren't competing against "Claude." They're competing against some of the usual suspects, but who happen to have a large source of new funding without all the (good and bad) encumbrances of regular academic practice. *That* is the real source of angst and not unique here.

120

Daniel Asarnow @dabiophysicist.bsky.social · 28/09/2026

You said elsewhere that it's not a coincidence someone's agent reiterated their argument. Likewise it's not a coincidence when someone who has previously discovered repetitive genetic elements, in the most famous repetitive genetic element lab in the world, runs an agent that discovers one too.

110

Daniel Asarnow @dabiophysicist.bsky.social · 28/09/2026

I don't think they were competing with *Claude* (and doubt they truly see it that way either). The folks on the preprint have a lot of experience, and funding. Unfortunately that research brief doesn't seem to have been shared.

000

Daniel Asarnow @dabiophysicist.bsky.social · 28/09/2026

Personal subs and the Claude Science access for individual PIs, paid by grant (procard) most common. Some orgs like Hughes make an enterprise account available. Uni-linked Gemini is usually for students, requires sending Google proof of enrollment.

020

Reposted by Daniel Asarnow

Stuart Turville @turville.bsky.social · 25/09/2026

SARS-CoV-2 phenotypic trajectory and the Omicron Paradox. Discussion of the following pre-print and the complex intertwined pathways that have dictated phenotypic outcomes for SARS-CoV-2. This can be otherwise termed the pandemic to endemic trajectory: www.biorxiv.org/content/10.6...

biorxiv.org

Proteolytic control of the SARS-CoV-2 furin cleavage site defines the phenotypic evolution from the pandemic to endemic state

The successive emergence of SARS-CoV-2 variants with altered tissue tropism has progressively decoupled transmissibility from lower respiratory tract pathogenicity. Omicron lineages transmit exception...

15119

Daniel Asarnow @dabiophysicist.bsky.social · 26/09/2026

I'm guessing his interview was never a 45+15 talk followed by 4-7 hours of 1x1s and possibly dinner

110

Daniel Asarnow @dabiophysicist.bsky.social · 25/09/2026

You'd die in an accident anyway, I don't think it even takes that long

000

Daniel Asarnow @dabiophysicist.bsky.social · 25/09/2026

You die in an accident eventually

000

Daniel Asarnow @dabiophysicist.bsky.social · 25/09/2026

Lol, however (PSA) it's a live vaccine that can still reactivate. A lot safer than natural infection but you still need a shingles vaccine!

000

Daniel Asarnow @dabiophysicist.bsky.social · 25/09/2026

People vaccinated for chickenpox still need a shingles vaccine!

000

Daniel Asarnow @dabiophysicist.bsky.social · 22/09/2026

Have you seen Google Scholar Labs?

000

Daniel Asarnow @dabiophysicist.bsky.social · 21/09/2026

It's also possible Dennett meant "buffalo" there, which would be a different matter. (I suspect only a handful of well-kept individuals might know they are "buffalo")

010

Daniel Asarnow @dabiophysicist.bsky.social · 21/09/2026

Agreed, but it's an easy save. If the units are small and build up general and specific capabilities, then we'd expect a continuum with some large gaps.

100

Daniel Asarnow @dabiophysicist.bsky.social · 21/09/2026

Submarines sail, even though they don't have one

050

Daniel Asarnow @dabiophysicist.bsky.social · 21/09/2026

came up in a presidential primary debate as I recall

020

Daniel Asarnow @dabiophysicist.bsky.social · 21/09/2026

I wonder if the unaccountability of others is not as significant as the analogy to the self, though

000