pmayrgundter

Do you know what the objective limits are with current mirrors? They just say that nanometer objectives are not possible with current mirrors, but no measure of SOTA

Good.

But it's only a few days with a million ISP from fusion.. let's raise the bar ;)

NASA/JPL predicted a major global warming anomaly arriving this past year due to the Tonga eruption of Jan 2022.

They gave a forcing estimate of +0.15 Wm−2, equivalent to half a decade of global human CO2 forcing, arriving mainly at 18-24 months after Jan 2022 (ie the past year) and lasting for 5-10 years

"the HT-HH eruption was unusual in that it injected extremely large amounts of H2O [into the stratosphere]. Preliminary climate model simulations suggest an effective radiative forcing at the tropopause of +0.15 Wm−2 due to the stratospheric H2O enhancement. For comparison, the radiative forcing increase due to the CO2 growth from 1996 to 2005 was about +0.26 Wm−2.

"The excess H2O could arrive in northern and southern midlatitudes in ∼18 and ∼24 months, respectively, over a broad domain in the upper stratosphere... The timescale for complete dissipation of the plume may be 5–10 years

https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/202...

Am I missing it? 3x faster than an Xbox 10 from 2020 sounds "just" like Moore's law growth

Cool! Thanks Apple!

I guess this is to be expected given your market position and billions in the bank?

Thanks so much for doing it. We'll definitely buy all your

Well, they can immediately charge a fee (personal experience) and also it may be applicable for other matters in-flight.. let's say you're applying for a visa but have overstayed the current one. If they don't check until re-entry, they wouldn't catch this

Nice to hear the octave relation used!

"blue above visible blue" is a good name.. hmm, a little web tool to name these would be neat ;)

Howdy sir! Great memory and fun times :)

Dataset was something from CAIDA, like this: https://www.caida.org/catalog/datasets/ipv4_prefix_probing_d...

IIRC we used the LGL algorithm (https://lgl.sourceforge.net/) while pinning any nodes we could get geolocations for, giving a nice hybrid geo/topo layout

I don't remember exactly how we got the geolocations, but often network routers have 3-letter airport codes in their DNS names, so maybe that? We may also have had a lookup table in el googz somewhere

Definitely a project whose time should again come! ;)

Hmm, how about watermarking the photo with some function of the OTP? (like bit steganography; invisible to normal human use)

Seems to preserve easy checking by device manufacturer, but not immediately clear to me how to replace the sensor in a way that gets a signed image.

And if it seems like a bit of trouble, I think it's still worth consideration by the manufacturer, as a feature for their images and support of a community of users who want to produce images that can be so validated

Ah no problem. Just a sketch..

Looking at C2PA, I'd now use "claim" to describe those tags. They're short and human readable but not self-contained. They'd have to be authenticated with tool signing to be very useful. As it also emphasizes, it's about establishing trust for a domain, eg can Samsung photo signing be trusted?

Eg for a normal photo, your device could sign it with.. and OTP coordinated with the manufacturer? Then include the signing metadata in the EXIF tags of the image. To validate, you send the image+signed_claims to manufacturer and they know the OTPs (?) and can verify the signing

Doesn't make it unforgeable, bc you could intercept the hardware otp/signing protocol.. but would raise the cost quite a bit. Maybe with frequent key resets you can raise this further

Something like that sounding more useful?

Thanks for the ref! Checking it out

I agree with the general idea of tagging content to help classify.

I'd given this some thought via MIME and ended up with a kind of BioNFT.. so named bc it uses NFTs piecewise, but tracking the creation events and agents types (bio, ai, etc) as part of the content lifecycle

https://twitter.com/PMayrgundter/status/1638016474483683328

Highlight..

What if:

  - devices sign source creations with a biosignature

  - editing tools sign input

  - media types include that, effectively saying:

    ai_edited(human_created(photo))

and do this under the experimental namespace in MIME:

  image/x.bio(pablo@example.com/photo123).html

  image/x.adobe.photoai(http://x.bio(pablo@example.com/photo123)).html

The [Study] has a better framing in its title "JWST Observations Reject Unrecognized Crowding of Cepheid Photometry as an Explanation for the Hubble Tension at 8σ Confidence".

It rejects the hypothesis that there was a systematic observational problem in observing Cepheid variables (CVs), which are in turn used to estimate distances to Type Ia Supernova (SNs), towards a long-term goal the Study concludes with of "Tying all of these together by observing large samples in common can lead to the calibration of ∼100 [SNs] and a <1% local measurement of [The Hubble Constant, H0], a landmark in our quest to understand the expansion of the Universe."

Notably the paper doesn't provide a new estimate of H0, but it does strengthen the case for CV/SN being at odds with other methods of estimating H0, a problem called the Hubble Tension.

JWST was built primarily to extend the sensitivity range for infrared observations, so we can see fainter sources from further away, or near sources with greater resolution. This study is about the latter.. the study of CVs and SNs in nearby galaxies.

"the significantly greater resolution of JWST over [Hubble Space Telescope] has greatly reduced—in practical terms, almost eliminated—the main source of noise in [Near-Infrared] photometry of [CVs] observed in the hosts of nearby [SNs]. The resolution of JWST provides the ability to cleanly separate these vital standard candles from surrounding photometric "chaff."

CVs and SNs are "standard candles", rungs on the Cosmic Distance Ladder[CDL], the framework we use to compare and contrast different astronomical distance measurements. The term "standard candle" is used for a physical process we think we understand well enough to use its appearance at astronomical distance to infer other properties of its observation, e.g. the candle's color shift towards red the further away it appears to be. ("appears" since we can't directly measure actual distances, but observe that galaxies get smaller/fainter/redder together)

Cepheids are a relatively common kind of star that pulsates regularly during its lifetime, while SuperNovas are much rarer one-time very bright events. SNs are really useful to see them far away bc how bright they are, but since there's so few of them, we calibrate nearby SN distances based on the many CVs in the host galaxy of the SN.

In all, this study starts by looking at CVs in NGC 4258 at 23 Million lightyears away, and then looks for photometric crowding of CVs at successively further steps away in NGC 5643 (41 Mly), NGC 1559 (48 Mly), NGC (1448 56 Mly) and NGC 5468 (140 Mly), but don't find evidence of crowding to account for apparent brightness/closeness of the CVs, so rejects that idea with a high confidence. Those galaxies are actually as far away as they appear to be if CVs are good standard candles.

These are at the nearer end of the CDL.. 140 Mly vs the observable Universe is thought to be at least 13 billion light years radius. But if it in turn makes us more confident about SNs, those go out to a current max of 16Bly[FarthestSN].

[Study] https://iopscience.iop.org/article/10.3847/2041-8213/ad1ddd

[CDL] https://en.wikipedia.org/wiki/Cosmic_distance_ladder

[HT] https://xkcd.com/2516/

[FarthestSN] https://en.wikipedia.org/wiki/SN_UDS10Wil

Thanks!

Is there an inverse to this, more on the publisher side? I find myself wanting to cross-post more as social changes, and would like to be more systematic with my OSS projects and company work as well. Seems like campaign mgmt SaaS, but I never feel inclined to go explore that market and your site feels better somehow.

e.g. I write a short summary for Twitter, slightly longer for mstdn, longer still for FB and full length for medium or a markdown readme. If I have media, include where possible.. incl primary screenshot/social preview selection, or multi-screenshot. Or video. Email support++

I see the value mainly as a time save, but also quality checker eg for staging w/previews, so the simplest version of this could be a static page that just help prepare the content.

It gets me thinking about managing/remembering logins, aggregating stats and responding to comments, etc.. I suppose CORS would kill this for the more proprietary channels. Should be APIs for the good ones.

And throw some AI in for good measure: during composition, let me start with what I have and fill out drafts for the rest. For responding to comments, maybe autosuggest short responses for common FAQs, or help me build FAQs as I go.

Sorry, don't mean to hijack.. haven't seen this being handled well.

"To avoid data collection from Soofa signs, they would either need to turn off their phone’s wifi detection and Bluetooth or avoid the signs as a whole."

These should be defaults.

Android has options to WiFi scanning off. Does OSX?

20 years staring at it and no, i don't. i usually have to work it out from context. i think if you have vi or ed sensory organs it might work better for ya. which.. i still chuckle that vi is the visual editor, bc ed lol

If you do an accounting of all the organ functions, and then ask if the cell has this function independently, nothing is left out... but only if you allow that intelligence arises from the cells.

So I believe intelligence arises from the cells and is an essential function of life, not only an emergent phenomena. The organs serve as division of labor amongst the cells in community for what they are already originally capable of themselves.

More musings in this direction https://sites.google.com/site/pablomayrgundter/mind

"Do you know how to read @@ -5,6 +5,7 @@ syntax? Difftastic shows the actual line numbers from your files, both before and after."

Preach!

Just dropped it in and did a git diff.. works like a charm!

Alexander Unziker. Also Eric Lerner, Pierre Robitaille

Could Barnard's Star Kill Planet Nine?

https://www.listennotes.com/podcasts/the-demystifysci/great-...

I wonder if an energy and work metric could be derived for gradient descent. This might be useful for a more rigorous approach to hyperparameter development, and maybe for characterizing the data being learned. We say that some datasets are harder to learn, or measure difficulty by the overall compute needed to hit a quality benchmark. Something more essential would be a step forward.

Like in ANN backprop, the gradient descent algorithm can use a momentum to overcome getting stuck in local minima. This was heuristically physical when I learned it.. perhaps it's been developed since. Maybe only allowing a "real" energy to the momentum would then align it with an ability to do work calculation. Might also help with ensemble/monte carlo methods, to maintain an energy account across the ensemble.

Actually, they say, first page

"Increasing CO2 has driven global greening: over the last 40 years, half of the world’s vegetated area has undergone greening, of which 70% is attributed to elevated CO2 (Zhu et al. 2016)"

Related study that blew my mind.. DOE & USDA funded a 2021 NBER study of CO₂ fertilization on US agriculture:

"we apply the CO₂ fertilization effect... backwards to 1940, and, assuming no other limiting factors, find that CO₂ was the dominant driver of yield growth

https://www.nber.org/system/files/working_papers/w29320/w293...

Yeah, think you're right.. was reading the intro to the paper wrong. Just saw the acko visualization and was reminded your point

https://acko.net/blog/how-to-fold-a-julia-fractal/

Cheers!

1) agreed. it's exciting seeing the same basic architecture broadly applied

2) not sure what you mean by "operate in discrete space"

I'd emphasize the potential similarity to biological recurrence. Deep ANNs don't need to have this explicitly (tho e.g. LSTM has explicit recurrence), but it is known that recurrent NNs can be emulated by unrolling, in a process similar to function currying. In this mode, a learned network would learn to recognize certain inputs and carry them across to other parts of the network that can be copies of the originator, thus achieving functional equivalence to self feedback, or neighbor feedback. It takes a lot of layers and nodes in theory, but ofc modern nets are getting very big.

Per some offline discussion, I'll note that while this paper is about the structure of hyperparams, it also starts with the analogy between Mandelbrot & Julia sets, Julia as the hyperparam space for Mandelbrot, the param space

Well, they both also have similar fractal dimensions. Mandlebrot's Hausdorff dimension is 2, Julia is 1-2.

I won't argue it here but just suggest that this is an important complexity relationship and that the neural net being fit may also have similar fractal complexity, and that the distinction between param and hyper-param in this sense may be somewhat a red-herring

One of Wofram's comments is that there appears to be much more internal structure in language semantics that we had expected, contra-Chomsky.

We also know the brain, cortex esp, is highly recurrent, so it should be primed for creating fractals and chaotic mixing.

So maybe the hidden structure is the set of neural hyperparams needed to put a given cluster of neurons into fractal/chaotic oscillations like this. Seems potentially more useful too.. way more information content than a configuration that yields a fast convergence to a fixed point.

Perhaps this is what learning deep NNs is doing: producing conditions where the substrate is at the tipping point, to get to a high-information generation condition, and then shaping this to fit the target system as well as it can with so many free parameters.

That suggests that using iterative generators that are somehow closer to the dynamics of real neurons would be more efficient for AI: it'd be easier to drive them to similar feedback conditions and patterns

Like matching resonators in any physical system

Yeah, saw that too, but it's quoted at $4-5B and 7-10 years

How much does a chip fab cost? Internet says $10-20B [cost]

The last paragraph says they "would need about a dozen new fabs to be built and in full volume production by 2030"

max out 12*20 = $240B, but they say total cost $500B. Supply chain really on par with fab cost? Hm.

[cost] - ChatG estimates $10-20 for SOTA on par with TSMC

- https://www.quora.com/How-much-does-it-cost-to-build-a-semic...

- transitively sourced to a "third-party analysis by IBS" https://www.granitefirm.com/blog/us/2023/04/29/cost-of-chip-...

I've been having problems with my Mac M1 laptop display (4 years old), and this kicked them back in.

But to be clear: display was already having issues and i have a service appointment to get it replaced