Jeff Kaufman

Pro AI Bots Scraping List Archives I'm on various mailing lists, and the archives are a trove of niche knowledge. A dance calling list I'm on is considering making archives subscriber-only, to keep AI bots from snarfing up this data. B...

I'm on various mailing lists, and the archives are a trove of niche knowledge. A dance calling list I'm on is considering making archives subscriber-only, to keep AI bots from snarfing up this data. But I think this harvesting is overall a good thing: I'd prefer future LLMs gave good dance advice.

Two Kinds of Do Overs One strategy we often find helpful with our kids is the

When something doesn't go well with the kids a "do over" can offer a chance to try again. It quickly breaks the bad pattern and replaces it with a better one. You're not digging into what should have happened, you just jump back and try again. It works whether they, I, or both of us messed up.

Context windows are now long enough, but the models still aren't able to handle the task well. A book might be 100k words, which translates to maybe 200k tokens, and flagship models are now up to 1M tokens.

(I expect they'll get there, though, especially with scaffolding and outlining)

Very Light Hardshell Suitcases I fly to a lot of gigs, and I check a bag with my pedalboard. Airlines generally limit checked bags to 50lb, and I'd like to be able to get as much stuff as possible into that 50lb; how light can a s...

I fly to a lot of gigs, and check my pedalboard. Out of a budget of 50lb, how little can I allocate to the suitcase?

It looks like all the <7lb options are Samsonite: Lite-Shock (5.5lb), Cosmolite 3.0 (5.8lb), C-Lite (6.3lb), Firelite (6.7), Proxis (6.8). All way expensive, but maybe one used?

Polyethylene Glycol is not Propylene Glycol Epistemic status: making things up About six months ago I got interested in glycol vapors for reducing infections in crowded indoor spaces. I decided to give it a try, but it went poorly. I was goi...

Don't substitute polyethylene glycol (Miralax) for propylene glycol (fog juice) if you're trying to reduce airborne pathogens. It turns out the syllables mean something.

Substack for Best Posts The most common feedback I get about my writing is that people like my posts but the fraction of interesting posts is too low. Some of this is hard to avoid, because I write about a wide variety of th...

The most common feedback I get about my writing is that people like my posts but the fraction of interesting posts is too low. I currently cross-post everything to a bunch of places; I'm going to switch my Substack mirror to just getting the best 10-25% of posts, about 2-4 per month.

Shallow Water is Dangerous Too Content warning: risk to children Julia and I know drowning is the biggest risk to US children under 5, and we try to take this seriously. But yesterday our 4yo came very close to drowning in a fount...

@juliawise.bsky.social and I know drowning is the biggest risk to US children under 5, and we try to take this seriously. But yesterday our 4yo came very close to drowning in a fountain. (She's fine now.)

PM2.5 Concentration and Composition in Subway Systems in the Northeastern United States | Environmental Health Perspectives | Vol. 129, No. 2 Objectives: The goals of this study were to assess the air quality in subway systems in the northeastern United States and estimate the health risks for transit workers and commuters. Methods: We re...

Reading ehp.niehs.nih.gov/doi/10.1289/... they definitely saw worse results than I did. I wonder how much of this is different equipment, different sampling locations, or the system improving ventilation sometime in the last 5 years?

Subway Particle Levels Aren't That High I recently read an article where a blogger described their decision to start masking on the subway: I found that the subway and stations had the worst air quality of my whole day by far, over 1k u...

This was wrong: www.jefftk.com/p/subway-par...

Subway Particle Levels Aren't That High I recently read an article where a blogger described their decision to start masking on the subway: I found that the subway and stations had the worst air quality of my whole day by far, over 1k u...

Subway pm2.5 levels peak closer to 75 ug/m3 than 1,000 ug/m3, and aren't something I'd wear a mask for. The post I shared earlier was misreading (my own) graph. Sorry!

Masking on the Subway Back when I was still masking on the subway for covid ( to avoid missing things) I also did some air quality measuring. I found that the subway and stations had the worst air quality of my whole day b...

I've started masking on the subway again. Not because of covid or anything we've found at the NAO, but because of the extremely high pm2.5 levels from brake dust.

Inflight Auctions Airplanes are strange places. Whatever you have on board when you take off is the most you'll have until you land. Want a sandwich? You can only have one if there's one on board. There are many things...

Imagine you load up your seatback entertainment, and one of the options is a burrito. But there's only one. You can put in a bid, and 1/3 of the way into the flight the person who bids the most gets it.

Even if profitable, though, I expect this would be widely hated.

Misconceptions on Affordable Housing draft post People often think 'affordable' housing is much more expensive than it actually is, and then conclude it's a scam to make housing for rich people. But this is often based on a misunderstan...

The legal term "affordable housing" is actually a decent operationalization of whether housing is something regular people can afford. If the area median income (AMI) is $100k, then a "50% AMI" unit is priced so someone earning 50% of the AMI ($50k) spends <1/3 of their income ($1,390/month).

At the NAO, we’ve created new estimates on how well wastewater sequencing detects different pathogens.

SARS-CoV-2 is again highly detectable, but common cold viruses are harder to detect.

We will use this research to compare wastewater sequencing to other detection strategies.

How Much Data From a Sequencing Run? Cross-posted from my NAO Notebook. Manufacturers often give optimistic estimates for how much data their systems produce, but performance in practice can be pretty different. What have we seen at the...

In sequencing wastewater we see slightly higher than advertised output for Illumina (~28B read pairs on a 25B flowcell), but much lower for ONT (~3Gbp on a PromethION where we'd hope to get 200Gbp).

That would be a good first step! If this general approach is going to work all these human-associated taxa should be decently correlated.

Though human 18S would probably work less well because (a) we're ribodepleting but (b) the depletion is inconsistent.

Multispecies Metagenomic Calibration Cross-posted from my NAO Notebook. This is something I wrote internally in late-2022. Sharing it now with light edits, additional context, and updated links after the idea came up at the Microbiology...

Metagenomic sequencing data is fundamentally relative: each observation is a fraction of all the observations in a sample. If you want to understand whether there's been an increase in the number of people with some infection you need to calibrate these observations: ...

Local Speech Recognition with Whisper draft post I've been a heavy user of dictation, off and on, as my wrists have gotten better and worse. I've mostly used the built-in Mac and Android recognition: Mac isn't great, Android is pretty go...

I've used dictation heavily on and off, but the built-in Mac and Android tools have been essentially stagnant since I started using them in 2020. I set up whisper.cpp on my Mac, and when combining it with Claude for cleanup it's excellent.

Grounding to Avoid Airplane Delays I recently flew through CLT and spent more time delayed than in the air. There were summer thunderstorms, and with the lightning it wasn't safe for workers to be out. This meant no loading, unloading,...

Lightning causes delays costing ~$50M annually at busy airports in stormy regions. Seems like you ought to be able to build a lightning-proof faraday cage of towers and cables for that much money. What am I missing?

Traveling to NC for Summer Soiree. Bad weather hit just as we were landing and we needed a go around. Landed fine the second time, but had to sit on the tarmac for 2.5hr before it was safe to unload. CLT customer service line is the longest I've seen, very glad we didn't have a connection!

I agree!

AI will likely grow to be a majority of the world's energy usage, but that will happen because of running tons of queries and replacing human employees (coupled with widespread technological unemployment, which I expect to be quite bad) and not because of even many individuals running a few queries.

AI in aggregate is a very small fraction of global energy usage, just like personal AI usage is a very small fraction of anyone's individual carbon footprint.

This is compatible with the increase at any individual tech company being large because their energy usage was relatively low to begin with.

Conflicted on AI Politics Many people are worried about AI, envisioning a large range of ways it could make the world worse. About twice as many Americans think AI is likely to have a negative effect as a positive one. At a h...

I'd love it if people thought hard about we should go with AI, and took both existential (pandemic generation) and everyday (unemployment) risks seriously. I'm very conflicted, though, on how much to push back on arguments where I agree with the bottom line while disagreeing with the specifics.

An American might use 30-50 kWh daily, of which 100 daily ChatGPT queries would be ~1%. In looking for places to cut, it's unlikely to be the best trade-off.

But to the extent that you're trying to minimize your individual climate change contribution, what matters is the proportion of your individual footprint that the activity represents.

Driving to the grocery store uses maybe 0.0005T of C02. All personal transportation in the US together releases ~1B T annually. Something can be a big deal at the national/global level while each individual's contribution is quite minor, since there are so many people.

Using ChatGPT is not bad for the environment And a plea to think seriously about climate change without getting distracted

The environmental impact of ai image generation is negligible: andymasley.substack.com/p/individual...

The human impact is more complicated, but for this kind of usage, where it's definitely not displacing artists, it seems clearly positive.

Ghiblification for Privacy I often want to include an image in my posts to give a sense of a situation. A photo communicates the most, but sometimes that's too much: some participants would rather remain anonymous. A friend sug...

A photo communicates a lot, but sometimes too much. For illustrating posts I've recently tried running pictures through an AI model to convert them into a Studio Ghibli-style cartoon. It gets a lot of things "wrong", but for my purposes that's a good thing, since it better obscures identity.

Busking with Kids Our older two, ages 11 and 9, have been learning fiddle, and are getting pretty good at it. When the weather's nice we'll occasionally go play somewhere public for tips (

Our older two (11y, 9y) have been learning fiddle. When the weather's nice we'll go busking. Better than practicing, builds performance skills, and the money is a good motivation!

I didn't anticipate the effect of their increased buying power, however, and am thinking about building responsibility.