jefftk
jefftk on Hacker News
Vowels definitely require tongue movement! Leave your mouth closed and whisper "a-e-i-o-u" and you can feel your tongue moving.
Vowels change the shape of your mouth's cavity to change which frequencies to emphasize ("formants") and we hear the differences in vowels by analyzing those formant patterns.
I don't think this would actually work; have you tried it, or do you know anyone who has?
The issue is the grid-tied solar. It looks for a grid signal, and sends power out to the grid. When you connect your F150's 240V bed outlet, that will look just like the grid, and if you're producing more than you're consuming the solar will try and send excess power to the truck. Maybe this damages the truck, maybe it just trips and shuts down, but either way your house goes dark.
Florida gets hot, but not wet-bulb-above-95F hot. The proximity to the ocean keeps it from getting quite that high. People lived there before AC.
I think they're proposing switching your furnace from hard wired to plugged, so you have the option to power it either from an outlet (grid) or solar/generator/etc. This can't backfeed the grid, and is very safe.
It looks like when it's against code it's "by default": if an appliance is not explicitly listed as acceptable for a plug it has to be hard wired. ND and OK have passed state level amendments allowing it, some places have done this locally, and in many places building inspectors allow it with their discretion.
Reading more about this, it doesn't look like it does what I want. I'm looking for:
1. House has solar.
2. House does not have batteries.
3. Grid is down.
4. When solar production exceeds house consumption, the extra goes to the electric vehicle.
5. When house consumption exceeds solar production, the extra comes from the electric vehicle.
This really doesn't look like something the F150 can do.
How does solar recharge your F150 with this setup?
As far as I can tell, the Leaf never had this in the US. The 1st and 2nd gen hard the hardware, but the chargers weren't there. And the 3rd gen has only announced V2L.
From my perspective, "zero-interrupt automatic transition" is not that important, but being able to have the system smoothly move between having the solar recharge the car and the car power the house during an extended grid outage is critical.
> others start making me wonder “how often does the power really go out”? It all seems silly for 3 days outage once every 5-10 years.I think very little is worth it for three days every few years. Instead, I see the benefit of batteries + solar or a serious generator as reducing tail risk from rare serious disasters that leave the grid down for months or longer. These are not likely, but if they happen I'd want to have something in place.
This is exactly what the post is trying to address: during a serious long-term outage, batteries on their own will run empty. What you need is the ability to recharge batteries from solar during an extended outage, while also flexibly drawing from the batteries when solar isn't enough.
Whether we call it "permanent summer time" or "aligning with the political region to the east" it's the same thing: late sunrise, late sunset.
(Spain is effectively using single-double DST, based on the delta from solar)
This seems like it's also an argument against heating rooms in winter?
An AirFanta 3Pro on low (2) is 5W, which is $10 on average (but more like $15 in high $/kWh places like MA and CA).
IQAir HyperHEPA is substantially worse than the AirFanta 3Pro. The 3Pro gives you about 10x more clean air per dollar (2 CFM/$ vs 0.2 CFM/$), and 2.5x more clean air at a given noise level (45dBA: 213 vs 86 CFM).
I agree big fans moving slowly is better all else equal, and something AirFanta-like that used a 12" fan could be even better.
My aunt is immunocompromised and carries one around with her in crowded situations, keeping it pointed right at her face 8-12in away. It works well for her, especially since she's had a lung transplant and doesn't do well with a mask.
It just looks grey, but the housing is also grey, so I don't think it generally reads to people as dirty.
I have three in my living room, and while they're not fashion they blend in fine:
https://photos.app.goo.gl/wvRc4hEYqQep4zgd9I leave them on setting 2 (unless someone is sick or we're having a lot of people over), and we get ~425 CFM for about $40/y in power (~5W each).
There's either a typo or something wrong with your sensor: you can't have pm10 < pm2.5.
(Equivalent to saying you have more people under 2.5 years old than under 10 years old.)
There's only a ~1 in 17 chance that they'd all be below 256 if 384 was the limit. My money is on accidentally generating a 1-byte random number.
> for now the autonomous flight control software is incapable of complex, dynamic missionsI give it 1-2y.
If you have not watched the ramp video, I really recommend it. It's very impressive, even if it doesn't make any sense in a consumer vehicle.
If you mean one that can do the stunts from the demo I'm in, but not for $10k: that's more than I'm willing to bet with people I don't know. How about $100, since most of the point of a bet is formalizing a disagreement and showing that you're serious?
> Insane acceleration to get up the rampIt climbed the ramp very slowly. It doesn't require insane acceleration at all, just a reasonable amount of power at the wheels combined with a fan that can hold it to the track.
This is why they temporarily spot weld manhole covers before street races.
Posted by
jefftk on Thursday October 15, 2026 @12:53PM
from the astralcodexten.com dept.
AI-written satire, but reasonably funny.
The math doesn't work, though: 21k bags at $1.80 donated is $37,800 in donations, but you could clearly donate much more than that working at Jane St.
Today I can write what I'm looking for, as if I'm writing a design doc, and the models can do the rest. A year ago it was faster for me to just write the code myself. Models today are enormously better than what we had in 2025.
As is standard for academic papers, this is an attempt to describe the state of the world as of when they did the research. But then it takes time to write it up and get it out, after which newspapers report it as current events.
In this case, we're looking at a picture from 2025. The models involved in the main agent results were primarily Claude 3.5 through Sonnet 4.5, GPT-4.1 through GPT-5, and Gemini 2.5. And the paper's design means you're mostly looking at the very initial effects of deployment, before companies developed good review processes.