thaeli

Yeah, my glasses are a substantial part of my personal style and I'd be sad to have to go back to something generic looking. It's been a minus with other smart glasses as well.

Also I need to wear safety glasses with side shields frequently and would really rather be able to integrate there. Particularly relevant here since that's also in noisy environments where regular speech recognition struggles.

I really like the distinction between "prediction" and "generation" here. Thank you for putting that into words so clearly.

#1 seems like a feature, not a bug. The real problem is that health insurers in their current form exist at all.

Not for me. I tried out the high refresh rate thing, switched back to 60.. can't tell any difference at all. Same for high DPI vs "regular" DPI screens.

Screen flicker on CRTs with low refresh rates was terrible for me, but with modern flat panels.. nope, can't really tell the difference.

Huh. Maybe that’s why I find Claudish fairly easy to read.

For most of my life, I never quite “got it” about the night sky. It was pretty, but not THAT different away from cities.

I only recently learned that visual snow syndrome is a thing, and that most people don’t have it. For me, there is a faint background of multicolor static (analog TV on a dead channel, for those who remember that) on everything, all the time. Light washes it out, so it’s only in the deep dark that I’m absolutely surrounded by it. (It’s basically visual tinnitus, which in turn lead me to discover that I’ve also had auditory tinnitus my whole life, it’s just a soft static not a “ringing” so I never realized it was the same phenomenon.)

It’s bizarre to me that Claude Code doesn’t have more guardrails on plan mode. I mostly use Copilot and the plan mode there has teeth. In plan the agent does not have permission to write to the filesystem outside of a temp directory and the plan document itself, and tool calling is heavily restricted.

There’s also a “ask” mode which is read only. Both are enforced by the permissions model, not just a system prompt instruction. I’ve seen the model “forget” and try to start coding - it bounces off a hard permissions failure and that “reminds” it that it’s in plan mode.

How it works in my state: For in person renewals/changes, they punch VOID into your old license, give it back to you, and give you a temporary license printout. Then your new license is mailed a couple weeks later.

For remote renewals/replacements, they just give you a temporary license to print out and then mail you a new one. The old license never gets physically cancelled.

There absolutely is a market - anti-AI sentiment among creatives isn't going away. Like any other "purity" branding, it's a real commitment and severely limits your future ability to pivot.

I don't know of anyone in the VC or traditional incubator communities who has an appetite for this. And the groups who are an ideological fit don't have the funds to be worth pursuing.

It's a niche. Like most niches, there's the opportunity to prosper, but yeah - bootstrapping (or angel-bootstrapping if you're lucky and have a personal connection) is pretty much the only way to go here.

I will take those concerns more seriously when we start hearing similar consensus from the Chinese side.

A routine urinalysis has always been part of my normal annual physicals.

It’s still the standard for “white goods” (larger home appliances) in general.

Same here, except we just defaulted everyone to Auto and expect the percentage of tokens spent via the auto router to be high.

At an operational level for the railroads, we have PSR, which is even worse.

Also, my phone follows me between vehicles and provides the exact same interface in all of them. Heck, I can switch vehicles and my podcast or music starts back up right where I left it. That alone is a major win for the phone-based approach - if I drive my spouse's vehicle for instance, or my old truck because I need to haul something, doesn't matter, they all have the same navigation and audio.

I'm the same way. The scanlined, subpixeled versions just look terrible to me.

FPGA based devices that can do this, and quite well, do exist, they're just expensive. The RetroTINK-4k Pro is the top of the line as of this writing but it's a $750 converter.

Emissions related components work very similarly, replace the software and it’s presumed to be a defeat device unless proved otherwise.

For emissions related components, EPA rules do kick in though. While the current administration appears to have paused enforcement, their position for many years has been that running anything except factory approved firmware on an ECU or other emissions related computer constitutes a “defeat device” and is illegal for an on road vehicle subject to emissions controls. (Granted, in practice 99% of the reason anyone installs new firmware on their ECU, or switches to an aftermarket ECU, is for a “tune” that does affect emissions. I’m sure there is some edge case exception, but it’s very rare in on road engines.)

The alternative, and there are a very few tunes that have done this, is to prove to regulators that the tune does not negatively affect emissions in any way. In practice this is done by getting a CARB exception since they’re the ones actually checking for tunes.

Well, many areas have banned app-only payment requirements (along with card-only) so it’s possible we’ll get some mandated alternatives.

Ironically, this check would be a pretty good use for a LLM.

Are there any additional mitigations folks are using for this? This issue is the only reason we can’t turn classic PATs off entirely.

Short lifetime mandatory reauth to enterprise SSO seems to be the best available, but it’s inconvenient for the single Classic PAT we actually need.

Yeah, interested to compare this with optimized and properly cached GitHub Hosted Runner builds.

The biggest problem with SE for me, and this is related to the culture issues you're talking about, is that the site has no good way of deprecating "formerly correct" answers. Even if a better, more correct answer is posted later, the reputation system has a huge incumbency bias in favor of older answers that have accumulated upvotes by being the best available answer at the time.

Their knowledge repository is slowly rotting under the weight of having to ask every time "okay, is this correct-sounding, highly upvoted answer actually (still) correct, or is it 10 years out of date?"

There was a common scam back then of sending cheap goods to someone, then billing them a high price, when they didn't order anything to begin with.

Compare this to EMS, where unless you're lucky enough to be fire or hospital based, worker protections are almost nonexistent.

Unless we're ever ready to start treating bike thieves like horse thieves, that's unlikely to ever change. And I doubt we'll ever have the political will to do that, or even a kinder gentler version of it.

The concentrated force for candidates to counteract this would be an industy-wide union. Which isn't happening.. but if you look at unionized industries this is one of the major benefits for candidates, or at least for candidates who make it into the union..

It's not ideal, per app would be preferable, but at least global can be a toggle in the global quick settings UI.

As a user, I don't care about your ML model. I care about not sending you personal info. Coarser info isn't good enough - you can convince me you actually deserve my real info, or you can get no info (preferable), or you can get fake info (alternative if you degrade or break my user experience because I wouldn't give you that info).

I understand there are some legit use cases for validated info. Unfortunately, targeted advertising and other types of profiling are also common use cases for the same info. It's a lot like MAC randomization on public wifi - it sucks, it breaks legit use cases, but it was needed because too many companies were using it to track people.