In recent months (I don't use it enough to confidently state when the change occurred) Google's LLM summary more often answers the wrong question rather than confidently answering the question wrong.
If it answers and the answer is for what I asked, it is right much more often than not. With that said, "trust but verify" is more like "don't trust and also verify" anyway, which is the same metric that should be used for most websites.
This is true, but it doesn't make it less cringe inducing. Even when discussing actual requests from actual users, we don't pretend to be them and talk in their first person point of view. We just say "Eva wants to" or "admins need to" etc.
I'd love to see some source for this, because from what I've seen, 59% of undocumented immigrants receive some form of welfare (the state of California spends roughly $10 billion a year alone on this) and documented migrants are at roughly the same rate, double the native born population's rate of 28%.
I can't find study now from CIS but when I dug into it, it lumped citizen children of undocumented immigrants into "undocumented immigrant", and other studies classify them as native/citizen. It also looked at estimated education costs since schools are required to take all school age kids without regard to immigration status and many of big school systems don't have a breakdown of immigration status. CIS argument was: If parents are immigrants, shouldn't be contribute the kids cost to them regardless of kids status?
Looking at CBO numbers, CBO reports that recent immigration into United States is drain, in particular, at state and local level since that's generally budget that gets impacted the most by people. https://www.cbo.gov/publication/61464
> I'd love to see some source for this [proceeds to throw out a bunch of source-free numbers]
I looked up the source for these numbers.
It's the Center for Immigration Studies, and they are notorious for being a right-wing think tank used to dress up up anti-immigration policy using misleading statistics.
How notorious are they? Well, they are listed as a hate group with the Southern Poverty Law Center[1].
The numbers you're citing aren't 'welfare,' the biggest thing is using the education system. The thing that, long term, makes they into productive taxpayers. CIS doesn't figure the long-term benefits into their numbers because that would destroy their thesis.
The HUD secretary mentioned the number, and the number was also included in a Congressional report, though it ended up being through testimony, also from that same source, which I hadn't realized. I'd been under the impression the one claim was independently sourced from actual federal data, rather being from CIS.
In any case, none of this amounts to immigrants using less welfare than native born residents, which is what I was asking sources for.
The vast majority of the difference, according to the article, is social security benefits and Medicare. The non-retirement difference is very small.
> Native-born Americans use an average of ... $3,638 in means-tested benefits in 2023. By comparison, immigrants used ... $3,370 in means-tested benefits.
The article also ignores state provided benefits, and thus the disparity between documented and undocumented migrant populations. The article 's "much less" headline is not really supported by the evidence provided.
It's been a few years now, but I know a major manufacturer had some windows XP boxes around because they were the newest hardware/software they could get with native parallel port support to interface with their machines.
To give you an idea of the complexity of the manufacturing lines, shutting the machines down and starting them back up again was measured in days. Upgrading them would be obscenely expensive, let alone replacing them, just in the opportunity cost of not making anything alone.
For anyone that needs true native serial & parallel ports, a Dell Latitude E series laptop (or Precision 7xx0) with docking station is an easily-available option.
The last machines with dock port have 6th gen Core CPUs (Latitude E5x70), or 7th gen for Precision 7x20.
I don't really know the full details, as it was my brother who worked there and we just chatted over beers, but I believe the software itself that communicated over the parallel ports was 16 bit, and that (along with the native ports) was the reason they kept the old windows XP machines around.
It is possible, perhaps almost a certainly, that the company did not own the source code for the application and could or would not attempt to rewrite it to work with a modern operating system.
This is a top brand name company, so it isn't like they couldn't throw money at the problem, but with the number of lines in the number of plants they had, there wasn't really any reason to not keep the old hardware running (sans Internet connection).
Why waste tokens and ops team time and developer bug chasing time when you can get a signed contract with guaranteed uptime and support, and have those people be building something you can sell instead?
It's all opportunity cost.
Edit: throw in mandated security audits and maintenance for SOC2 compliance or whatever else you're on too, depending on the industry you're in.
Most states also rely on sales taxes and various fees, and assign limited property taxes. A Georgist land value tax would have much greater effects than the current tax regimes.
I had a similar experience earlier this year. Cheaper situation than a mainframe, but... why do some companies make it hard for them to take my money? I'm already a customer ffs.
I'm stuck on Windows, so oh-my-pi has been really nice. The others I've tried such as kilo do alright but tool calling can mess up a bit.
Only complaint is that connecting the agent harness to my local model took more work getting configured right than I'd like, but that's been true of most harnesses I've tried as well. Most assume you're using a cloud model and local model configuration is a bit of an afterthought.
I tried many of these pi extensions but they have tons of paper cuts. Don't remember which one had which issue but here are some I ran into:
* commands run by me (! prefix) are also sandbox blocked
* agent has no way to _request_ unsandboxed execution (e.g. if `kubectl whatever` is rejected by the sandbox, the model should have the chance to request permission)
* does not understand shell composition patterns (e.g. if `git status` is allowed and `git log` is allowed, then `git status && git log` should be allowed automatically)
* sandbox only supported on mac or linux. not both
All of that can be fixed by yourself. That's certainly the spirit of pi. But if you want strong defaults and batteries included (like omp promises) then that's just annoying.
- instruct model to write a markdown file with a phased plan to implement whatever feature or change I want
- start a new context, instruct model to implement one phase of the file
- review changes manually, then start a new context and have it do the next phase
- repeat as needed
I've never seen omp touch a file outside of the directory I start it up in, and the few times where I've been unhappy with a change git has been there to revert.
This could easily be a case of survivor bias but I've not had an issue with letting it go yolo yet.
Sometimes agents create huge half-minified one-off python/bash scripts to do some data processing. I'd prefer to neither review nor yolo these. Sandbox restricts reads/writes to designated directories, so at least there's no `rm -rf /` in there.
If it answers and the answer is for what I asked, it is right much more often than not. With that said, "trust but verify" is more like "don't trust and also verify" anyway, which is the same metric that should be used for most websites.
reply