> There isn't, pretty much everyone wants the best of the best.
For direct user interaction or coding problems, perhaps. But as API calls get cheaper, it becomes more realistic to use them for completely automated workflows against data-sets, or as sub-agents called from expensive SOTA models.
For example, in Claude, using Opus as an orchestrator to call Sonnet sub-agents, is a popular usage "hack." That only gets more powerful, as the Sonnet equivalent model gets cheaper. Now you can spawn entire teams of small specialized sub-agents with small context windows but limited scope.
I did create my own MCP with custom agents that combine several tools into a single one. For example, all WebSearch, WebFetch, Context7 exposed as a single "web research" tool, backed by the cheapest model that passes evaluation. The same for a codebase research
Use it with both Claude and Opencode saves a lot of time and tokens.
> But as API calls get cheaper, it becomes more realistic to use them for completely automated workflows against data-sets
Seems like a huge waste of money and electricity for processes that can be implemented as a traditional deterministic program. One would hope that tools would identify recurrent jobs that can be turned into simple scripts.
For example: "Here our dataset that contains customer feedback comment fields; look through them, draw out themes, associations, and look for trends." Solving that with a deterministic program isn't a trivial problem, and it is likely cheaper solved via LLM.
It makes sense if the dataset is so large that LLM cost is a prohibitive factor. Otherwise a frontier LLM has the advantage of producing a better result.
Yep, there is like two brands that specialize in three across.
A lot of new parents haven't yet realized that a carseat is wider than the average adult. Meaning that cramped middle seat isn't getting a third seat without very careful consideration and the right vehicle.
It's coming to mainstream seats though. Graco has had the Slimfit3 LX out for a couple years now; we had 6 (3-across in a Toyota Corolla and Honda Fit) and they worked great except for the kids not really liking them. Britax has the Poplar and One4Life Slim; Chicco apparently added the Fit3x when I wasn't looking.
The question itself feels like it calls for "Schoolhouse Rock" level basics about how the federal government works.
The federal government does not work like a private escrow account where a fee collected for X automatically goes to Y. Tax revenue comes in to the Treasury, and Congress decides what agencies are allowed to spend. So even if TSA screening is funded in part by a per-ticket user fee, TSA still does not get to just collect that money and use it directly. Congress has to authorize and appropriate it.
On a practical level, imagine the chaos if every federal department acted as its own tax collector and then set its own spending priorities. That is basically an argument for gutting Congress's oversight of TSA and treating it like an independent agency, just because Congress and the executive branch invented the modern shutdown in the 1980s.
Keep in mind shutdowns are a fairly new concept, that nearly no other country has. The US also didn't have it for most of its history. Congress could stop at any time it wanted.
Many government services do directly use fees to pay for operating the service. Those services can operate with no government appropriation, and the funds never leave the specific agencies' own accounts.
SFO doesn’t use TSA for security and works like this with whatever contractor it hire. I wish all airports would just use private security funded via usage fees, then we would never be held hostage because some whacko wanted to use masked thugs to beat up and shoot Americans.
From like the way it is and has been for many years?
From their website:
Covenant Aviation Security, a private company under contract with the Transportation Security Administration (TSA), provides passenger and baggage screening at SFO.
Visit Covenant Aviation Security to learn more about security screening at SFO.
I thought it was common knowledge, but anyone who has been through SFO in the last decade should know that they aren’t using TSA for the checks to get to gates. Always seems fast.
You’re right that that’s the default state. However, Congress could have set things up such that the fees would pass through to TSA’s budget (i.e. earmarked) but chose not to.
No, it's accurate, because TSA (or at least the feds) ultimately pay for it, but the company has some runway it can spend to keep the employees working on the assumption it'll be paid later, I.E. a buffer.
Windows' search has been broken for multiple generations now. Some people inside Microsoft seemingly even know, that's why the PowerToys team created "PowerToys Run." A Windows Search that actually basically functions correctly.
People act like it sudden was broken in Windows 11 when in reality it never worked correctly in 7, 8, 8.1, or 10 either. Instead of fixing it, they've only made it worse. It seems like nobody in Microsoft works on core stuff anymore.
If memory serves, Windows 2000 was the last version where search worked reliably. It was a simple linear search through files which could take a while on larger folders, but was reliable and predictable since it did not rely on a background indexing service which seems to get stale or just plain wrong most of the time.
If I search for “foo”, I’d like to get all files containing “foo” please, without a shadow of a doubt that some files were skipped, including those that I have recently created. I still can’t get that as of Windows 11!
> It was a simple linear search through files which could take a while on larger folders, but was reliable and predictable since it did not rely on a background indexing service which seems to get stale or just plain wrong most of the time.
It would be easy to have your cake and eat it too. Have the file search default to the index. Allow frustrated users to then click a button that says "search harder" which would initiate the full enumeration of the relevant filesystems. Of course some UX professional will tell me I'm wrong, they don't like anything they didn't think of themselves.
Almost everyone's search has been broken like this. I don't trust Windows search, I don't trust search in Explorer - but I also don't trust search in my Samsung at OS level, in Google Drive, in OneDrive, in Dropbox, and in just about any other webshit there is.
I can't put my finger on what's going on exactly; there seems to be some design choice commonly made, that makes search behave (or appear) inconsistently to the point you can't really trust it to be exhaustive. I.e. "if I search for term 'foo' and it finds nothing, it means 'foo' thing is not on file". It's a fundamental guarantee search systems should deliver, but these days most don't. Without this guarantee, I always seek out means to manually walk the resource catalog (e.g. filesystem tree).
Yeah, I've never experienced Windows search ever working. Even on XP, it couldn't find commonly opened folders or programs for me. It always felt like some sort of joke feature just meant to fool me into wasting time.
As far as I remember it was working well in 7 and 8 (deterministic and shows programs that you expect it to show). From 10 it started behaving erratically (same time it got binged but maybe unrelated).
It had problems in 8. I would frequently type my search term, see it was the number one result. I would then attempt to arrow or tab down and hit enter to launch that result. Between arrowing down and hitting enter, the result list would update/reorder and suddenly I'm launching some unknown program. Happened all the time.
I don't know how but it works beautifully for me on windows 11. What I mean is, I have been using windows for decades and I do not like any changes at all, they are all forced on me. But this change successfully turned me around. I find I rarely use File Explorer/file managers any more and access most applications and documents through the search.
I do remember it sucking on previous versions. I did use winaero tweaker to turn off the web results (and many other annoyances).
Context management is a core skill of using an LLM. So if it loses key context (e.g. tasks, instructions, or constraints), I screwed up, and I need to up my game.
Just throwing stuff into an LLM and expecting it to remember what you want it to without any involvement from yourself isn't how the technology works (or could ever work).
An LLM is a tool, not a person, so I don't have an emotional response to hitting its innate limitations. If you get "deeply frustrated" or feel "helpless anger", instead of just working the problem, that feels like it would be an unconstructive reaction to say the least.
LLMs are a limited tool, just learn what they can and cannot do, and how you can get the best out of them and leave emotions at the door. Getting upset a tool won't do anything.
In general, I can. In LaGuardia? Aside from right after 9/11 and during COVID-19 when almost all commercial travel stopped, I cannot.
I don't think people saying this stuff quite understand how busy LGA is even at night. I'd even go as far as to say that three minimum on duty with two in the tower at all times (for a ground/air split), would be the bare minimum for any hour or situation at LGA.
It does quiet down eventually. There's no scheduled departures 22:55-5:45 and only a handful of arrivals 23:59-6:45.
However, arrivals stay pretty heavy right up until 23:59 even on schedule and if you've got a lot of delayed flights (not exactly uncommon at LGA) - you may still have a lot of departures going out in the 23:00 hour.
I would not be surprised to learn that they're staffed to an appropriate level for what the schedule says is supposed to be operating at that time, but a very inadequate level for what actually winds up operating at that time on many days.
Initial analysis suggests they were running about 75% of full capacity in flight ops in the 15min prior to the accident. I doubt they were staffed to 75% of the daytime peak.
Yep. People who have never tried to add Mac support to an existing organization do not realize how freaking expensive it is.
There are basically two cases. If you use Microsoft, you are often already paying for Entra ID and Intune, then still adding the Apple-side pieces for Mac support: Apple Business Manager and often Jamf or Kandji. If you do not use Microsoft, you are buying the full stack yourself: Okta or JumpCloud for identity, Jamf or Kandji for device management, and Apple Business Manager for enrollment. Apple Business Manager is free, but the rest is not, and the cost adds up fast.
This means that, in practice, a managed Mac can easily end up costing close to twice as much to support as a Windows device.
Actually Intune handles MacOS reasonably well, you don’t need Jamf; that’s the way we went, and it’s okay-ish for the most part. By far the annoyingest thing is getting Macs bought before we went down the Business Manager integration route into MDM.
You think there’s a standard way to do that? Just install company portal? That worked in exactly 1/20 cases. It’s an exciting new error on every single device. Awful. Just awful.
The only thing you need out of any of those to correctly support the Mac is an MDM, of which there are free ones and expensive ones and everything in between. So long as it can deploy configuration profiles and declarative management configs, you can spin up Munki to be your pkg/script runner and script the rest. Installomator to install and patch applications.
But if you also wanted identity, there are plenty of free selfhostable SSO/ID providers out there. If you're just starting out and not at the scale where a big Microsoft CoPilotM365OfficeWhatever contract makes sense, you probably don't even really have a need for a lot of this stuff. A minimum contract for Jamf Pro is like $5k a year or something. That's two well kitted developer MacBook Pros per year in license costs.
Codex? I know OpenAI is really politically unpopular right now, but it has very high usage limits for the $20 plan. Claude ($20) and Codex ($20) are hard to beat in terms of pure value. Just set Codex on Thinking-High/Extra-High, and it is Opus 4.6 levels for sure (although both have their niche, where they're superior).
For direct user interaction or coding problems, perhaps. But as API calls get cheaper, it becomes more realistic to use them for completely automated workflows against data-sets, or as sub-agents called from expensive SOTA models.
For example, in Claude, using Opus as an orchestrator to call Sonnet sub-agents, is a popular usage "hack." That only gets more powerful, as the Sonnet equivalent model gets cheaper. Now you can spawn entire teams of small specialized sub-agents with small context windows but limited scope.
reply