But companies don't have to bear the cost of raising a human from birth, or training them. They only pay the cost of hiring them, and that includes cost of maintenence.
Add to that the fact that we can't blindly trust LLM output just yet, so we need a mearbag to review it.
LLM will always be more expensive than human +LLM, until we're at a stage where we can remove the human from the loop
> But companies don't have to bear the cost of raising a human from birth, or training them.
The costs do exist somewhere though, and must be paid by someone. There's no free lunch, and the human lunch is very likely far more costly than the LLM lunch.
> Add to that the fact that we can't blindly trust LLM output just yet
Can't blindly trust human output either. That's why there are various tiers in roles, from junior-equivalent to senior-equivalent, and the actual user of the product is always the final arbiter. There's ultimately nothing different, except that the LLM iterates on issue resolution in seconds to minutes, whereas the human equivalent takes hours to days.
How? The government runs it, and/or heavily regulates it, and shifts costs.
> and they're better off for it
In the US, the cost of medical care rose in step with inflation until 1968. After that, it rose at a much steeper rate, and has not slowed down. 1968 was when the government got involved with health care.
Countries with a heavily-interfered health care system are poorer as a result.
I wish all model providers would converge on a standard set of files, so I could switch easily from Claude to Codex to Cursor to Opencode depending on the situation
Issue is that both harness and specific model matters a lot in what type of instruction works best, if you were to use Anthrophic's models together with the best way to do prompting with Codex and GPT models, you'd get a lot worse results compared to if you use GPT models with Codex, prompted in the way GPTs react best to them.
I don't think people realize exactly how important the specific prompts are, with the same prompt you'd get wildly different results for different models, and when you're iterating on a prompt (say for some processing), you'd do different changes depending on what model is being used.
Having experimented with soft-linking AGENTS.md into CLAUDE.md and GEMINI.md, this lines up well with my experience. I now just let each time maintain it's own files and don't try to combine them. If it's something like my custom "## Agent Instructions" then I just copy-pasta and it's not been hard, and since that section is mostly identical I just treat AGENTS.md as the canonical and copy/paste any changes over to the others.
If you create a CLAUDE.md with contents of @AGENTS.md that works and has the added benefit of allowing Claude specific instructions too, to be added below.
I think one of the main examples that i saw in a swyx article a while back is that using the sort of ALL CAPS and *IMPORTANT* language that works decently with claude will actually detune the codex models and make them perform worse. I will see if I can find the post
Because that just does it for you, it doesn't help me understand how to write better prompts.
Actually, I can just read the skill with my own eyes and then I can also learn. So, thank you for sharing. It's interesting to read through what it suggests for different models - it fits for the ones I work with regularly, but there are many I don't know the strengths and weaknesses of.
Cursor supports all the Claude file patterns, including plugins and marketplaces. We leverage that to support both Claude and Cursor with same instructions and skills
Is this an outdated requirement? What's the attack surface of an email vs fax? Unless they ban phones at the office, someone could just take a photo of the documents the patient faxed or mailed them
It's a current requirement. (Source: I'm adjacent to a doctor's office.) Two big advantages of faxes are that 1) they're point-to-point, and 2) there's zero caching between the sender and receiver.
If everyone had a fax machine such that you'd commonly get a working fax receiver if you mis-entered the recipient's number, then #1 wouldn't be such a big deal. But in reality, if you enter a fax number, and the other end actually answers and responds with a screech, it's extremely likely that you're connected to the right party. (Also, I bet 99% of modern faxing is triggered by a nearby computer, or by pressing one of the preprogrammed speed dial buttons on the fax. There aren't that many opportunities to misdial the number in the first place.)
That second is also a big deal. There are no intermediate servers which may be caching and inappropriately storing the data, except maybe the NSA, but what can ya do. The sender may have a cache, in the form of a print spooler. The receiver may have a cache where it temporarily stores inbound faxes and prints them asynchronously. But since both of those devices are owned and controlled by the parties in the communication, that's not a legal issue.
I'm not advocating for faxes. They're a slow, clunky, lossy, pain in the ass. And yet, they do have specific properties that are pretty sweet. I guess the equivalent would be if I could ask you to send a PDF to my specific IPv6 address, and you could peer-to-peer shoot it directly to me. If I typoed the address at all, it's statistically "unlikely" that another person would be listening on that specific IP a that specific time. And if it were truly P2P, then you and I would be the only 2 who ever touched the file, except maybe the NSA, but what can ya do. Alas, I don't see that replacing fax machines any time soon.
Amazingly enough, this is actually not true. Many smaller doctors' offices still have a physical fax machine. I work on automation for certain processes in healthcare and a very large proportion of the faxes we receive come from physical fax machines. You can see artifacts on the fax itself and sometimes the cover letter will have a scribbled note.
Do they still have POTS lines? Telecom companies are shutting down old SS7 switches so eventually all faxes are ultimately being sent over the Internet and it will be entirely vestigial.
> I guess the equivalent would be if I could ask you to send a PDF to my specific IPv6 address, and you could peer-to-peer shoot it directly to me.
That's not exactly complicated if either party owns a web server. Which - last I checked - the government has.
Just give the person who needs to send the sensitive documents a short link like uploaddocuments.gov, have that page ask for some basic identifying info, and have a box for the user to drag and drop a file. At which point the browser will p2p upload that file over HTTPS.
That’s kinda true, but adds a few steps over cmd-P “print to fax”, paste in a phone number, done. And when done, the fax workflow has been tested and approved in courts. It’s a known entity.
I don’t love faxes. This isn’t me saying we should keep them forever. We shouldn’t. Still, there are reasons they’re still widely used for medical stuff today. If CMS or HHS rolled out a new method and told doctor’s offices to start using it if they want to get paid, the industry would switch in a heartbeat. Short of that, any other alternative will take approximately forever.
Possibly! I haven’t used my Verizon CO badge to work on telco equipment in a few years. How is it fundamentally different now so that my brief description is wrong? I like to learn new stuff!
Output: I’m thrilled to announce that I’ve just successfully completed a major internal release. This process required significant focus, a commitment to letting go of what no longer serves me, and a streamlined approach to output. It’s a reminder that sometimes, the most impactful growth comes from a high-volume delivery. Feeling lighter and ready to tackle the next challenge! #GrowthMindset #Efficiency #OutputMatters #Relief
I assume that if you take a break you'll have missed a lot when you come back, at the pace things are evolving. Which is OK for some people like OP but maybe not for simonw
reply