roncesvalles
Entrepreneurship isn't about building a company or even making money anymore, it's about the prestige of announcing that you either got into a selective incubator or that you raised. Minds corrupted by the panopticon.
I rushed to the Claude platform console and was expecting to see a 4 figure usage bill or something. Luckily there was no usage.
But imagine if I were traveling. Even a rot13 of the key would've bought me enough time to rotate.
2 lessons:
- don't name your env var "CLAUDE_API_KEY". Call it "ARG_1" or something
- at least do some obscurin' so that the value doesn't start with "sk-ant..."
LLMs for many months have been at a level where if you know exactly what you want, you can simply let the LLM bang out an implementation and never have to read it. But outside of simple problems, you probably don't know what you exactly want.
That being said, I've had considerable success ending every prompt with "ask me for any further details that I haven't considered" or some variation thereof.
This has very interesting implications for everything in the world, btw. Correctness does not track majority consensus.
This Microsoft deal for the large part is not being offered to software engineering orgs.
For work in which you are in the loop, absolutely. Do people not value their own time or something? Why in the world would you use less than the best LLM available to you? You're not going to run out of the $20 plan even if you use Opus 5.5 at xhigh and code 8 hours/day (assuming you're doing it piecewise using a Claude product) so what exactly are you conserving by using something inferior?
For large continuous volumes of programmatically submitted LLM tasks, sure.
Like that old adage "that's a feature not a product" - Jev is just a system prompt, not a product.
The latter still gives you many-X productivity improvement while the quality is pretty close to handwritten code.
LLMs are still not good enough where you can unleash it upon a huge codebase and get good results btw. Quality goes to shit once a context uses up around 50k-100k tokens. That's often only enough for one feature change and whatever validation the LLM wants to do for it.
It's fair to say that most new software is written with AI help, but that doesn't mean someone just prompted a few things and pressed deploy.
Claude Code (Opus 5.5 xhigh), when it makes changes, tests the relevant endpoints for behavior based on what it touched. You could also just explicitly ask it to as a follow-up prompt. You do need to have instructions on how to set up a test environment or use infra for which dev env testing is well-documented online.
And yes, it's non-deterministic. It's not gonna test everything exhaustively or systematically like a fuzzer. It's basically doing what a manual QA would, at best. It probably doesn't even cover every edge case each diff. But I'm fine with this tradeoff for not having unit tests.
Caveat emptor:
1. These are relatively small-medium programs (~10k LOC) where the frontend and backend code reside in the same repo, and the frontend is strictly a web app, and where 100% of the functionality can be tested in the dev environment (incl the database).
2. I'm the only person working on these repos right now.
3. I read every line written by the LLM and have things I don't like refactored or stashed and rewritten with a modified prompt, mostly based on intuition.
>How many tokens does this process take?
I genuinely believe anyone who exceeds usage on a $20 plan doesn't use Claude Code correctly. Always use a new context wherever possible. Always make surgical changes (or, broad changes with well-defined success criteria). For anything complex, get a plan from the LLM at least so that you know it understood your ask. Ctrl+C and `git stash` are your best friends. Keep everything clean, don't build castles on garbage. In the LLM era ironically you should never put off tech debt because it will just compound.
Unit testing imo is pre-LLM. It was mostly to keep the human honest because the human can't possibly read all the code to make sure nothing broke.
Basically, don't let the AI coding era peer-pressure you into being a bad engineer.
This is what Terence Tao's ChatGPT chats look like: https://chatgpt.com/share/6a5fdc7a-d6f8-83e8-bbea-8deb42cfed.... Can a layperson prompt the LLM like that?
"Garbage in, garbage out" is a polite way of saying the LLM's output is a reflection of the prompter's own intelligence.
To me the subtext of the movie was that a sociopath's manipulation is so strong that even against the heavy counterforce of the filmmaker's bias, it makes it through. I agree with TFA's general premise that the film forces you to confront how easy it is to deceive you -- but it's not the filmmaker who's doing the deceiving. Sociopathic manipulation is about sparking doubt. Gaslighters don't need to prove anything, they just need to plant a "but what if" in some corner of your mind. The fact that the author felt compelled to write TFA at all is ironically a product of that manipulative power.
Another way to think of it is: the documentary makes you think whether, if you were in the room with her in the years that Theranos was operating, would you have fallen for it? If you feel the need to write apologia after watching a documentary that's basically a hit piece, you know the answer.
[1]Lots of other cars have technically faster times than Teslas, but most of them require launch control, good tires, and sound like the engine's about to explode. You can floor the Tesla at any time and the software will deliver however much power it can without slipping.
True racism is about exclusion. If you'd never meaningfully associate with a person of some certain race(s), you're a racist. Doesn't matter if you're the most polite soul in the world.
There are people who have traveled the whole world, eat ethnic food all the time, do yoga, vote left etc but would never make friends with someone not of their own race. They are the real racists. Racism is simply defined as whether you consider race to be a meaningful way to classify people.