I’ve recently changed dozens of accounts to use Mozilla email masks. Most websites accept them, and the ones that don’t I think twice if I actually need that service. I have Simple Login and 3 custom domains if I really want to, before I give out a personally identifiable name. I’ve only seen one service that was super strict and only allowed gmail, outlook/hotmail, and yahoo.
I’ve been thinking about this essay for a few hours now, and I can’t shake the feeling that, if we’re using “tacit knowledge” as an argument for retaining seniors engineers, it’ll fail miserably.
The examples have that “I don’t like it, but I can’t tell you why yet. Give me an hour.” pattern to them. The author says this knowledge can’t be extracted or transcribed, but isn’t that exactly what that senior engineer does after that hour has passed?
That kind of work is impossible for an AI. It’s also impossible to prompt an AI into doing, because the input my colleague used — a constellation of subtle features, a year-long history of similar bugs, a half-conscious memory of “where pain has come from before” — isn’t anywhere it can be fed in. It lived in him. He’d built it the slow way, over a decade.
Well, it’s “impossible” in the same way that it’s impossible to a newcomer: because it’s unwritten; it requires knowledge from other sources. Currently. There’s nothing inherent about this kind of knowledge that prevents it from being written in the first place. The fact this knowledge is missing only tells us that someone has never bothered writing it down before. Perhaps they didn’t see the need, perhaps that wasn’t a priority. With automated systems writing code and documentation, well-managed projects will also be better at tracking this kind of knowledge and - finally - writing it down.
You’re the one who can look at a PR and say “this fits us” or “this doesn’t.” That sentence is mostly tacit.
If you’re a senior engineer and can’t argue why something is or isn’t a good fit, and yet you are locked on either one of those… I don’t know what to tell you other than that, maybe, you lack communication skills.
I do think that projects have this sort of tacit knowledge to them, I’m not questioning that. But:
I don’t believe this is an argument in favor of senior engineers because it may be misinterpreted as ignorance (“If you can’t explain something to a first year student, then you haven’t really understood.” - Richard Feynman);
I’m not convinced this kind of knowledge would be required for a project to succeed;
I’m not convinced AI itself isn’t able to extract patterns from the project that we would call tacit knowledge. Pattern recognition is kind of their thing, after all.
yup, it helps reducing wrist pain from typing
tbh I’d rather have a 5% chance of being rickrolled for each link I click for the rest of my life than having this stupid invasive captcha everywhere
crosspost it to c/actually_infuriating
the original comment says “We should be able to retrain local models so they can develop an actual experience without prefilling the context.” - it turns out we can. Not sure why you’re trying to attach labels of user vs creator, when the premise already mentions retraining.
Btrfs is usually last in perf benchmarks, so I wouldn’t do it myself
in fact, I have regretted using btrfs because it made my laptop noticeably slower
the phoronix benchmark posted in this thread uses artificial loads and doesn’t compare performance against other file systems, just compression on vs off for btrfs
but you can, as long as it’s open weight. Fine tuning and training are pretty much the same process
lol how do you think LLMs are trained in the first place?
I don’t see the probulent here
heavily depends on the model and quantization level
choose the model you want on this website and it’ll give you some specs likely to run it
any/most distros will do, especially if you run it on Docker
if you’re going with intel cards (best $ per GB VRAM right now), you could get a decent machine under $3k
that’s what happens when the western alternatives are 10x more expensive
that’s just good form, regardless of who’s reading the md
that whole line is 3 tokens btw, they’re wasting more time and energy just discussing that
If you do named parameters in rust, you would need to set every parameter.
If that’s the condition, yeah, it’s bad. The good thing about how Python handles it is that they can still be mandatory or optional. And that makes it a more flexible option than the builder pattern because it applies to any function or method.
you can have a better data structure in any language, but rarely someone will bother doing that for booleans
This! There’s also a longstanding open issue on rust to allow a similar way to pass keyword/named arguments. IMO every modern language should have something like this. Makes all the difference when reading code.
Inlay hints from your editor are a good help, but this should be built into the language.
either you didn’t get the idea, or you don’t know what’s identity theft