hansvm

61 ms·

931.
by hansvm 2y ago
In the worst case, that must have intermediate space requirements equal to the entire collections, right?
932.
by hansvm 2y ago
Keep in mind, LLMs can't introspect. They integrate your prompt with their training data. If Python is an extremely popular programming language people like to write about using (it is), you'll expect a lot of people to state that
933.
by hansvm 2y ago
You can even have accrued sick days, starting at 0. Catch covid week 3 at Amazon? You'll start having problems above and beyond the illness in a hurry.
934.
by hansvm 2y ago
As a small amount of data, my asking salary has never been less than the highest paid H1B at a company plus another $30k/yr, and I've always been offered more than I've asked, despite the H1B position being more senior. That
935.
by hansvm 2y ago
> 1.5y vs 4y That's pretty common. Details vary, but having 1-3y of credit you can transfer in and then only having to stick it out for 1-3y to navigate the maze of upper-level course requirements happens all the time. In my case, M
936.
by hansvm 2y ago
> months to hire I've seen that at BigCo, but that's the exception. Every other place strongly prefers a start date of ASAP, with O(weeks) from initial contact as a next-best option. If you state that you aren't available
937.
by hansvm 2y ago
This is a fun technique that's surprisingly general-purpose (describing your error function differentiably and throwing your favorite optimizer at it). E.g., graph layouts are notoriously hard to glean any good information from, especi
938.
by hansvm 2y ago
HN ostensibly tries to have thoughtful, engaged conversations. LLM grammatical cleanups and translations are absolutely fair game. The comment has certain hallmarks (broad, sweeping summary, certain word choices, ...) suggesting that an LLM
939.
by hansvm 2y ago
They don't think they have the time, but that's because they view task completions as purely additive. Imagine you're working on this or that feature, and find a stumbling block in the legacy codebase (e.g., a poorly thought
940.
by hansvm 2y ago
And Gen-Beta are still shitting their diapers and crying unintelligibly when they don't get their way. An age-period-cohort analysis is appropriate when looking at that sort of data.
941.
by hansvm 2y ago
That quartile reservoir sampler example is ... intriguing? My experience with LLM code is that it can't come up with anything even remotely novel. If I say "make it run in amortized O(1)" then 99 times out of 100 I'll ge
942.
by hansvm 2y ago
> parametricity That feels like the wrong word for the thing you're describing. Linguistic arguments aside, yes, you're absolutely right. In Zig though, that issue is completely orthogonal to generics. The first implementation
943.
by hansvm 2y ago
Zig is lazy, and C++ is eager. I can define an infinite set of mutually recursive types in Zig, and only the ones I actually use will be instantiated (not an everyday need, but occasionally interesting -- I had fun building an autodiff pack
944.
by hansvm 2y ago
The problem is that "sufficiently smart" does a lot of heavy lifting. One way to look at the problem is to go build a sufficiently smart generic protocol and write down everything that's challenging to support in v1. You have
945.
by hansvm 2y ago
The gold standard is a purpose-built protocol for each message, usually coming in ~20x faster and ~2-8x smaller than a comparable proto (it's perhaps obvious why Google doesn't do this, since the developer workload is increased fo
946.
by hansvm 2y ago
It's comparable in the way described in sentence #2: > An individual company having specific needs isn't at odds with a general statement being broadly true. Google needs certain things more than reduced carbon emissions, and J
947.
by hansvm 2y ago
The same reason Google burns $50M+ in electricity each year using protobufs instead of a more efficient format. An individual company having specific needs isn't at odds with a general statement being broadly true.
948.
by hansvm 2y ago
I'd love a version of this tool that takes into account the data not being directly linearizable. E.g.: Suppose your plan is to suck it up in tech for 5yrs and retire. You want to know the place to live to maximize return-adjusted post
949.
by hansvm 2y ago
I asume that you mean only the weight loss pill exists? Would that be something like manjaro (inhibiting the desire to eat) or one of the few _actual_ metabolism boosters in existence (which all make you produce insane amounts of heat and h
950.
by hansvm 2y ago
Often. If you want expert results, you want to exploit the portion of the weights with expert viewpoints. That isn't always what you're after. You can, e.g., ask the same question many different times and get a distribution of &qu
951.
by hansvm 2y ago
740 kcal of pasta and cheese went into the dish, and under half (370 kcal) ended up on that plate. People vary, but even short, old people with no exercise have a maintenance metabolism of 3x that. To maintain my weight I need 10x that. I s
952.
by hansvm 2y ago
> I'd argue that floating point often gets used out of familiarity and convenience You're probably right. I can't tell you how often I've seen floats used in places they really shouldn't have been, or where nobod
953.
by hansvm 2y ago
LLMs (IME) aren't stellar at most tasks, cooking included. For that particular prompt, I'm a bit surprised. With small models and/or naive prompts, I see a lot of "Give me a recipe for pork-free <foobar>" that
954.
by hansvm 2y ago
> SMS 2FA is better than no 2FA Depending on the implementation it's occasionally more secure. For me it's never "better." A significant fraction of banks, retirement accounts, financial web services, ..., can fully r
955.
by hansvm 2y ago
> this is why ordinary users turn on biometric authentication despite some here questioning its security That's part of the reason. Another part is BigCo spamming the users asking for biometrics or whatever the current promotion-dri
956.
by hansvm 2y ago
For the normal borrowed-from-French idiom, yes, but somehow I don't mind both English words with their normal definitions sitting there.
957.
by hansvm 2y ago
It distilled bulk information into a form the author cared about. If nothing else it was probably fun, and a personal report on the things you care about can save minutes each day. I did something similar awhile back without LLMs. I enjoy k
958.
by hansvm 2y ago
Correct. Much like in Python or C, you can explicitly handle every error/break/early-return condition and manually free the resource. Each language offers a tool to make that easier (and more likely to be correct, especially as th
959.
by hansvm 2y ago
IME, not long. ChatGPT (v3) did poorly with Zig when it first came out, despite some Zig in its training code, and despite equivalent Python tasks being handled trivially. I worked around that for a bit by including a moderately lengthy pro
960.
by hansvm 2y ago
> 10 recovery codes Mostly unrelated, I fondly remember Google doing one of those "prove you're you" checks when I was in a different state, allowing me to waste one of my 10 recovery codes as one of the options, and still

More ›