hansvm

28 ms·

481.
by hansvm 1y ago
We don't have anything close to the totality of human knowledge digitized, much less in a form that LLMs can easily take advantage of. Even for easily verifiable facts powering modern industry, details like appropriate lube/speeds
482.
by hansvm 1y ago
You're not wrong. This is just a storage/retrieval problem. But ... the current systems have limits. If you want commercial success in <3yrs, are any of those ideas remotely viable?
483.
by hansvm 1y ago
A few points: 1. It's not necessarily different. Your ISP has monopolistic power over you, and it should be regulated more aggressively. 2. A non-mobile ISP is currently much less important than an Apple/Google account for interac
484.
by hansvm 1y ago
A subtlety still exists there. Copy-pasting is fine. What you're trying to prevent with DRY is two physical locations in your codebase referring to the same semantic context (i.e., when you should change "the thing" you have
485.
by hansvm 1y ago
Alright, suppose I'm passionate and an expert, and I want to write a blog going into a deep dive about logarithm approximations: How do I frame it? Is this for a math audience or a programming audience? If for programmers, are we talki
486.
by hansvm 1y ago
Yeah, we do 100k ML inferences per second. It's not a single server, but the architecture isn't much more complicated than that. With today's computers, indexing the entire internet and serving 100k QPS also isn't really
487.
by hansvm 1y ago
That's not the issue. Take pure blue wavelengths as an example. They also activate your green cones a little. Increasing the frequency activates the blue cones less and the green cones much less, yielding a higher ratio of blue to gree
488.
by hansvm 1y ago
This is something I'm hopeful will fall out accidentally from Zig's new IO interface. If everyone doing IO has the implementation injected, mocks (fault injection, etc) become trivial, along with any other sort of feature you migh
489.
by hansvm 1y ago
That's currently unknown. For all current practical purposes, kind of. The plank length sets a limit on spatial resolution of any information, so a finite region with (universally) bounded entropy per conceivable bucket on that scale s
490.
by hansvm 1y ago
Semantically it's just a table from instruction location to branch probability. Some nuances exist in: - Table overflow mitigation: multi-leveled tables, not wasting space on 100% predicted branches, etc - Table eviction: Rolling count
491.
by hansvm 1y ago
I don't think that meaningfully engages with the author's point. If you already know how to compute `len` on some arbitrary syntax soup then the difference is just a minor annoyance where you have to jump back in your editor, add
492.
by hansvm 1y ago
Cycles are often not what you're trying to measure with something like this. You care about whether the program has higher latency, higher inverse throughput, and other metrics denominated in wall-clock time. Cycles are a fine thing to
493.
by hansvm 1y ago
On top of that, dry docks are a common free amenity in boating/fishing towns, just using the tide to do your dirty work.
494.
by hansvm 1y ago
In activities prone to heatstroke, the advice is similar. Even for people with normal thirst detection capabilities, by the time that you realize you're thirsty it's likely too late. You need to be proactive about drinking enough
495.
by hansvm 1y ago
Sure: - There are how many computer architectures? A compile-once-run-anywhere binary looks closer to shipping a fancy interpreter with your code than shipping a compiled project. Runtime bytecode generation is one technique for making that
496.
by hansvm 1y ago
Yes, and if you're paranoid you can write if (comptime T == u8) { // code } to guarantee that if you're wrong about how the compiler behaves then you'll get a compiler error.
497.
by hansvm 1y ago
Or at runtime, if you'd like. You can create a generic binary that runs faster on supported platforms.
498.
by hansvm 1y ago
That reads more as an indictment against hyperscaled grocers than an argument that the current price gouging is unavoidable.
499.
by hansvm 1y ago
Easy. Don't apply the policy to new homes.
500.
by hansvm 1y ago
That's just a potential explanation for why they haven't bothered. I don't think we're disagreeing.
501.
by hansvm 1y ago
> You seem to suppose that they actually perform addition internally, rather than simply having a model of the concept that humans sometimes do addition and use it to compute results. Why? Nothing of the sort. They're _capable_ of d
502.
by hansvm 1y ago
Common misconception. That just means the algorithm for counting letters can't be as simple as adding 1 for every token. The number of distinct tokens is tiny compared to the parameter space, and it's not infeasible to store a map
503.
by hansvm 1y ago
It's easy to accidentally accumulate more than you need. E.g., when I got married we each were previously living our own lives. Big things like mattresses we consolidated, but an extra set of sheets is zero maintenance. It's also
504.
by hansvm 1y ago
Mine's 167-481ms (high jitter). It's the best internet I can get right now, a few suburbs south of San Francisco. Comcast was okayish, lower mean latency, but it had enough other problems that T-Mobile home internet was a small im
505.
by hansvm 1y ago
Doesn't that also count the investment losses they made last year, the cost of gobbling up other companies, the interest from the debt they acquired with previous expansionary practices, etc?
506.
by hansvm 1y ago
Costco is an interesting one to me. They're widely touted as being cheaper, but most foods there have some sort of upscaling factor -- "organic", some sort of regional variation like "sockeye" salmon instead of farm
507.
by hansvm 1y ago
When you control most of the market it works out a little differently. People can only eat so much food, but you can price gouge them relatively easily.
508.
by hansvm 1y ago
"compatible error type" Which is prone to causing propagating changes if you're not comfortable slapping duck tape on every conversion.
509.
by hansvm 1y ago
At a practical level, most of the language doesn't care about the distinction between errors and other types. You mostly just have to consider `try/catch/errdefer`. Your question then, mildly restated, is "how do people
510.
by hansvm 1y ago
Except, that sort of thing is important in places like gevent, pytest, and numba, and that functionality isn't easy to replace without a lot of additional language/stdlib work (no sane developer would reach for it if other APIs su

More ›