habitue

22 ms·

271.
by habitue 4y ago
Making customers happy isn't totally orthogonal to making money right? In midwit meme format: Low wit: Make customers happy. Making customers happy is how you sell products and win customers from competitors Midwit: Screw over customer
272.
by habitue 4y ago
Do travel agents have some benefit when travelling internationally or something? I have not heard about travel agents since the 90s. Maybe I'm missing some incredible deals? I kinda expected them to not exist any more
273.
by habitue 4y ago
3.5 was the gpt-3 pretrained model plus instruction tuning and rlhf
274.
by habitue 4y ago
This is true, and so you have to look at it case by case. I think if most developers look at the switching costs of bazel vs. just using a decent build tool like cargo, they wont make that tradeoff. Similarly, they might choose some other t
275.
by habitue 4y ago
I don't think it's arrogance, it's just that blaze inside google is really nice, and they assume that using the tool outside google will also be really nice. The issue is that to make it really nice: - you have to convert ev
276.
by habitue 4y ago
Hey, what a surprise, ex-google engineers dragging everyone into bazel... Maybe you have some really complicated use case that makes cargo insufficient, but really, it's pretty damn nice and it's integrated with everything. There
277.
by habitue 4y ago
I mean, estimating that there are a couple thousand c++ developers is pretty funny.
278.
by habitue 4y ago
Yeah, everyone should think about what kind of company they want to work for. Maybe you want a big name on your resume, a couple years at stripe might not be a bad idea. Similarly, if they get a bad anti-employee rep, they'll suffer in
279.
by habitue 4y ago
Yeah, I find llms are really useful for taking premises you supply, and trying them out, or exploring inconsistencies in them. It's decent at the synthesis kind of creativity (mashups, remixes, etc) but it isn't smart enough yet t
280.
by habitue 4y ago
Don't fall victim to the halo effect! Both Epic and Apple are right on some issues, and horrifically wrong on others.
281.
by habitue 4y ago
If you go public, there are a bunch of requirements on you as a company. you have to regularly tell everyone your P&L and a bunch of other things Stripe would rather not tell anyone.
282.
by habitue 4y ago
> ChatGPT cannot give an answer that is the result of a "reasoning" before laying out the "reasoning". This is slightly too strong a statement: it can give an answer before reasoning it out, but it only gets a singl
283.
by habitue 4y ago
pre-push hook?
284.
by habitue 4y ago
I interviewed an artist who uses nannou to make art daily, back in 2021: https://blog.abor.dev/p/making-generative-art-with-rust
285.
by habitue 4y ago
It does have a repl! https://github.com/evcxr/evcxr Also, while cold compile times can be very slow of you have a lot of dependencies with macros etc, incremental compile time is often very fast. Rust has one of the
286.
by habitue 4y ago
I think they're referring to the query language itself not being json or yaml
287.
by habitue 4y ago
There's more to life than fitting in. Letting kids from spmenother family dictate what you spend your money on is lame. As others have said, the issue is the kid's feelings and self confidence not actually technology. That applies
288.
by habitue 4y ago
certainly, in terms of writing scripts, deno seems nicer. Like, if I want to write typescript that just runs and does something, without having to carry around a node_modules folder, etc, deno seems like it might be nice
289.
by habitue 4y ago
In your 3llmtricks link, you are missing a hyperlink to the HyDe paper: > There’s been a lot of attention around the HyDe paper recently [LINK]
290.
by habitue 4y ago
Ah! Interesting, I thought the capability was capped by parameters, but you're saying you can keep getting more capability from a fixed parameter size by continuing to train past what the chinchilla paper specifies. That's really
291.
by habitue 4y ago
I just meant chinchilla optimal in terms of the corrected scaling curves from the chinchilla paper. The original GPT-3 was way larger than it needed to be for the amount of data they put into it based on the curves from the chinchilla paper
292.
by habitue 4y ago
Speculation: GPT-turbo is a new chinchilla optimal model with the equivalent capabilities as GPT-3.5. So it's literally just smaller, faster and cheaper to run. The reason I don't think it's just loss-leading is that they mad
293.
by habitue 4y ago
So ambient uses WebGPU, which on rust usually means the wgpu library, which has a vulkan backend for platforms that support vulkan. It's possible that's what you're thinking of. The issue is that vulkan is really low level. S
294.
by habitue 4y ago
The key insight here seems to be that you usually need to specify the tasks for a repo twice: once in your task runner (Makefile, package.json, bazel BUILD file) and then again in your readme so that people can learn what custom tasks exist
295.
by habitue 4y ago
I think the point being made is about the tooling being available to do it easily.
296.
by habitue 4y ago
WebGPU makes cross-platform seamless because it's got multiple backends, and it's quite capable. It can be extended to add native platform features
297.
by habitue 4y ago
At the call site, unless it's a keyword argument, you don't know what the parameter is named. IDEs help with this of course, but it's not just the "2 options becomes 3 options" situation that recommends against them
298.
by habitue 4y ago
The consequence of cutesy errors is: the person seeing the error might not like cutesy stuff. The consequence of having a personality is: the person you're speaking with might not like your personality. Seems cut and dried!
299.
by habitue 4y ago
But they should.
300.
by habitue 4y ago
Ok, yes, people said both of these things, but `Sundar doesn't react fast enough` is the correct take. That's the side Google's institutions heavily skew it towards. In fact, the "it's no where close to perfect&quot

More ›