HNHacker News
TopNewBestAskShowJobs

ladberg

3,485 karma · joined August 10, 2017

Contact me at hackernews@leoadberg.com
submissionscomments
ladberg··on Getting a job at Apple without going to college or doing LeetCode
I think "unique among big tech companies" and "used to be the norm" can both be true!

It definitely has pros and cons though: I've worked on a handful of teams at Apple and was very lucky to have been able to work on super fun and interesting stuff on all of them, but an outsider confronted with 1000+ vague listings on jobs.apple.com is effectively playing the lottery of whether they'll get to interview for a team that's a good match for them.

ladberg··on How Is LLaMa.cpp Possible?
You're thinking about the wrong bandwidth here. The article is talking about going from the GPU's RAM <-> GPU cores (i.e. through load/store instructions in a cuda kernel), not from CPU's RAM <-> GPU's RAM. That kind of bandwidth is still important but usually not the bottleneck on most ML workloads.
ladberg··on My (Herb Sutter's) C++ Now 2023 talk is online: “A TypeScript for C++”
I kinda feel like this is DOA until modules are more widespread. Any new cppfront code in an existing C++ codebase can't include any existing header files which automatically excludes it from any C++ codebase I've worked on (or any I expect to in the next few years) without a significant amount of developer time spent making the migration to modules-only.

Would love to try it out though when that happens, I do hope that modules will eventually get mainstream enough to make cppfront viable.

ladberg··on Automatically classifying the content of sound files using ML
I think they were hoping that the off-the-shelf model the author used was a novel technique trained on unlabeled data, not a dataset labeled by humans.
ladberg··on What If All the Gasoline Cars Became Electric Vehicles?
Define city speeds? Because this is pretty objectively false at the speeds cars usually drive in all the cities I've lived in. Even the pdf you linked has an EV being recorded louder than an ICE at only a single speed out of all the different speeds and cars they chose.
ladberg··on Speed Up C++ Compilation
Would recommend Build Bench to help gain a mental model for how different styles of code impact build times: https://build-bench.com/
ladberg··on Ask HN: What are some reasonable black swan tech predictions for 2023?
Eh, CO2 is a pretty low energy molecule so doing anything with it usually takes energy input. I'd believe in a ton of other wacky sci-fi ideas turning into reality before that one.

That said, I'm hopeful we could use cheap energy (e.g. the sun) to do something useful with carbon in the air, like plants do. See Running Tide as an example, they're trying to sequester carbon in a completely hands-off way using no added energy input other than the sun.

ladberg··on Pixar, Adobe, Apple, Autodesk, and Nvidia form alliance for OpenUSD
Yes, but using .reality files is better for stuff like games in RealityKit. For example it would contain a GPU-ready texture instead of a jpeg (and a lot of other differences), so it's more in-line with how assets are typically loaded in games.
ladberg··on Accessing Llama 2 from the command-line with the LLM-replicate plugin
Idk, to me this reads pretty clearly remote (for this model in particularly):

> My LLM tool provides command-line access to a wide variety of language models, both via web APIs and self-hosted on your own machine ... The brand new llm-replicate plugin provides CLI access to models hosted on Replicate, and this morning a16z-infra released a16z-infra/llama13b-v2-chat which provides Replicate API access to the new Llama 2 13B chat model.

I'm also unsure what this would have to do with the license, when you install any other software do you expect the license to tell you whether processing happens locally or in the cloud?

ladberg··on {n} times faster than C
C doesn't even originally have true/false, I think you may be conflating the two concepts that "any nonzero int is truthy" and "boolean expressions evaluate to ints". The standard mandates that boolean expressions like equality always evaluate to 0/1.
ladberg··on $900k Median Package for Engineers at OpenAI
I feel like person you're replying to specifically implied that OpenAI does it differently in that aspect, and they were also high up at Netflix so is likely very familiar :)
ladberg··on Developer tools to create spatial experiences for Apple Vision Pro
> Eyes are not required for proprioception.

No, but they will mess with and override proprioception. I've used the Vision Pro and the first time you try to take a drink with it you'll probably miss your mouth. I think the main reason is that you see the cup coming towards your slightly offset "eyes" and your body automatically adjusts to match.

I think if I had closed my eyes (I didn't try that) I would have done better.

ladberg··on Do you know how much your computer can do in a second? (2015)
The 68k is per millisecond, which lines up with 68M per second.
ladberg··on Apple Vision Pro – Why It May Be Lousy for Watching Movies on a Plane
Does anyone actually care about their looks on a plane? I can't imagine prioritizing what other plane passengers think of me over my personal comfort and ability to get through sitting in a cramped seat all day...
ladberg··on Humans aren’t mentally ready for an AI-saturated ‘post-truth world’
Bleecker and 7th Ave

For comparison: I'm trying to beat the numerous dollar pizza and food carts nearby, not normal "fast food" like the more expensive Five Guys on that intersection.

ladberg··on Humans aren’t mentally ready for an AI-saturated ‘post-truth world’
Not sure if you're being facetious or just don't understand the reality of living in NYC...

I can probably walk to a dozen different big grocery stores in 15m and they're ALL more expensive than the cheap fast food in the same area. Not including the smaller expensive bodegas where you can pick up stuff 24/7 every block (kinda like the equivalent of a gas station). A half gallon of milk is $4 at any of the big stores and even more at a smaller place.

Anything cheaper requires a subway ride, which adds more walking and is annoying to do with multiple grocery bags, not to mention adding a flat ~$5 additional cost.

ladberg··on Humans aren’t mentally ready for an AI-saturated ‘post-truth world’
I don't mean the cost of my time that work pays me, just how much money I'd personally pay to avoid doing something something I don't like (schlepping grocery bags on the subway, doing dishes).

In general cooking is the fun part and that's what makes it a treat, not the rest of it.

Some prepared food within walking distance most definitely is cheaper than all unprepared food within walking distance. It works out just because the places I can walk to for groceries are incredibly overpriced and the restaurants obviously don't source their food there.

ladberg··on Humans aren’t mentally ready for an AI-saturated ‘post-truth world’
> There are even people so confused about food prices that they make claims that fast food is cheaper than groceries.

I live in an expensive part of NYC and have to go decently far out (by subway, I don't have a car) to find groceries that are cheaper than local fast food unless I want to eat mostly rice and beans.

Add in the costs of my time to shop, transport groceries, cook, and clean, it's significantly cheaper to eat out most of the time. Even subtracting the one task I actually enjoy (cooking), it's still not worth it most of the time.

The result is that cooking in becomes our "treat" that we do a few times a week and we end up buying the more expensive ingredients within walking distance.

ladberg··on Facebook almost acquired Waze, but we ended up with Google
Yep, I pretty much exclusively use Apple Maps in the US and it's great but when I'm on vacation it's often unusable.

Most recently, I was in Georgia (country) and navigation of any kind is nonexistent. Like often times I would look something up and want walking directions there but it simply isn't available. Google Maps worked great.

ladberg··on Goodbye, San Francisco
> Not sure if all this is true, but I did not feel like finding out.

It is most definitely not true

ladberg··on Stackoverflow is investing into baking GenAI
In my opinion it's significantly more meaningful than blockchain hype. Blockchain doesn't add any value whatsoever for 99% of the use-cases that companies were advertising it for, but AI can immediately and cheaply create output that previously would have been expensive, time-consuming, or impossible.
ladberg··on Karl Guttag on Apple Vision Pro (Part 1)
I don't think web apps like VSCode will be at a disadvantage. If they're doing "normal" web rendering and not rolling their own canvas renderer like Google Docs then it should just work fine.
ladberg··on Karl Guttag on Apple Vision Pro (Part 1)
Foveated rendering is supported by default. Most apps will rely on the OS's renderer and not their own so they don't have to do anything with eye tracking to have it enabled.

EDIT: See cma's comment below which adds more info: apps that use their own render still support foveated rendering too.

ladberg··on Karl Guttag on Apple Vision Pro (Part 1)
You can do VSCode Remote in a browser on an iPad so I'm not too worried.

I think the bigger factor is that Apple is trying to push this as more of a workplace productivity machine than the iPad, so third party developers will hopefully be more amenable to porting those types of apps than they were to iPadOS.

ladberg··on How some common material imbalances affect your win-rate
I mean I agree that's the ideal but not necessarily true in practice. I bet if you took two random players of the same Elo and pit them against each other many times, it would not be 50/50.
ladberg··on How some common material imbalances affect your win-rate
The probability of player A being up a rook over player B is higher when player A is better than player B, even if their Elo is the same. So the author has picked scenarios where player A is already more likely to be better than player B but assumed they were of equal skill, which will cause the results to be biased.

If I have time later I'll try to whip up a small coin-tossing demo in Python that demonstrates what I mean, I'm definitely not the best explainer!

ladberg··on How some common material imbalances affect your win-rate
Yeah I guess my claim is that two people of the same Elo on lichess are likely to not be perfectly equal. If the average winrate of any random two players with the same Elo is something like 60/40 (random speculation) then you'd bias the results in the article enough to create the kind of graphs shown.

The problem is that this doesn't really average out with more samples.

ladberg··on Jim Keller on AI, RISC-V, Tenstorrent’s Move to Edge IP
That's not exactly my takeaway, e.g. he says this in that interview which is pretty consistent:

> So if I was just going to say if I want to build a computer really fast today, and I want it to go fast, RISC-V is the easiest one to choose. It’s the simplest one, it has got all the right features, it has got the right top eight instructions that you actually need to optimize for, and it doesn't have too much junk.

ladberg··on How some common material imbalances affect your win-rate
Am I misunderstanding something or is there a glaring statistical error here? I think this analysis works if the matches are all between two players of truly identical skill that will roughly win 50% of the time against the other, but that's not really the case here. They attempt to do that by controlling for Elo but I'm decently sure that you could pick any two players of the same Elo on lichess and one will win the majority of games.

Now if we assume there is often a slightly better player in these games, the better player will more likely get into an advantageous position early in the game and win more often, but not entirely because they got into the more advantageous position. What I'm trying to say is that someone who is up a rook will win partially because they're up a rook and partially because they are likely a better player in the first place and will continue to play better than their opponent.

I think to do this study correctly you'd need to place new players in a random position drawn from the dataset and actually evaluate the win rate without the confounding factor of having gotten into that position in the first place.

ladberg··on Apollo’s Christian Selig explains his fight with Reddit – and why users revolted
The straight answer is that ads are embedded in the "HTTP API" and the vast majority of browser users don't have adblock, and Reddit can't force ads on any API users (except through the official app).
← PreviousPage 5 of 24Next →