HNHacker News
TopNewBestAskShowJobs

mistercow

11,087 karma · joined May 9, 2009

submissionscomments
mistercow··on Judge dismisses majority of GitHub Copilot copyright claims
That’s an interesting rhetorical trick, because answering the question “what is an LLM” requires a lecture series, not a hacker news comment.
mistercow··on Judge dismisses majority of GitHub Copilot copyright claims
It’s trained on publicly available code, so what would be the point of that? If you’re looking to specifically infringe the copyright of code available on the open web, using an LLM code completion engine is just about the most roundabout and unreliable way to achieve that.
mistercow··on Bug squash: An underrated interview question
There is a vast distance between “accurate self-assessment” and “no idea what the interviewer is looking for”.

I’ve already explained why I think obscuring poor performance to preserve candidate confidence is crucial. If you think that’s a “terrible interview”, maybe you could elaborate on why, rather than just asserting it.

mistercow··on Bug squash: An underrated interview question
What an incredibly bad faith reply. Candidates should always be given an answer within a few days of finishing the interview. But it’s not good for a candidate to think they’ve bombed during the interview.

If someone fails my interview, it is completely possible that they’ll still get an offer. I have one data point, and my colleagues are going to collect more. I’ve changed my vote from no to yes in debriefs many times once I saw other feedback. But that’s a lot less likely to happen if my interview wrecks their confidence.

mistercow··on Bug squash: An underrated interview question
What you want is to prevent floundering death spirals. They need to understand the criteria, but if you need to hint and nudge them, they need to not feel like that means they’re bombing. Of course, you can absolutely do that in a bug squash interview, but my point is that it defeats this claimed advantage.
mistercow··on Bug squash: An underrated interview question
> It’s easy for the candidate to self-assess their own progress.⊕ When the candidate isn’t doing well on it, they probably already knew it without needing to be told as much by the recruiter. This is a much better candidate experience than the whiplash of thinking you solved a question perfectly only to realize that the interviewer was looking for something else entirely.

Not to say that I think this is a bad type of question overall, but IMO, this is an anti-feature. The candidate does not need to accurately self-assess their performance on the day. They need to have their confidence preserved so they don’t tilt and tank the signal for the entire interview round.

mistercow··on Yesterday's Pixels, Today
Varying pixel sizes, pixels offset from their natural grid, and even rotated pixels were actually not uncommon in retro games. Take a look at just about any SNES game that did advanced Mode 7 effects, like Pilot Wings or Super Mario Kart, and you'll see examples of pixels being rendered at varying scales outside their native resolution.

It's not the same look as modern games, because the screen resolution was much lower than what we have today, but to me, high resolution rendering of transformed low resolution sprites is a perfectly valid successor aesthetic. It's not the same as what a game would have looked like in the 90s, but as others have pointed out in this thread, neither are crisp full-color pixels rendered as perfect squares. It's all an interpretation.

On the other hand, what I cannot abide are "low poly" models that use polygons other than triangles and convex quads. I will die on this hill.

mistercow··on Toothpaste Null-Terminator
> When I find that the only tube of toothpaste left is that special, sentinel toothpaste, then I know it is time to buy more toothpaste. And then I have ample time to do so.

I might suggest phrasing this a little differently. Obviously, when you find that any tube is the only tube left, you know it’s time to buy more toothpaste. The problem is that you realize this early in the morning or right before bed, and so you’re likely to forget. The advantage of the “sentinel toothpaste” is that it continues to remind you after that moment, so that you’ll actually remember when you have time to do something about it.

And I think that captures the key thing about a sentinel value: it captures state in a place where you’re already storing state, without requiring you to wrap the state in some external structure, e.g. putting a sticky note on your mirror or storing a string’s length in a separate field.

mistercow··on "Frost crack" sounds may come from sky, not trees
Wouldn’t the tree explanation be refuted by simply recording the sounds in an open field, far away from trees?
mistercow··on I put a toaster in the dishwasher (2012)
> It's not a myth.

Yes it is. A tip: if someone tells you something is a myth, your next step should be to google it. This is intro level parallel circuit stuff.

Hell, just think about the circuitry in your house. If you have your phone charging on one outlet, and you plug a hair dryer into the adjacent outlet, does the phone stop charging? Does the hair dryer not run? Do you think those things have exactly equal resistance? Of course not. Current flows to both of them, but more current flows to the hair dryer.

> That is, by definition, NOT the "least resistance path".

I don’t think you know what “by definition” means. You could be the only path for the electricity to flow, and if there’s high enough resistance, it won’t hurt you.

Please stop repeating this nonsensical folk wisdom. It can get people killed.

mistercow··on All of Earth's water in a single sphere (2019)
Well, let me put it this way: that apparently little fresh water sphere? If the ISS passed over that spot, it would hit that.

This is the point: everything on the surface of the Earth is essentially two dimensional at this scale. Seeing a sphere with a 400 km radius placed on the Earth doesn’t give any better intuitions for the scale of water resources than simply reading the number “35 quintillion liters”. If you think it’s given you a better intuition, you’ve been misled.

mistercow··on I put a toaster in the dishwasher (2012)
This is a dangerous myth. Electricity takes all available paths, with less current flowing through the paths with greater resistance.

As far as I understand, the main reason people don’t get electrocuted even with improperly grounded shower head heaters is that (as noted in the post), tap water is generally not very conductive in the first place, so you’ve effectively got a gigantic resistor between you and the wires (assuming there’s a continuous path through the water at all). So even if the least resistance path is through you, the amount of current will be tiny due to Ohm’s law.

mistercow··on All of Earth's water in a single sphere (2019)
I think it’s important to keep in mind that if you did this same visualization on a planet with ten times Earth’s radius, but the same ocean depth and water distribution, then the water blobs would seem even smaller in comparison to the planet.

I’m just not sure it’s a particularly useful illustration to compare the volume of water on a planet to that of the planet itself.

mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
You can argue that having an extra number to juggle around is somehow less awkward because, under specific and subjective criteria, it’s “expected”, but given that the whole hook for tau is “we keep having to put a multiplier of two everywhere”, I don’t find it very compelling.

I can also make arguments that pi/2 would have been a better constant from a teaching perspective. The pi/2 version of the Euler identity, for example, would give you all the tools you need to link complex multiplication to rotation.

But at the end of the day, the choice of multiple used for the constant is a convention. None is going to be ideal in every case, and no math fundamentally changes because of a particular choice. Trying to argue for a change in convention at this point is just silly.

mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
The entire tau manifesto is basically an exercise in how you can come up with rationalizations for just about any aesthetic preference, and the “area of a circle section” is a perfect example of how far you can go with the gymnastics.
mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
No, like objectively, a dimensionless number lining up with a meaningful constant is more likely to be because of some underlying mathematical connection, and a dimensioned number lining up is more likely to be a coincidence. There are only a handful of ways for a unit’s heritage to have a connection to a local physical phenomenon like the post describes, and that’s what it takes to have a unit-dependent non-coincidence. That’s not dependent on your perspective.

The thing that’s interesting in this case is that the meter’s connection to g is obscured by history, whereas most of the time a unit’s heritage is well known. Nobody is going to be surprised by constants coming out of amps, ohms, and volts, for example, because we know that those units are defined to have a clean relationship.

mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
What are you disputing about the explanation given in the post? As far as I can tell, it’s basically accurate (although the pendulum unit was called the toise, and the meter seems to have targeted half a toise). If you accept that account, it’s not a coincidence.
mistercow··on Grace Hopper, Nvidia's Halfway APU
Even so, massive compute doesn’t necessarily mean GPU-friendly compute. We could see a breakthrough in analog or neuromorphic hardware, for example, where Nvidia isn’t well positioned. Or we could see a training breakthrough which is far more efficient, but bottlenecked on single core performance, or just branch-heavy performance. You can imagine scenarios like that where GPUs still play a role, but where even today’s top of the line GPUs are way over the top compared to the CPU bottleneck.

If one of those scenarios happens, maybe Nvidia can pivot, or if we see analog take over, we could see something really bizarre like a dark horse like Seagate taking over by pivoting from SSDs, just because their manufacturing pipeline is more compatible.

mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
I’m saying that the single SI length unit could have been defined precisely as the light nanosecond, or “metric foot”, had people known that that length fit closely to an existing unit back around 1790.

There would still be one unit with prefixes added, but that unit would have a really clean correspondence to physics rather than a hacky conversion factor.

But you have to go back that far in time for it to work, because it’s a fraction of a percent off of the current standard foot. They were happy to make those kinds of changes (as in the case of defining the meter to be ~0.51 toises) back when all of the existing measurements were pretty imprecise to begin with.

Of course, that’s why it could never have worked out this way. By the time we could measure a light nanosecond, we were already committed to defining units very closely to their existing usage.

mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
I seriously doubt you could define any system of units that has zero coincidences, even with significant computational effort. Some things in the real world are just going to happen to line up close to round numbers, or important mathematical constants, or powers or roots of mathematical constants, and then you’ll have some coincidences.

There are just too many physical quantities we find significant, and too many ways to mix numbers together to make expressions that look notable.

mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
As far as I can tell, the most recent experiment to measure the mass of the Earth by 1790, when they decided on the definition of the meter, was the 1772 Schiehallion experiment, which gave a value 20% below the actual value. So if pi^2 were to somehow fall out of that it would likely be so far off as to be unrecognizable.

But even that doesn’t matter, because the mass of the Earth didn’t play a direct role in the definition of the meter. If you take out the whole thing about the meter’s definition targeting half a toise, then all you have is “related to the circumference of the Earth”, and it would be a monumental coincidence if the mass of the earth and gravitational constant just conspired to somehow drop an unadulterated pi^2 out of the math.

mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
I really wish we had known this back before it was way too late to seriously change our units around. It would mean that our SI length units wouldn't have to have some absolutely ridiculous denominator to derive them from physical constants, and also the term "metric foot" is pretty fun.
mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
Eh, you can find plenty of cases where tau is just as awkward as pi is elsewhere. Right off the bat, the area of a circle becomes more awkward with tau, becoming (tau*r^2)/2, and in general, the volume of an n-ball gains weird powers and roots of two in its denominator as n increases if you switch to tau. In general, I don't think you can really claim either one is "more fundamental". It's just a matter of framing.
mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
The problem with that is that writers are not consistent with comma usage either, particularly when it comes to informal writing, where prescriptive rules are out the window anyway. And I would argue that it would be a bit of a norm violation even in informal writing to introduce this new point at the end of a paragraph rather than starting a new one, which makes me think that that was not the author's intent.
mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
I wonder if this is related, but imperial measurements with a 5 in the numerator (and a power of two in the denominator) are generally just under a power of two number of millimeters.

The reason is fun, and as far as I know, historically unintentional. To convert from 5/(2^n) inches to mm, we multiply by 25.4 mm/in. So we get 5*25.4/(2^n) mm, or 127/(2^n) mm. This is just under (2^7)/(2^n) mm, which simplifies to 2^(7 - n) mm.

This is actually super handy if you're a maker in North America, and you want to use metric in CAD, but source local hardware. Stock up on 5/16" and 5/8" bolts, and just slap 8 mm and 16 mm holes in your designs, and your bolts will fit with just a little bit of slop.

mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
That analogy is so bizarre that I have no idea how to respond to it.
mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
Huh, interesting point. Writing unambiguously is ridiculously hard.
mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
Right, that connection is not a coincidence. The connection the previous commenter drew between the meter, pi, and the circumference of the earth is a coincidence.
mistercow··on Grace Hopper, Nvidia's Halfway APU
We might, but it’s also plausible that it would change the ecosystem so much that centralized models are no longer so prominent. For example, suppose that with much cheaper training, most training is on your specific data and behaviors so that you have a model (or ensemble of models) tailored to your own needs. You still need a foundation model, but those are much smaller so that they can run on device, so even with overparameterization and distillation, the training costs are orders of magnitude smaller.

Or, in the small business case (mind you, “long term” for tech reaching small businesses is looooong), these businesses again need much smaller models because a) they don’t need a model well versed in Shakespeare and multi variable calculus, and b) they want inference to be as low cost as possible.

These are just scenarios off the top of my head. The broader point is that a dramatic drop in training cost is a wildcard whose effects are really hard to predict.

mistercow··on A wonderful coincidence or an expected connection: why π² ≈ g
> Or after-atmosphere insolation being somewhat on average 1kw/m2.

I’m kind of inclined to say that this one isn’t so much of a coincidence as it is another implicit “unit” in the form of a rule of thumb. Peak insolation is so variable that giving a precise value isn’t really useful; you’re going to be using that in rough calculations anyway, so we might as well have a “unit” which cancels nicely. The only thing that’s missing is a catchy name for the derived unit. I propose “solatrons”.

← PreviousPage 6 of 34Next →