HNHacker News
TopNewBestAskShowJobs

ngriffiths

336 karma · joined October 16, 2024

Twitter: @ng_griff

Website: griffens.net

submissionscomments
ngriffiths··on Tour de France confronts a new threat: Are cyclists using tiny motors?
Betteridge's Law strikes again
ngriffiths··on Facts don't change minds, structure does
In practice I think people often don't see the full structure of their own belief graph. Parts of it are clear but for 99% of important issues, it's more fuzzy than portrayed in the figures here. I still think this is an illuminating way of looking at it!

Another major factor is that while the graph may be fuzzy, the people we trust are clear. Only those people are allowed to "fill in" the missing pieces, and I think it takes a lot of work to do that, so it totally makes sense.

If the takeaway is "don't expect conflicting facts to convince your audience" I agree with that, but the reason is they don't trust you, not the conflicting graphs, and the trust is not really a consequence of the graph structure.

(Also, I was writing about similar stuff recently here: https://blog.griffens.net/blog/no-one-reads-page-28/)

ngriffiths··on The Krull dimension of the semiring of natural numbers is equal to 2
Interesting, makes me curious about geometric ways of looking at semirings. Krull dimension is an algebraic way of capturing the dimension of corresponding geometric objects, so is there some way of doing that with semirings? Or any more intuitive reason why we'd get dimension 2 here? The papers I found in a quick search are way over my head.
ngriffiths··on Could a Paper Plane Thrown from the ISS Survive the Flight?
I feel like the news article doesn't emphasize that you aren't "launching" the plane any more than you would by setting it on the roof. So it's really surprising that it all happens in <4 days.

I thought atmospheric effects were much lower at that altitude, but apparently even the ISS loses about 3km every month (enough to deorbit in ~15 months).[1]

The ballistic coefficient of the ISS is a whopping >500 times greater than the plane so the plane drops really fast.

[1]: https://space.stackexchange.com/questions/9482/how-long-woul...

ngriffiths··on Is the doc bot docs, or not?
The doc bot goes in the same category as asking a human who has read the docs. In order of helpfulness you could get:

- "Oh yeah just write this," except the person is not an expert and it's either wrong or not idiomatic

- An answer that is reliably correct enough of the time

- An answer in the form "read this page" or quotes the docs

The last one is so much better because it directly solves the problem, which is fundamentally a search problem. And it places the responsibility for accuracy where it belongs (on the written docs).

ngriffiths··on Hilbert's sixth problem: derivation of fluid equations via Boltzmann's theory
John Baez wrote a Mastodon thread on this paper here:

https://mathstodon.xyz/@johncarlosbaez/114618637031193532

He references a posted comment by Shan Gao[^1] and writes that the problem still seems open, even if this is some good work.

[^1]: https://arxiv.org/abs/2504.06297

ngriffiths··on Engineered Addictions
I think in this context "competing" means having a meaningful market share, which would help reduce time the world spends on the alternative useless gamified/addictive apps
ngriffiths··on Engineered Addictions
I agree. I see this becoming a bigger and bigger problem unless someone steps in with significant regulation or major changes like the article says.

The other challenge is the regulation part is much easier when the product is, say, heroin. Algorithms are technically complex (hard for policymakers to grasp), flexible (can be tweaked to work around guidelines?), and operating in the digital world (harder to monitor/block).

Maybe a major factor here is social acceptance vs stigma. In the future will it be considered extremely weird and antisocial to be on your phone nonstop?

ngriffiths··on Writing toy software is a joy
Actually, someone should design one that can pull in quotes. Like as a separate tool the LLM uses to quote that has guarantees, it's just a copy and can't be hallucinated. Then you could see the primary info when it's needed/asked for, similar to articles
ngriffiths··on Writing toy software is a joy
I agree and am perfectly happy using it as fuzzier web search, because it works really well for me.

Finding references is often my main goal, but other times I just want some quick facts, in which case I'll be as thorough with checking as I would when reading a random blog with spelling errors.

ngriffiths··on Writing toy software is a joy
I don't think anyone can make a Google today that works as well as it did back then. Google shaped how new content was created, and that was probably a much bigger deal than any changes to the tool itself
ngriffiths··on Writing toy software is a joy
Hopefully not that fast, but I'm pessimistic. The cost of the human bloat will far surpass the current cost of hallucinations. And like we saw with Google, that bloat can become a feature of the content itself, not just contained in the tool.
ngriffiths··on Writing toy software is a joy
I use it the same way. The feeling is that I'm back in ~2010 when Googling stuff felt like a superpower. I could find anything back then.

Of course, it didn't last long, and trying to Google now is an exercise in pain and frustration. Lots of people have complained about the various things Google and marketers have done to get there, idk, I just don't like how it works now.

Top LLMs feel amazingly good at rapidly surfacing info online, and as I go through the references they're usually pretty good. I guess the same forces as before will apply, and there will be some window of opportunity before it all goes away again.

ngriffiths··on An election forecast that’s 50-50 is not “giving up”
Oh you're right, it is basically the same, the undervalued stock is the one to buy. I guess I was thinking about another difference which is that the value of stocks seems more vibes-based and a really stupid product (from the investor's perspective) can sometimes become popular. In prediction markets the outcome is more concrete.
ngriffiths··on An election forecast that’s 50-50 is not “giving up”
Yes but unlike with stocks, you don't want to agree with them: if you think people overrate the chances of an event, you'd against it.
ngriffiths··on An election forecast that’s 50-50 is not “giving up”
It matters a lot what the price is of "putin loses." If Putin does in fact lose 5% of the time, but the odds are 3%, you should bet on it. It may feel silly to do that in this instance but consider what happens if you regularly bet in prediction markets. Always betting on the "putin wins" outcome loses money in the long run.
ngriffiths··on An election forecast that’s 50-50 is not “giving up”
I don't think it's that simple - if other people find out English bookies overrate England, then they make a profit by betting against. I'm not saying every market is perfectly efficient but it would be surprising if a major inefficiency lasted for a while. And if "enough" bets are placed the set odds will always be an unbiased estimate of the true odds.
ngriffiths··on How did places like Bell Labs know how to ask the right questions? (2023)
> The one hard and fast rule Labs did seem to have was that you could not say no to any request for help from any of the applied folks — or other researchers for that matter.

Nice example of a simple policy that changes the culture in exactly the way you want it to. The university researcher types want to say "ugh, stop interrupting me," but when that's not an option everybody keeps each other on track.

ngriffiths··on Markov Chains Explained Visually (2014)
Markov chains are super useful in statistics but it isn't obvious at first what problem they solve and how - some further reading that I found helpful:

https://twiecki.io/blog/2015/11/10/mcmc-sampling/

Note that the point of the markov chain is it's possible to compute relative probabilities between two given points in the posterior even when you don't have a closed form expression for the posterior.

Also, the reason behind separating the proposal distribution and the acceptance probability is that it's a convenient method to make the Markov process stationary, which isn't true in general. (Wikipedia page on MCMC is also useful here).

ngriffiths··on A data analysis of speeches at the Oscars
Very cool! I think mentions per word could be a good metric for some of these, otherwise a main takeaway is just "everyone crams in more stuff now."
ngriffiths··on Blunder Free Chess – visualize which squares are attacked
I have, and it's like magic to me. I have no idea how it works!
ngriffiths··on Blunder Free Chess – visualize which squares are attacked
Yeah I've heard this description too, and it makes sense. But I also get the sense that the "chunks" must not just be simplifications of parts of the board, in fact the experts must "see more at once" than the newbie right? Like inside of the chunks there is somehow tons of intricate detail that the newbies can't even see, let alone assimilate into one memory or instinct or thought.

Maybe like a tourist who notices one interesting building or sign. The local doesn't notice anything, just like "cool, I remember this place," except then they notice if one small detail changes compared to the last time they visited. So in a sense they "saw" way more than the tourist.

I am also terrible at chess. Curious if this is anything like how the experts experience it.

ngriffiths··on I Went to SQL Injection Court
> Congratulations! You now understand databases.

Data engineering: doing a lot of fancy work to make a very simple product

ngriffiths··on Blunder Free Chess – visualize which squares are attacked
Makes me wonder how good players "see" the board. Like what parts are "highlighted" and is it always changing as you ponder different aspects of the game, or more static?
ngriffiths··on Introduction to Stochastic Calculus
This is such a good model for how to write a beginner friendly introduction. Especially the motivation for the Ito lemma, with the dW^2 term remaining important even though it disappears in regular calculus, and the conversion to Stratonovich is really nice.
ngriffiths··on Despite sticker prices, the real cost of getting a degree has been going down
This is super unexpected to me. Sort of tangential but it reminds me of healthcare costs discussions, where orgs like National Nurses United will point to rising sticker prices and say that health systems are screwing us all over. But it's more important to understand actual spending trends and sticker price is just not a great metric to have that discussion.
ngriffiths··on Please Commit More Blatant Academic Fraud (2021)
I had two experiences at the polar opposites of the spectrum - one research team I worked on had very high standards and was comfortable being patient for material that had value. The other involved an approach that obviously stood no chance to be useful to anyone.

Some differences:

- The first one was in a space with more low hanging fruit

- The first one was after large effect sizes, not the kind where you can massage the statistical model

- The second one was a topic with far higher public interest

- The second one was primarily an analytic project, whereas the first one was primarily experimental

I feel like bad science lives in the middle of a spectrum - you have young fields/subfields with boring but impressive experimental breakthroughs, and on the other end you have highly political questions that have been argued to death without resolution. Bad science is about borrowing some of the strategies used in politics because all the important experiments have already been done.

ngriffiths··on Does X cause Y? An in-depth evidence review (2021)
And all this before you even get to "how much of an impact of X on Y should there be before it is even close to a bottleneck that's worth actually acting on, and do we think it reaches that threshold?"
ngriffiths··on Basis of the Kalman Filter [pdf]
Thanks for sharing this - saving this paper too (from link in the github page): https://people.ee.ethz.ch/~loeliger/localpapers/FactorGraphs...

I'm not familiar with these techniques at all but seems like they have a ton of useful applications.

ngriffiths··on Visualizing data is an art
> And creating art is about more than just precision. It's about evoking emotion, telling stories, and sparking curiosity. It's about making people think and feel.

For sure, the best artists are the ones who take their whole perception of stuff, with the feelings and everything else, and put it directly in someone else's brain. I would maybe say that some data visualization is supposed to do that, particularly data journalism, but definitely not all data visualization.

I think not all data visualization is art in the same sense that not all writing is literature, and in fact almost all isn't, because it's more like "I am in a shared info network with my boss/colleagues/clients/readers, and we need to communicate as rapidly and directly as possible, so we've agreed on a shared visual language that is higher bandwidth." Emails are slightly less exciting to read than a novel but like a million times faster to write per word.

Which is kind of sad. It would be fun to hone that skill and there is no room for that at my job.

(Side note, it seems like everyone always says that the pie chart problem is about lengths vs angles... but isn't it just that one chart has a shared baseline and the other doesn't? Taking the middle out [donut chart] to emphasize arc length doesn't actually solve the problem. Conversely, a stacked bar is more like a pie chart than a bar chart, except with a weird aspect ratio if you only have one stacked bar).

← PreviousPage 4 of 5Next →