sfink

24 ms·

361.
by sfink 1y ago
I have mixed feelings about AI, but love the posts and papers that dig into how they work. Except, as this post shows, I seem to vastly prefer Anthropic's posts to OpenAI's. > Claim: Hallucinations are inevitable. > Finding:
362.
by sfink 1y ago
If they made it as far as being able to lose something they were working on, then it's less likely to have been a missing library problem. But I don't know what it was; people do successfully run Firefox on 32-bit Linux. Firefox d
363.
by sfink 1y ago
You still need matter to carry the heat, and matter is heavy (by definition!) Radiating the heat isn't going to be enough if you want to send anything other than a blob of plasma. Which is an interesting thought when considered as a we
364.
by sfink 1y ago
No, you have to zap them early. Lasers still spread out (see "beam divergence"), so you have to impart most of the momentum while they're still close enough for your lasers to hit with enough power. Sadly, that also means you
365.
by sfink 1y ago
Not to mention that I'm not sure if all photons are guaranteed to hit anything. I have no intuition for calculating how likely that is for things in our neighborhood, but if you think of the star that is closest to the edge of the univ
366.
by sfink 1y ago
Yep. From a light emitter's perpective, it is directly embedded in all of the places surrounding it that its light would eventually reach. Your eyeball, a distant dust spec 4 million light years away, and a black hole are all directly
367.
by sfink 1y ago
Heh, sorry. The `-A` part went whooshing over my head until I had already written up the rest. I just saw `jj new` and got triggered into a "well akshually..."
368.
by sfink 1y ago
Oh, I missed this part. I think jj is better here in at least one scenario. Specifically, I believe the scenario you're talking about is: change file1 build, producing binary.out squash the change down (leaving your wor
369.
by sfink 1y ago
> Complicated workflows requiring deep understanding might be harder to learn because of a higher abstraction. That's a fair complaint, and where other git porcelains have failed in the past. I think the trick is that jj isn't
370.
by sfink 1y ago
I'm mostly git -> mercurial -> sapling -> jj. For me, Mercurial -> Sapling was a mostly lateral, slightly upward move. Sapling -> jj felt really good, though oddly not because anything felt horribly wrong or missing in S
371.
by sfink 1y ago
You can have most of that stuff, but it doesn't sound like that would sway you anyway. (And those 4 missing features may truly be showstoppers for you, in which case you're absolutely correct to stick with git.) There are early ad
372.
by sfink 1y ago
Technically, it's z-k (z=0, k=15). It makes for a nice root change ID: zzzzzzzz. Less choking than kkkkkkkk. The relevance, btw, is that you can use git commit IDs in place of change IDs anywhere and they won't conflict; jj knows
373.
by sfink 1y ago
> Would it be accurate to describe Jujutsu as "a Mercurial-inspired frontend for Git"? In my opinion: yes. (Maybe with a slight caveat that I think some of the Mercurial inspiration came through Sapling.) And it's not limi
374.
by sfink 1y ago
My personal opinion: `jj new` is better because it's non-modal. If you use `jj edit`, you're sort of switching to "edit mode": any change you make will trigger a rebase of all descendants.[1] You're live-mutating th
375.
by sfink 1y ago
Yes, and this is the most commonly used workflow these days with jj (the "squash workflow"). You have a top commit, which is also your working directory, and you make changes freely. To "stage" something, you squash it d
376.
by sfink 1y ago
> Sure, we may see many lacking fundamentals, but I suppose that isn't so different from the criticism I heard when I wrote most of my first web software in PHP. It's not just the fundamentals, though you're right that is
377.
by sfink 1y ago
> We are headed in a direction where written code is no longer a time sink. Written code has never been a time sink. The actual time that software developers have spent actually writing code has always been a very low percentage of total
378.
by sfink 1y ago
Apologies for arguing from first principles, but for anything that spurs a lot of activity, the only alternative to a bubble is this: people ramp up investment and activity and enthusiasm only as much as the underlying thing can handle, the
379.
by sfink 1y ago
Ironically, this sort of thing could have the side effect of turning web page usability issues into security issues. If your web page is a pain to use, then your users will rely on insecure automation to avoid dealing with it. I'd like
380.
by sfink 1y ago
I think something like cargo vet is the way forward: https://mozilla.github.io/cargo-vet/ Yes, it's a ton of overhead, and an equivalent will be needed for every language ecosystem. The internet was great too, bef
381.
by sfink 1y ago
Carriage returns are good enough for progress bars, and seem to work fine in my emacs shell at least: % echo -n "loading..."; sleep 1; echo -en "\rABORT ABORT"; sleep 1; echo -e "\rTerminated" works fi
382.
by sfink 1y ago
That's how it happens in OSS development. For proprietary development, it happens because one big account depends on some obscure feature, or because it is frowned upon to interfere with getting people to upgrade to the latest paid ver
383.
by sfink 1y ago
I solved it on my dog walk as a recurrence. But to move the spoiler down a little, my initial solution was to pour all the bottles out into a barrel, mix them together, and refill the bottles. Nothing explodes since you never measure anythi
384.
by sfink 1y ago
I don't see a lot of value in generic code reviewers. I want the reviewers to be actively engaged in writing somewhat related code themselves, otherwise the value of their opinions will decline over time. As for prioritization... isn&#
385.
by sfink 1y ago
> I think the article underestimates how much intent can be grasped from code alone. Even without comments. I agree, the human thought process always ends up getting embedded in a which of several possible ways any one thing might be don
386.
by sfink 1y ago
The author was just saying that this will become the norm, not the exception. Ok, not this bad -- we can at least expect to have access to the AI-generated code for at least a little while longer. (I can imagine a future where AI programm
387.
by sfink 1y ago
> Vector embeddings are lossy encodings of documents roughly in the same way a SHA256 hash is a lossy encoding. Incorrect. With a hash, I need to have the identical input to know whether it matches. If I'm one bit off, I get no info
388.
by sfink 1y ago
Awesome visualizations. The part that I was expecting to see but didn't: how can you move at a constant speed? For the original purpose of positioning objects along a path, it doesn't matter. But when moving, you can see it's
389.
by sfink 1y ago
> What if "increasing intelligence", which is a very vague goal, has diminishing returns, making recursive self-improvement incredibly slow? This is sort of what I subscribe to as the main limiting factor, though I'd descr
390.
by sfink 1y ago
It's a subset of profile-guided optimization, but I fairly frequently will gather summary histograms from more or less representative cases and then use those to validate my assumptions. (And once in a while, write a benchmark that pro

More ›