evouga

Sure but... this is the technology at the most expensive it will ever be. I'm impressed that o3 was able to achieve such high performance at all, and am not too pessimistic about costs decreasing over time.

It's funny; I spent a couple of hours last week helping some students debug out-of-bounds indices in their Rust code.

I've written bugs that would have been caught by the compiler in a memory-safe language. I think the last time was maybe in 2012 or 2013? I still write plenty of bugs today but they're almost all logic errors that nothing (short of AI tools) could have caught.

I think it's simply that the blog author and commentators have an unrealistic threat model when it comes to how the legal profession uses MD5s.

After the first high-profile case where authenticity of evidence gets called into question because a seized electronic document was deliberately doctored to allow for a hash collision (if that ever happens), there will be a will to change to something new.

Who picks the reviewers and nags them to complete their reviews? This is the principal actual useful work journals do these days.

The situation is now counterintuitive in the other direction: if Monty Hall had opened those 48 doors at random and they just happened to not contain the car, then there is no advantage to switching, though many people would insist otherwise.

It's not so easy for several reasons:

1. It's easy enough for a lone entrepreneur with no investors to make principled long-term decisions. It's not so trivial when you owe a fiduciary duty to a board of venture capitalists.

2. Your team of engineers (which has grown steadily as you've scaled up) have built The Thing and everyone loves it. Now what? You only need 5% of the team to maintain the software. You could fire 95% of the team, which will make your company mighty unpopular to future hires, and moreover your best developers won't want to stay and do maintenance for the next decade. Easier to have them work on gratuitous frontend redesigns, bloated features that increase engagement metrics, etc. In a restaurant the contractors who build and furnish the place aren't your employees.

I've spent hours helping my students debug their Rust programs of incorrect array-indexing logic, incorrectly-reasoned loop invariants, incorrect conditional logic, and everything in between. They're not double-deleting raw pointers, but then again, most C++ programs don't use raw pointers these days either...

"Rust makes a few classes of bugs impossible to write by construction" is a very nice feature of the language. But I'm really turned off by how insufferable the Rust community is (which shines through in the OP's blog post and several comments elsewhere in the responses): it's never the quote at the start of this paragraph; it's always "writing software in anything but Rust is fundamentally irresponsible!" or "if a Rust program compiles, it must be correct!" No, not even close.

The store may or may not be telling the truth. Either way, it's undeniable that SF has changed dramatically over the last several years, with dramatic upswing in property crime and complete unwillingness on the part of law enforcement to fight it.

Because after your lean, highly-productive startup team creates the app that everyone loves, you get a bunch of funding and hire thousands of extraneous software developers and then have to find something for them to do.

How much of the cooling cost of current chips is due to wires, vs. the transistors themselves?

Superconductors will not help with the latter.

“Most important one in physics this century”: I’m hoping the room-temperature superconductor pans out, but even if it does, the century is young!

This time in the 20th century, relativity was “obviously” the most important discovery of the century, with the new theory of quantum mechanics right around the corner.

Arxiv has its own problems.

You can find dozens of proofs on the arxiv that P=NP and that P!=NP. What does peer “review” look like here? Most people won’t bother to write an article specifically rebutting one random incorrect proof.

Now somebody tomorrow posts a proof that P!=NP. Will people pay attention? If it’s a “big name” or somebody at a university with a good PR office then yes; otherwise no. Doesn’t seem ideal to me.

What I really like about this explanation is that it highlights the fact that entropy is not a natural property of the physics system: entropy is only defined with respect to some coarse-graining operation applied by an imperfect observer of the system. So as Sabine points out it seems we should really be talking about multiple different entropies, each of which corresponds to a different mechanism for coarse-graining microstates into macrostates, with each different entropy changing at different rates depending on the coarse-graining mechanism and physical system. (And in particular, God observing the universe would not see entropy change at all; even if there were uncertainty in the initial conditions of the universe, God would see that uncertainty perfectly propagated with no loss of information, in a way made precise by Liouville's Theorem.)

But even this is not the full story, because I can take a mass-spring network, and no matter how I choose to coarse-grain it, I will not see the entropy corresponding to that coarse-graining increase, because the trajectory of a mass-spring system is periodic. Entropy increase requires that the system is ergodic with respect to the chosen coarse-graining operation, i.e. that over long times the trajectory visits the coarse-grained states in a "random" and uniform way. It's not at all obvious to me why the dynamics of particles bouncing around in a box have this property, and particles attached in a mass-spring network do not; and neither the Sabine nor the Veritaserum videos address this or why we should expect all practical real-world physical systems to be ergotic with respect to practical coarse-graining mechanisms.

Well since circa mid-2010s the Internet is apps. Or app-like centralized web sites (like Hacker News).

People want to install a single “Lemmy” app on their phone that shows them all Lemmy content and lets them post to all instances, using a single account and a *simple* user interface. If Lemmy cannot provide that user experience (abstracting away the federation), then yes it is doomed.

What’s the endgame here? Who is buying ripened accounts with lots of karma and why?

Leading up to the blackout, there were multiple threads on Reddit asking for alternatives to visit instead.

None of those threads received any non-joke replies.

The only thing that will make Reddit change course is serious competition. Probably not even that, since as long as the exodus happens after the IPO, it's the public that will be left holding the bag and not the investors.

There was a magic period of time lasting about a decade when internet search Just Worked. If it was on the public web Google would find it. Search worked so well I took it for granted.

Today I avoid Google search at all costs, and use it mainly to find Wikipedia pages or to search Reddit.

The Googlers blame SEO and there is some truth to that; but Google has on retainer a huge stable of the world’s best-paid engineers, and still couldn’t be bothered to invest in their flagship service.

What you say is true, but the rolling hash solution is susceptible to potential hash collisions while the trie is not.

The AI boom has been 6 years of massively-overhyped vaporware followed by 6 months of massively-overhyped services that occasionally deliver the goods. People need some time to recalibrate.

Seems like a flagrant strawman to characterize AI as a “well-poisoning machine.”

I passed Calculus BC almost 20 years ago. All this time I could have been designing and implementing a practical self-improving AI architecture? I must really be slacking.

“no one emails”: that hasn’t been my experience or observation at all. Email remains by far the most common channel for professional communication.

Sure, my coworkers will send me Slack DMs. Random professional acquaintances do not.

It makes sense to me that the economy overall is doing well but that the tech sector specifically over invested during the pandemic and is now contracting.

There is not really any new insight in the post, though cellular automata are indeed a good way to explain entropy, coarse-graining, etc.

The breakthrough (if it exists) must be in his computational irreducibility theorem, which he alludes to in many places. Unfortunately the link is to a book chapter which doesn't render properly on mobile and seems more of a long ramble than a formal proof, so I didn't dig into it.

An effective principle along the lines of, "if the output of a sufficiently simple program 'seems random,' then it is indistinguishable from random by all other programs," would be extraordinarily powerful and would immediately close many famous open problems, such as the normality of pi (and other numbers like sqrt(2)), infinitude of primes of the form x^2+1, the twin prime conjecture (and Dickson's conjecture), etc. It's somewhat telling that these are still open problems.

I think a lot of what's happening is that there is not necessarily a correlation between the technical quality of a piece of art and its cultural influence and legacy. The former is far, far easier to evaluate "in the moment" than the latter is to predict.

For The Beatles specifically, I don't think it's a particularly controversial take to say that while the Beatles made great music, they (with perhaps the exception of Ringo) were unremarkable musicians. Lennon's butchery of the bass line in "The Long and Winding Road" is one famous example but hardly an outlier. I don't think we need to be too harsh on critics who compiled a long lists of technical failings in the latest Beatles album without recognizing its importance to the trajectory of pop music.

The counterexamples are perhaps even more interesting: cases like The Ramones's debut album that were commercial failures but recognized, correctly, by contemporary critics as the start of a momentous change despite its technical failings.

This doesn’t make any sense to me.

My private forks are *mine* and I most certainly do not want GitHub guessing at whether and when to permanently delete them without my consent.

Companies of course have the right to manage access to their proprietary source code, for example by only giving access to corporate accounts under their control and reclaiming those accounts when an employee leaves.

Well, this is a very interesting theoretical breakthrough, but it remains to be seen how it benchmarks in practice. The constant and log factors might be steep for practical inputs.

I was talking to an NSA recruiter last year, though, and they told me they routinely report applicants to the FBI for confessing to crimes during a polygraph. So damned if you do, damned if you don’t…

(For a recruiter he did an uncanny job at convincing me never to work for the NSA)

You could even have it write unit tests based on the JIRA ticket description.

This is a wonderful point: writing unit tests is exactly the kind of mind-numbing tedium that I'm super excited to automate away.

Right. Would we be impressed if a layman could pass the bar, given infinite time and access to the entire Internet (including the copious amount of bar exam study guides and worked example problems)? If not why are we impressed that a language model trained on that data can?

Meanwhile when I ask ChatGPT which of six numbers are odd, it confidently reports a mix of even numbers, odd numbers, and letters.

This is a fun milestone but the angst above about the “end of commoditized intelligence” etc. is unwarranted.