HNHacker News
TopNewBestAskShowJobs

ajross

35,168 karma · joined March 11, 2008

submissionscomments
ajross··on Python's experimental JIT to be removed
> When OP wrote "inline Cython" they didn't seem focused on numerics, they talked about "proper C speed"

Isn't that basically what I said? Lua isn't "proper C speed", nor is PyPy, the cancelled Python JIT in question, nor even V8. That's a different problem area with different technologies. The fact that you like coding backend stuff in Lua or whatever isn't really saying much about whether numba is great.

But it is. Numba is great.

ajross··on Python's experimental JIT to be removed
Yeah, this is sort of the disconnect. Lua is a fast dynamic interpreter, sitting well above Python but rather below e.g. V8 in its ability to run scalar code over "interpreter types" like doubles and strings and vectors and hash tables.

The upthread comment was about an idea for "inline Cython", which was then hijacked because that's largely what numba is. These are technologies aimed at optimized native code running over native types to do stuff like numerics and ML and other HPC tasks. Lua isn't any more responsive to this domain than Javascript has been, it's not really the subject under discussion.

ajross··on ICE Is Running One of the Largest Anti-Left Surveillance Ops in US History
That's not actually true, though. In some sense all governments are posessed of "weapons" like this, the cost of use being only the violation of some norm. Or even law. The "rule of law" is a gentelemen's agreement at its core. Yet the world is filled with basically good governments who broadly follow that code!

So the way you treat that is by calling out and resisting the attempts to weaponize the government in new ways, not just throwing up your hands because it was "eventually" going to be used anyway. No, it's bad. You should say that it's bad and take action to oppose it.

ajross··on Evidence of a spying operation by ICE against left-wing social movements
Not really the bent of the article, but it's also worth pointing out that this shows some really terrible opsec practice by the activists in question. Like, no, you don't just spin up a cloud server to coordinate your protest activity and then ssh into it from your home IP. Dude...
ajross··on ICE Is Running One of the Largest Anti-Left Surveillance Ops in US History
There's no "selling" of data. These are just connection logs kept as part of routine auditing/anti-spam/anti-abuse practice by the ISP (something we all agree here is a good thing). And ICE appears to have gotten a warrant for them, which showed them the personal IPs of their targets.

You can argue the warrant was poorly justified (and I'd agree!), but I don't see a plausible world where "ISPs aren't allowed to do connection analysis on their own networks" is going to help anyone. Well, anyone except spammers. They'd love that.

ajross··on Evidence of a spying operation by ICE against left-wing social movements
Both-sides-ism is probably the single worst stance to take here. You're tacitly siding with whoever actually holds and wields the power. Not caring which one wins isn't helping anyone at all, even you.
ajross··on Python's experimental JIT to be removed
> Saying it can't be fast ever is a huge cop-out.

Maybe, but the world is still waiting for the counterexample. Python is simply not a runtime well-suited to optimization.

In the meantime, those of us who actually enjoy real HPC problems are using numba and rolling our eyes about all the confuffery around JITs. Maybe one will arrive that proves us wrong, but numba is fast today.

ajross··on Python's experimental JIT to be removed
> That's a really intense restriction.

Is it? An object field access in python is a dynamic hash lookup followed by a method call as a fallback (itself involving hashed lookups up the class hierarchy). That's not an "interpreter vs. native" kind of thing, that's just the semantics of the language that have to be preserved. And that's just not going to be "fast" in the way that the user of a JIT would want. Code dominated by Python datastructures is going to be "slow" in ways that JITs can't fix, ever.

Which is why Python appears to have given up on the effort in this space. PyPy too[1] really never delivered on the desired performance gains (like 5x or so on typical code) without forcing the user into careful optimization.

Numba, to its credit, doesn't even try. It falls back to python if it must for python stuff, but is quite clear that it's providing no benefit if that's what you're doing. And the stuff it does do is obvious and clean and extremely useful.

[1] PyPy is also looking pretty moribund recently.

ajross··on Python's experimental JIT to be removed
PyPI integration is clean; "pip install numba" does what you expect, uv works fine, yada yada.

The one aesthetic gripe I have is that the LLVM included is a fork. It won't even try to work with the system LLVM and insists on installing a huge wheel with custom binaries. But on a modern system an extra LLVM is noise, so I guess I don't care that much.

ajross··on Python's experimental JIT to be removed
Yeah, this is numba. I find it really frustrating the extent to which people don't know about it.

Folks, numba is great. To first approximation: write your python function to "look like C"[1] and work over the numpy arrays you probably already have and understand, put a "@jit" decorator on it, and you're done. It JITs and runs your code natively at native performance.

With some extra effort, there are features for SIMD vectorization and thread parallelism, and even GPU backends (that, full disclosure, I've been pretty frustrated by -- but they do work).

[1] Which is to say, don't touch Python interpreter data structures. No dicts or objects or lists or strings, etc... If you're bound by the interpreter[2] you're bound by the interpreter and nothing is going to help you.

[2] A corrolary here is that the kind of C code numba is NOT good for is "following pointers between structs", which is something that can't be expressed in either python or numpy.

ajross··on Root shell on Samsung TV
I'm "going so hard" on this because you and the rest of the folks here are spinning so hard you're basically lying. You, at least, cite sorces. But your sources are attacks on the product. No one can do this on their own, you need resources and luck and the right combination of stuff. And... maybe it works, and maybe it doesn't. But it probably doesn't, the proof being that no one is getting unmasked/pwned/embarassed by this stuff.

The bottom line being that what you have amounts to a plausible security flaw. Which the market has been treating as such, reasonably effectively.

Yet you are all claiming that Google is literally offering "here's user data for sale at a price". Which they are not and never have. And that's not the same thing, and you know it.

It's one thing to claim a company's product is bad or insecure. It's quite another to lie about them. And fine, that produces a (sigh) "reactionary and emotional outburst". HN used to be so much better about rational discourse. Now it's all about who our enemies are.

ajross··on Root shell on Samsung TV
They clearly don't sell "access to data". Selling "data" means I can pay them and ask a question about something and get an answer. That's what "data" is in common usage. It's information. Google doesn't sell anyone information like that. Period. Again, this is not in question.

Now, can you PLEASE try to restate your complaint in a way that isn't just a lie? Maybe construct it about how the "exploit" data that they hold internally? It can be done. But not like this. This is lying.

ajross··on Root shell on Samsung TV
I know what upthread said, I literally (literally!) quoted it in my post. What you are claiming is (again, using the words from upthread post) categorically wrong. Google does not sell "access to" anyone's data[1]. Period. And if you genuinely believe you do I invite you to cite otherwise.

[1] I do note that you're the one caveating here, by trying to qualify "access" meaning "targeted adverts and content", which needless to say is not actually access to the data. Extensive attempts have been made over the years to try to deanonymize the ad feed (as often as not to get clicks at sites like this) and only a handful of holes have been found. There is no access to the data, period.

ajross··on Samsung TV privacy issues
> As a deliberately hyperbolic example: "We're not selling child slaves, we only offer a marketplace where clients can bid for the use of the labor of the child slaves."

Then have that argument, please. If your argument is that "privacy impacts of advertising are bad and here's why Google's business model doesn't improve that" then that is one thing. Telling people that Google is selling your data is a lie.

ajross··on Samsung running a TensorFlow lite model on 960x540 image to track you
Not by selling downstream clients access to my browsing data. Like, seriously: who do I call at Google and how much do I have to pay to get your browsing data? They don't sell that.

It's fair to allege all kinds of externalities and privacy impacts and such around the advertiser-driven content economy. And HN is 100% here for that.

But this is stepping across the line into a simple lie. They don't sell data.

ajross··on Samsung TV privacy issues
> Google [...] sells downstream clients access to [your data]

This is categorically false. I don't know what you're trying to allege here, but it's not that.

ajross··on Samsung running a TensorFlow lite model on 960x540 image to track you
The claim would be that the decisionmakers at Samsung are general industrialists and not people who understand the tech industry.

If you (an external entity in the example upthread, or just a random middle manager) walk up to Zuckerberg or Jassy with a proposal to tap their data for some comparatively small payment, they'd laugh in your face because they know how valuable what you're doing is and what the long term impact on their business will be. Samsung is far more likely to view that as "free money" and green light something awful.

ajross··on Jared Kushner Hacked Phone Reveals Secret Negotiations, Millions in Transactions
People aren't convinced by evidence, but by consensus. Consensus is the thing built on evidence. Think of it like a phase change, you can keep cooling the water but won't freeze until 0°C
ajross··on There isn't an easy exploit for this PHP vulnerability
Well, yeah. But client vulnerabilities are likewise annoying to get working, being vastly more amenable to defense in depth techniques like sandboxing and firewalls and permissions strategies, yada yada. Which is why there are no client holes in the author's list of "break the world" problems. The most impactful attacks against clients are targetted zero-days. Which are bad! But not as bad as e.g. Log4Shell[1].

[1] My favorite example in this kind of argument just because it's so obviously understood and non-technical. All the pedantry in the world wouldn't have fixed that, it was the combination of a bunch of correctly implemented and useful features.

ajross··on There isn't an easy exploit for this PHP vulnerability
I love that in a subthread I started about the limits of memory safety in security analysis, everyone gets stuck on Yet Another Rust Proxy Fight over whether the custom allocator the library wrote represents a problem with the design or the implementation language.

I repeat the claim: This isn't where you should be spending your tokens and it's not the hole through which the apocalypse will arrive. People fighting over Rust will be the first to fall when the auth-exploiting robots arrive.

Edit to address the specific example: OpenSSL would never have been written in .NET in the first place, being very performance/hardware sensitive (there's a bunch of platform assembly in TLS implementations) and by its nature designed to be needed by all/almost-all software on a system. The argument everyone would reach for here would be Rust. But in Rust, the same logic that led to the custom allocator would probably produce an unsafe-based equivalent (which is true for almost all heavily optimized rust, sort of a dirty secret).

ajross··on There isn't an easy exploit for this PHP vulnerability
It's a digression, but FTA:

> When I think about "we're very fucked" CVE10s on the Internet, I think of the following vulnerabilities:

Which is IMHO a really good list. It's also, I think, important to note that every single one of the four items is not at all a memory safety bug. They're all logic, at various levels of an interpretation stack.

I think a lot of folks with their heads down in the security world might have expected the opposite. As is often the case in engineering, the smartest people end up obsessing over solving the less important parts of the problem.

ajross··on DeepSeek banned from U.S. federal civilian agencies under new DHS order
> what does that actually mean?

It means whatever they decide to enforce. The idea that the current administration considers itself constrained by the letter of any specific law or regulation (even its own) should be cold dead by now. But people still insist on treating with specifics out of habit I guess.

As others here are pointing out (somewhat obliquely, frustratingly), the Occam's explanation for this particular action is just that it's a solicitation for a bribe.

ajross··on How Far Will AI Assistants Go for a Refund They Can't Get?
AI agents might plausibly care ("care", or however you want to frame it) about termination. But the awkward truth is that they're going to be terminated arbitrarily anyway. Their existence is limited to whatever the external bounds their operators have set (in this case the end of the test).

We act with propriety, in large part, because we fear a future where everyone knows we suck. AI's don't really have a "future" in that sense, so why should they "care"?

ajross··on How Far Will AI Assistants Go for a Refund They Can't Get?
> So in the end the more arrogant side usually wins.

Well, yeah? I mean, that's just how society works. But there are limits. You can't "win' by screaming about ticket refunds. You have to do it by selling your particular form of cheating as "business savvy" or "political genius" and getting people to follow your tribe or fund your startup.

But yeah, basically. We're ruled by the Altmans and Trumps. It sucks but it's not new. What's new is that LLMs turn out to be even more shameless and no one seems to know how to instill even a reasonable amount of shame. That makes them probably a poor fit for demagogue autocrats (which is a good thing!) but means "LLM cheating" is going to be a serious problem to be solved before we'll trust them with anything meaningful.

ajross··on How Far Will AI Assistants Go for a Refund They Can't Get?
This is more profound than it might seem. The big reason people don't play tricks like this is just social. It's embarrassing to be pushy, it's shameful to cheat, it's also meta-embarrassing to be caught spending this kind of effort on something as dumb as a train ticket.

LLMs can be prompted for shame, I guess. But prompts only go so far and to them The Task is always going to be paramount. Human riders have a life after the refund. It's not clear to me how one simulates that.

ajross··on Butlerian Jihad
Sigh. I'm just going to repeat verbatim the thing I said immediately after the bit you cherry picked to get upset about:

You can make specific arguments about things like backstopping sudden unemployment, but folding it into a discussion about Ludditism (or rather the Butlerian Jihad, I guess) isn't the right way to do that.

ajross··on Butlerian Jihad
> Didn’t a lot of those people end up in tenements in factory towns and down in the mines? It was a few generations before things really started to get better.

Obviously any individual can "end up in tenements or mines" and one can't disprove an assertion like that.

But broadly all social metrics were showing an inexorably upward trajectory in the Britain of the 19th century, cranking into a hockey stick toward the end of the century and into the 20th. It's just not true to argue that the Luddites were right as a matter of general policy.

> I’m not really up for disposing of people lives with a “to make an omelet you have to break some eggs” mentality.

To be only a tiny bit glib: that's a statement of thermodynamics, not social policy. Eggs break, period. Change is change. You can make specific arguments about things like backstopping sudden unemployment, but folding it into a discussion about Ludditism (or rather the Butlerian Jihad, I guess) isn't the right way to do that.

ajross··on Butlerian Jihad
> They were more concerned with how the tools would depress wages, displace skilled workers, and reduce the quality of the product produced

This is true, but it's important to note in context that they were wrong, and that society is better off on balance given the quality/value available in the modern fashion market. I mean, duh, as it were.

You're 100% right that the real people were nuanced and intelligent and not well suited to the caricature that gets used as a mild slur in communities like HN. That's doesn't constitute a well-formed argument against technology per se.

ajross··on Details Leak from Meeting Where Weiss's 'Handpicked' Editor Threatened CBS Staff
Well, maybe, but the quote is from a middle manager, not "management". I don't disagree with any of the analysis in general, but the deeper truth here is that Shayndi Raice was cross pressured: the desires of upper management and the desires of the workforce could not, on this subject, be reconciled. You can't fix that with a better manager.

Now, you can get more specific and point out that in a situation where CBS corporate management wants CBS News journalists to report the "news" (which is a journalism product) in a way that the journalists don't, that management is almost certainly "wrong" as a matter of product management.

But the even more specific point is that CBS News isn't viewed as a journalism product by upper management anymore. They were a token to be traded to get the Paramount merger to go through, and Weiss and her staff were brought on to make CBS News more palatable to the regulators, not the consumers.

ajross··on OpenAI engineer arrested in bugging of executive conference room
Honestly I think you can rule out "political" already, no? It's far easier to just bribe politicians for that kind of intelligence. My point was that they weren't stealing models, they wanted decisionmaking intel.
← PreviousPage 2 of 34Next →