HNHacker News
TopNewBestAskShowJobs

magicalist

13,114 karma · joined December 14, 2010

submissionscomments
magicalist··on Tao: Open math problems being non-renewably mined by AI
> Soldiers rarely know the history of war, and war historians are rarely soldiers.

I mean, besides the empty platitude that we have no reason to assume applies here, we can easily search and find Tao commenting on the history and philosophy of mathematics.

This is a really weird subthread.

magicalist··on ChatGPT Images 2.5
> Then will we continue to see the slop accusations?

If it doesn't look like the food you're about to purchase, yes? Why is this even a question in this context.

magicalist··on ChatGPT Images 2.5
I mean, I actively advocate for better fuel efficiency standards but also haven't entirely eliminated red meat from my diet. If I'm discussing fuel efficiency standards and you call me hypocritical, even if I granted that, it doesn't seem like a useful turn of the conversation for anyone unless it's a rhetorical move to change the topic of conversation?
magicalist··on Navier-Stokes – Tristan Buckmaster [pdf]
> If this is true he should release the actual emails

these were statements while on a call, and at least the career comment Bubeck has admitted to while doing damage control ("I deeply apologize for this extremely poor choice of words, it is the opposite of what I was trying to convey. (I should say that I retracted them on the spot by the way.)"[1]).

[1] https://xcancel.com/SebastienBubeck/status/20973794116915163...

magicalist··on On the Navier–Stokes Millennium Prize Problem
> identify any of their de-identified data that came from their usage of ChatGPT

"de-identified" seems more of a euphemism than normal in this context, given the very unique work they were doing.

magicalist··on On the Navier–Stokes Millennium Prize Problem
> Buried under the drama is the fact that OpenAI is claiming that an internal model they’ve been training for less than two weeks is more than twice as capable in mathematics as Astra.

Is this buried under the drama or are the major OpenAI twitter accounts from the people involved in the drama desperately attempting to make this the story after everything else obviously got away from them?

magicalist··on Recreating Minecraft Is Not a Benchmark
That's playing RuneScape though? Unless the connection is just that these are both RuneScape related?
magicalist··on Actively exploited sandbox RCE in all Chromium versions
I believe that means the v8 sandbox, not a renderer sandbox, based on the v8CTF reference.
magicalist··on Record-High 89% in U.S. Say Government Corruption Widespread
> “We found the scapegoat, blame that guy!”

Yeah, that's not what an scapegoat is.

magicalist··on GPT-6 Astra
> You could script this in with a time database, illustrations, and appropriate harness, tests, and editing passes.

Yes, and I could script a truly marvelous proof if this textarea were but a little larger :)

Hand waving doesn't count for much these days when you could spin these things up quite quickly to prove the point, so the GP's claim seems much stronger than whatever you're not convinced of?

magicalist··on Claude Fable 5.1 and Claude Mythos 5.1
> I suggest you to give a look to the MCP protocol for hardware that is being proposed by Anthropic. The hardware will be the next harness of LLMs, they will be able to operate machines to reinforce their theories.

Yeah, that's called an API. Again.

The actual hard problem that this hand waves is making (and funding the making of) hardware to reliably do the things you need it to do.

magicalist··on METR and Redwood Offer Holy %^ Postmortem of the HuggingFace Hack
> Uhhh… how would literally any finite number of humans actually read and comprehend the log outputs of even a single agent, never mind hundreds or thousands of them interacting with each other over weeks across disparate systems

I mean there's quite a lot of people in the world whose specialty are to dig through logs from "hundreds or thousands" of clients, including intentionally deceptive ones, to spot problems.

The ridiculous thing is to mythologize these pretty standard hacking approaches. It's shocking/amazing/whatever that automated agents were doing this, but they weren't doing it through some inscrutable method beyond human understanding.

magicalist··on METR and Redwood Offer Holy %^ Postmortem of the HuggingFace Hack
> This is the case with all complex system failures. There were always obvious fixes that could’ve prevented it.

From this writeup and the Black Hat talk I'd really disagree. That would be like saying my hospital getting ransomwared because we didn't update our version of MSSQL because no one in particular was in charge of keeping dependencies up to date.

Sure systems are complex, but this is well trodden territory. Agents aren't the first things trying to break in or out of sandboxes, or the first ones to have done it, and based on these reports the reason they were able to work on this for so long was not because of super human intelligence.

magicalist··on METR and Redwood Offer Holy %^ Postmortem of the HuggingFace Hack
> Those turned out to be inadequate against proto-GPT-6 agents

This is nonsense.

Gross negligence in the sandbox and system aside, humans literally noticed the agents in action doing what they should not be able to do in their sandbox and decided not to act upon it. It's difficult to explain that except if safety and security is simply not part of their engineering culture.

magicalist··on Gemini Omni 1.1 Flash
> See current situation where the Carpenters' Union in California is pushing hard against an initiative that would make it much easier to build more housing in the state because it would also reduce requirements for union labor

If you're talking about AB 1751 (Missing Middle Townhome Ownership Act) I think you have this exactly backwards.

> Unions are very much an in-group vs out-group phenomenon

As opposed to what? You and the GGP phrase this as a dichotomy, but I'm really curious what the other side of this is, because "not being in a union" doesn't erase your self interest or all the many many groups of people doing the same that don't feel like they're apparently being overly self interested by forming corporations, governments, non-profits, advocacy groups, etc, etc, etc

magicalist··on OpenAI Jalapeño: Better than Nvidia Blackwell
> Since the current NVL72 are still at the ~15%/yr failure rate it's not clear your new data center is going to have half it's compute in 3 years

This would be a stupidly bad failure rate, basically the worst business decision you could make, especially if you're somehow on the hook for eating those losses (which seems to be the implication?). Is there a linkable source on this?

The only thing I could find is SemiAnalysis claims that 15% of Blackwells end up RMAed[1]. That appears to be a total failure rate, though, and if you're RMAing them, you're getting replacements. So that appears to be a pretty different state of things.

[1] https://www.dwarkesh.com/p/dylan-patel#:~:text=GPUs%20are%20...

magicalist··on Firefox 157 will include JPEG XL by default on all platforms
> As much as I enjoy rust, there is no reason why a c++ library can’t be optimised to match the rust implementation.

I don't think anyone is claiming that. I took the GP's point as what was desired was a "safe, performant, compact, and compatible JPEG XL decoder in Rust" and "performant" (as defined as the speed of the c++ version) was passed two months ago.

> They usually come from the fact that rust implementations are usually a second (or 3rd, or 4th) iteration of the design, and the lessons learned help performance.

Sure, and looking at the recent commit histories, the team is focusing on getting the rust version ready for this milestone for obvious reasons, and is less concerned about immediate parity in the c++ codebase.

magicalist··on Oceans hit highest temperature on record
> The RWE paid 1.26 billion to acquire the lease on offshore wind. Payment is a refund.

The Trump administration told them it could continue to make the process drag out for more years as it has been or they could spend the money on fossil fuels. It was not any kind of market decision.

magicalist··on OpenAI: GPT 5.6 Sol price reduction (until at least Nov 21)
> Meanwhile Nvidia just announced a space-optimized Vera Rubin designed with SpaceX designed for orbital datacenters.

well, no, they announced the concept of a space-optimized Vera Rubin designed with SpaceX:

> NVIDIA and SpaceXAI are working to adapt that foundation to the requirements of orbital computing while preserving a common NVIDIA architecture and software ecosystem.

The press release is really announcing that SpaceX's terrestrial data centres are going to use Vera Rubin.

https://nvidianews.nvidia.com/news/spacexai-adopts-nvidia-ve...

magicalist··on Pacing model development in an era of cyber-critical capabilities
Not sure what you're arguing here (we shouldn't even try?), but even going fatalistic, that's still fully compatible with "Treat models as untrusted and potentially compromised/hostile and proceed accordingly".

Of course you shouldn't fully trust a model to properly redteam your sandbox, but that doesn't mean you shouldn't redteam your sandbox, including using your own security models to do so.

magicalist··on Pacing model development in an era of cyber-critical capabilities
> what if they know they are in a sandbox and that they should fail the test in order to escape in the future.

What if they're able to find hardware exploits and commandeer nearby access points across an air gap? What if they hack my brain waves to indoctrinate me? Etc

You still have to start with the basics regardless of speculative unknowns.

Treat models as untrusted and potentially compromised/hostile and proceed accordingly.

magicalist··on Field measurements of neighborhood-scale air temperature impacts of data centers
When Amazon gets to run an offgrid power plant for a new data centre that's permitted to release more co2 than any other single power plant in the US[1], you'll pardon me for wondering what the rush is and why it becomes literally everyone else's problem as a result.

Build the data centre, sure, but if it takes you five years to build out the grid and sufficient generating capacity in a non stupid way in order to do that, sorry, but that's by far my preferred route to get there.

[1] https://www.nytimes.com/2026/08/08/climate/amazon-data-cente...

magicalist··on Universal health coverage could save $1T and 114k lives a year: study
Like this one: https://jamanetwork.com/journals/jamasurgery/article-abstrac...

> Among patients undergoing cancer surgery, waiting times to initiation of first-course therapy have steadily increased since 2012, particularly at high-volume academic centers and among patients referred for definitive care. With continued consolidation and expansion of health systems, system-level strategies are urgently needed to monitor and mitigate delays in the delivery of surgical care for cancer.

> Delays were more pronounced at academic compared with community hospitals and among patients referred for care. Predictors of longer waiting time included Medicaid insurance (5 of 6 cancers), lowest-quartile income (6 of 6 cancers), Black race (5 of 6 cancers), increased travel distance (4 of 6 cancers), care in the West region (6 of 6 cancers), and treatment at academic institutions (6 of 6 cancers). Receipt of robotic operations was linked with longer waiting times for nonbreast malignancies (5 of 5 cancers).

magicalist··on Universal health coverage could save $1T and 114k lives a year: study
Indeed: https://arstechnica.com/health/2026/08/us-wait-times-for-can...
magicalist··on On AI regulation and messaging
> Maybe I'm overly cynical, but I think charities exist to exist. They don't have a strong incentive to actually deliver on their mission. Everyone at the org may be fully bought it and obviously want to cure cancer, but as an organization it would fail to deliver on the promise. Like most organisms, their actual purpose is to continue to be influential, grow and continue to exist.

I mean, yes, "I haven't looked into the specifics, but working from the first principles I came up with in 3 seconds, cancer research charities are in fact cancer on society" does come off a bit reductively cynical.

magicalist··on Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models
> Wasn't LLaMA or Llama 2 the first ever open weight LLM through a leak

No

> and weren't there speculations that Zuckerberg himself might be involved in it

and no. Meta legal was issuing DMCA notices within days (so, plenty of time for execs, including Zuckerberg, to sign off on, or even propose, that strategy).

> He does seem like a good-ish guy at the most crucial moments of truths.

Does he? I can't seem to recall any particular crucial moments that revealed any underlying character there.

magicalist··on Mark Zuckerberg attacks 'closed' AI rivals as Meta returns to open models
> It depends on if you can say things like "it is good they are supporting this" or "I'm glad they're doing this" or "good for them" without having to couch it in a bunch of stuff about how terrible they are.

That is a weird definition of intellectual honesty. If you hold both of those opinions, what's wrong with stating them both?

If the answer has to do with the rhetorical effect, that seems a lot harder to justify as intellectual dishonesty, but I don't want to put words in your mouth.

magicalist··on The Claudyssey: A line-for-line translation of Homer's Odyssey by Claude Fable 5
> The only thing Fable was given was a clean of copy of the ancient Greek. This is not a bastardization of other translations

FWIW I gave GPT-5.5 two lines of William Cowper's translation (long out of copyright to avoid any filters) without web search and not only was it able to identify it, it was able to continue it verbatim for roughly 30 lines at which point it went off course, but only by skipping like two books ahead. If I had supplied the original text as an anchor it presumably could have kept going.

I didn't try it with Fable because I'm lazy, but all these translations (presumably the ones within copyright too, given the lawsuits) are definitely well known by these models.

magicalist··on Tax cuts for the wealthy only benefit the rich (2023)
> All else being equal, you want to produce more with less labor and other inputs

Is that actually true? Maybe I'm being dense right now but that's not completely apparent to me.

More abundance per person does seem generally better (albeit there's eventually diminishing marginal utility), but that's not the same thing as society as a whole producing more with less labor.

Maybe "all else being equal" is the operative condition here, like assuming that the economy would be able to absorb newly freed labor into new economic activity?

You can also optimize some good things out of existence in this manner, and it's still not clear that robots are going to be able to swoop in and save that kind of newly uneconomical activity, at least within my lifetime.

magicalist··on Responding to the next frontier of critical cyber capabilities
So they found their agents had RCE'd Artifactory once, reported it and got the fix, continued using Artifactory for their sandbox, and left it unmonitored for days despite the earlier exploits? They really do come out looking totally incompetent.

I stress about my agent sandboxes all the time and the only models I run have the default heavy handed guardrails, and I don't leave them running persistently.

Edit: not to mention, why is your first cybergym not your own sandbox??

← PreviousPage 3 of 34Next →