topsycatt

That's not accurate. Take a look at the paper to see what it is trained on! And specifically decontamination is called out in A.4

https://microsoft.ai/wp-content/uploads/2026/06/main_2026060...

I work on that team! We're growing extremely rapidly (I joined 4 months ago from GDM and I'm at the 50th percentile of tenure), if you're interested in E2E model training reach out at https://microsoft.ai/careers/

That's a very good point. Let me speak with some folks and see what I can do.

I have not. I'll take a look, thanks!

Sorry for the delay in replying!

We actually use gVisor (as stated in the article) and it has a very nifty feature called checkpoint_restore (https://gvisor.dev/docs/user_guide/checkpoint_restore/) which lets us start up sandboxes extremely efficiently. Then the filesystem is just a CoW overlay.

Thank you very much dang!

Thanks!

We could, it's just not high up on the priority list. Any particular reason you want php?

I submitted this HN link with a title that exactly matches the one on the article, but I didn't write the title on the article. AFAIK HN posts should match the title of the article they link to.

Canvas actually has a mix of this sandbox (with a different container) and fully client-side.

The "run" option for generated code was removed due to underutilization, but the sandbox is still used for things like the data analysis workflow and running extensions amongst other things. It's really just a general purpose sandbox for running untrusted code server-side.

I'm on the Google side of the equation. I think the title is a bit sensationalized, but that's the author's prerogative.

The google

That's a good question! It's not incompatible, it's just a matter of getting the flow right. I can't comment too much on that process but I'm excited for the possibilities there.

That's the system I work on! Please feel free to ask any questions. All opinions are my own and do not represent those of my employer.

I noticed seemingly all deaths in Manhattan are labeled as having occurred in flushing, a nearby neighborhood. Perhaps an off by one error?

Oh hey! I ghost-wrote an earlier article that references the same idea (https://blog.google/technology/ai/bard-improved-reasoning-go...). I wonder if they came to it independently or if they read mine and it planted some seeds? Either way, super cool work.

If only it was called GNP...

Very cool, but some basic stuff is broken having used the app. For example, I'm unable to create a new conversation since I'm unable to click on a contact after having searched for their name in the "New Chat" flow. Looking forward to using it once this is fixed!

Edit: Seems like I have to double tap instead? Not super easy to use but it works so I'll take it!

Hey I work on that! I can only share information that's already public unfortunately, which isn't much beyond what's on that page...

Gotcha, that makes a lot of sense. Thanks!

I'm working on it, but part of the motivation for the question is wanting to find people to get ideas / guidance / friendship from who are more interesting than myself.

Kind of an “If you're the most interesting person in the room, you're in the wrong room.” vibe.

Interesting. Can you elaborate on "become the 'host'"? Maybe a concrete example would help me understand what you mean

Great suggestions, thank you!

Defining interesting is an exercise that's left up to the reader ;)

I personally think it's fascinating to talk to people who are unlike myself, but your approach is definitely valid (if a bit "draw the rest of the f**ing owl") for a given interest.

No, I actually do conform to a rote/traditional lifestyle fairly closely. I'm actively trying to branch out with some success, but I'm also looking to talk to people who are unlike myself to get outside perspective, ideas for growth, and just to have a fun discussion!

I like to think I don't come across as aloof, self-centered, and dismissive, but -- like I mentioned in another comment -- who really knows what another thinks of them? Perhaps that's a good area to validate my assumptions so that I'm not shooting myself in the foot.

And on your last point, I definitely agree. Sometimes however you can find yourself in a situation where those barriers don't really apply and you can have a deep, interesting conversation with little prompting. I guess the heart of my question is about where people find those interesting conversations with greater regularity.

Both are fair questions. For the former, I'm personally looking for new friends and the novelty of meeting strangers. Some of my favorite memories are from meeting strangers and getting to know them over the course of an evening.

For the latter, I like to think not (but who really knows someone else's experience of themselves?). I enjoy talking to strangers for it's own sake and I'm not generally bracing for boredom, my question was primarily about good places to explore that have a higher ratio of interesting interactions to non-interesting interactions.

I understand the classification is very subjective, so I left the question somewhat subjective to hear not only answers to the core question, but also answers to the implicit question of "what do you find to be interesting".

Link to the current list of restrictions: http://pe.usps.com/cpim/ftp/manuals/imm/full/imm.pdf

A few actual quotes that I've found in just 10 minutes of browsing:

San Marino: "Toys not made wholly of wood"

Qatar: "Musical greeting cards"

Albania: "Extravagant clothes and other articles contrary to Albanians' taste" (already in the article but I thought the literal quote was even better)

The list is a gold mine, it seems every 5th country has some strange restriction

Yes, they still are useless, at least for the -O3 and -O4 sections.

The compiler removing the loop is due to a bad test setup in which the function in the loop does nothing and compiler can figure that out and skip the whole calculation. In a real scenario the function in the loop would (presumably) actually do some work so it could not be elided. Also, note that the compiler does this for BOTH the Duff's device version and the non Duff's device version.

Also, I'm neither attacking nor defending Duff's device. I'm just saying that we can't draw any useful conclusions from this article because of poor methodology.

I think the author is making a fairly critical mistake here. If I understand correctly, they missed the fact that the -O2 and -O3 results are several orders of magnitude faster... Likely meaning the compiler optimized away the whole loop in those cases, rendering their results useless.

Edit: Yeah, looking at the godbolt link they provided, at least the execute_loop_basic function does essentially no work.

Interesting article! I think there's an error though, where the first constraint should be expressed as "(5-x) + (x-5) = 10" (the author has "(x-5) + (x-5) = 10" currently).