Hacker Newsnew | past | comments | ask | show | jobs | submit | sigmar's commentslogin

When did it happen before? The openai navier stokes proof doesn't look anything like the other methodology.

Nice. What did they use for the statement of the problem? Was it pulled from the 10ish line deepmind formalization of the hypothesis? https://github.com/google-deepmind/formal-conjectures/blob/m...

It uses the one in Mathlib: https://leanprover-community.github.io/mathlib4_docs/Mathlib...

Which is referenced on the Deepmind formalisation.

But no, the DeepMind formalisation file is not there. I doubt that Alpöge would put out something that has errors in the proof statement.


>remote sensing is difficult and any archaeologist knows to triple check before saying something is "new"

>Francisco was very kind and shared his publication image with me

...can she post the pdf?

Seems likely the google SWE that wrote the blog wouldn't have claimed this was new if there was a public way to access this information. I understand archaeologists have valid reasons to keep location details private, but complaining the OP didn't "check" thoroughly (when she can't link to the data he should checked in order to validate it was new) reads more like a shortcoming in how this knowledge is shared than a failure of OP to check.



Not to agents

I told my agent to bypass Cloudflare captcha's and it happily did so.

which agent, out of curiosity?

Probably Sonnet. Just take one of the Cloudflare scrape python modules.

He expects me to click this link with my own finger? And then read it with my own eyes?

> So far, only 2 of the 442 have been exploited in the wild, just 0.45%.

> almost none are being exploited

Isn't the whole point of glasswing to patch stuff way before it can be exploited? If the patched code is distributed quickly- there's no opportunity (or incentive) for bad actors to craft and distribute an exploit. This reads like an attempt to spin them as not being "real" vulnerabilities.


I see at least two more CVEs in that https://github.com/patrickmgarrity/Anthropic-Credited-CVEs repository that were most definitely "exploited in the wild" (https://www.cve.org/CVERecord?id=CVE-2026-93292 and https://www.cve.org/CVERecord?id=CVE-2026-92729) -- by virtue of being the one at my org who had to deal with it at the time. So I'm not sure what that person is basing this on.

Exactly what I was thinking ...

    But are these vulnerabilities actually useful to attackers?
I thought the idea is that they are finding them before they are used in the wild -- how would an unfound vulnerability be useful to an attacker?

This question is a rewording of "are these questions actually high impact?", and is being asked because Anthropic said almost everything they found was high impact.

It’s almost as if the author thinks Glasswing should arming attackers with vulnerabilities.

This assumes that all vulnerable devices update, and do so quickly, which is not the case.

The FAA, responsible for all American air travel, still uses floppy disks.


>Nearly 80 of the foundation’s 250 or so employees voted to nix the suicide pledge after the draft was released, according to the lawsuit, but leadership at the Ethereum Foundation ignored their objections and included the pledge word-for-word in the final version of the mandate.

Contacts are still contracts even if you put pictures in it. You definitely don't understand how the world works if you don't get that.


>That $1 of compute is not taking into account R&D spending, infrastructure, etc

People assume this, but it is wrong. Check the S-1 for Q2, the $2.08 in revenue is for every $1.00 spent on "compute and infrastructure expenses" which also include R&D compute. (Chart is on page 137)

"Compute and infrastructure expenses primarily consist of costs attributable to operating our technology infrastructure, which supports the full range of workloads for our models, including research, model development, serving customers, and internal usage of Claude." page 146 in the S-1/A-1


It is not really clear. In the graph on page 137 does the "dollar of compute" includes only compute required for inference to provide the answer? Or does this include training as well?

It would be logical to assume that they show expense of compute required to run inference. Leaving training compute (which is huge expense) out.


One of the most important (of many) questions the board will have is "did Jensen give her any privileged information?" which means it is also an investigation of him.

>It's simply not autocompleted.

Inaccurate. According to swipeblocked.com it is blocked with both swiping and autocomplete.


Not sure what you think is "hilariously cagey", but why would they only be "cagey" if the bug was from ai-features? Also it says paint versions are "affected from 11.0.0.0 before 11.2605.91.0" which I take to mean the bug has been there since Windows 11 was released in 2021


>Google/Alphabet is conflating GenAI/LLMs with traditional machine learning

I see people try so hard to make this distinction when it really doesn't exist. Transformers are only a little different than long short-term-memory architecture (parallelized, with attention mechanism and no recurrance).

What is the point at which is departed from tradition machine learning? Seems more like people retrofitting a moral decision to not like LLMs and trying to dress it up as a technical distinction.


determinism


This is not inherently a property of either LLMs or "classical" ML.


I can run a linear regression and get the same result every time or I can go to chat.com and get A/B tested upon. yes these are cherry picked examples. they're also the ones that apply to me personally. for a higher level version of this argument please see my response to the other responder.


linear regression on commodity hardware is generally not going to be truly deterministic delta implementation details


Transformers are plenty deterministic. We just decide to run an rng against their output to get the final token in most cases.


when concept A applies to concept B in practice, a lack of theoretical connection between A and B is basically irrelevant to me as a lay person. people don't seem to grasp this argument well here so I'll be very clear, yes you are entirely correct, concurrently there isn't much reason for me to care about correctness when my eyes and ears tell me something else


Non-associativeness of floating point arithmetic plays a major role as well.


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: