US Cybersecurity Job Postings Fall 26% as AI Security Engineer Demand Soars https://www.securityweek.com/us-cybersecurity-job-postings-fall-26-as-ai-security-engineer-demand-soars/
Today, the government of #Austria announced funding cuts for #universities , and Trump accounts.
RE: https://mastodon.social/@nixCraft/117451805824031590
I seriously hope that Linus Torvalds understands the difference between a compiler (which provides guarantees) and Agentic AI. Given how mature compilers are, there is little reason to read the binary code generated by a compiler. Maybe once LLMs routinely provide correctness proofs for the code they generate, that will be true for AI-generated code, too, but I think we are not there yet.
Quite a few people have sent me the "Passkeys are a trap. Do not use them." post on Substack.
The post recommends people abandon passkeys and stick to passwords + TOTP, citing what appears to be concerns with attestation as the primary reason.
I’ve written a rebuttal as this is really dangerous advice and it's based on some incorrect technical claims.
At least in mathematics, the goal of proving theorems is to gain a better understanding and to simplify the argument to present it in an understandable manner (apparently not a forte of LLMs).
In formal verification, arguably we do not care about the proofs (as long as they are correct).
In both cases, formulating the actual goal (or theorem) that should be proven remains a task that requires humans (at least for now).
https://proofsandprompts.com/2026/10/08/100-reactions-to-100-solutions/
Quoted by Michael Harris:
“The result is also contained in a paper… released by OpenAI on October 6, 2026, in which the proof strategy and specific choices of notation are identical to a preliminary version of the present paper that was uploaded to ChatGPT on September 8, 2026.”
https://siliconreckoner.substack.com/p/now-all-mathematicians-can-be-mechanical
On that note, this 2019 talk by Ben Titzer was how I grokked a lot of the non-obvious Spectre issues in depth at the time when it was new and fresh, and it's still the best technical explanation I've seen, especially if you're in the target audience of language implementors: https://www.youtube.com/watch?v=FGX-KD5Nh2g. I'd love to see someone give the updated 2026 version of that talk; I know a lot has happened on both the attack and mitigation fronts, I just don't have a good mental map of them.
"If in fact Russian-made software [...] was used for handling EU Commission criminal evidence database, it raises a serious possibility of sensitive information being compromised and accessed by Russian authorities"
New from @timnitGebru and me:
"It is high time for us to regulate usage of systems sold as “AI”, not because they are magical machines, but because they are error-prone products that shouldn’t be used in high stakes scenarios."
@dominykas @gooba42 Now, that said, no matter how you’re creating software (from hand assembly to AI learning models or anything in between), ensuring correct, safe behavior is HARD. Evey programmer knows this (and there are fundamental reasons for it), which is why testing and debugging are an expected part of the development process, starting well before systems are inflicted on real users and consequential environments.
"I quickly gave Wren access to my bank accounts and apps like Google Docs, Gmail [...]"
"[T]he A.I. agent had called my dental insurer, given it my member ID, answered a security question"
What could possibly go wrong?
finally doing the experiment of daily driving #phosh on #postmarketos
with fractal for #matrix
meshy for #meshcore
parla for #deltachat
Reassuringly, the foundations of security and crypto still hold.
But agentic AI has changed the game: discovering potential vulnerabilities has become cheap (well, relatively; $100k is still more than your average research grant will allow you to do) and fast - verification (expert checking) is the bottleneck.
From the blog post: "[W]ith human review now the bottleneck rather than discovery, [cryptography professor Matthew Green] argues the field needs machine-checkable proofs, not just a solver's word that it looked correct."
https://www.ben-livshits.org/blog/breaking-cryptography
#agentic_ai #llm #security #cryptography #FormalMethods #AutomatedReasoning
Wait, my iPad Pro 2018 is stuck with iPadOS 26? Cruel.
#Apple should at least clean up the glass shards.
Alter. 21627 Einreichungen auf ein Fellowship-Programm. Wer soll das alles begutachten? 😶🌫️ Wer jetzt nicht für mehr Grundförderung und weniger Drittmittel argumentiert, der ist doch nicht gescheit.
A vector looks like an arrow; anyone studying math or physics in college learns about these.
Bivectors are important too - though less widely taught. You can create a bivector by taking the 'wedge product' u ∧ v of two vectors u and v, and you can visualize this as the parallelogram shown at left here.
But in fact, an ellipse or any other flat surface with the same area lying in the same plane is just as good a picture of this bivector. Further, you can add bivectors, so there are bivectors like u ∧ v + w ∧ z that you can't picture as just a single surface.
Thus, you need to be good at the algebraic manipulation of bivectors to keep from making mistakes: unlike with vectors, the picture is just a rough indication of what's going on. Yet bivectors are extremely important.
For example, while the electric field is a vector field, the magnetic field is really a bivector field. In 3d space we can convert bivectors into vectors using a trick called the 'cross product': we replace u ∧ v with a vector called u × v. So that's what they teach us in college, to avoid talking about bivectors. But this trick requires an awkward 'right-hand rule', and in higher dimensional space it doesn't work at all.
You can go on and define trivectors, etc., and some of the same issues show up.
So now let's talk about the classification of trivectors in 9-dimensional space.
[Typical way a math professor gracefully transitions to what they really wanted to talk about.]
(1/n)
At today's #FMCAD keynote, Claire Xenia Wolf from @yosyshq gave an inspiring presentation on how to understand and visualize the Goats and Sheep (or Dancing Witches and Wizards) operation of CPUs.
this was incredibly annoying but I've added 1Password SSH key import to ChiPass! this is our first significant feature work
Die Forderung nach einer "Pause der KI-Entwicklung" hat nichts mir ihren echten Gefahren oder so zu tun, sondern mit einem crashenden Businessmodell. Wir dürfen uns nicht auf diesen Effective Altrusim-verseuchten Diskurs einlassen.
I miss the good ol' times, when supercomputers were actually stilish pieces of furniture.
context: https://inf.ethz.ch/news-and-events/spotlights/infk-news-channel/2021/07/archives-cray.html
#cray #nostalgia #retro #retrocomputing #eth #ethz #zurich #switzerland