Pinned Tweet
Black October: an optimistic (yes) wake up call
As the cybersecurities events unfold in real time, with breaches and vulnerabilities that would have been deemed "black swan events", it's apparent to anybody in the field that black swans are now flying in flocks.
Many color-pilled circles saw it coming, either with fear and Cassandre-y certainty, or with a feeling of "dooming-but-still-unreal".
Now, there's no coming back to normal.
And yet, I'm still very optimistic about the future (let's say, medium-term future) of software, the web, and cybersecurity as a whole.
1/ Every CVE is a lesson.
One software pays full price for it (early 26 Node, PHP etc) but each and every one of this event leaves us, as a field, more prepared, and not less.
We do discover new failure modes of software previously seen as safe & proven. A newfound and probably useful mistrust in decade-old ways of securing & architecturing our code, products.
Even better, we have both a new direction in which to look, and the tools to do it right away.
"But those are the same tools that put us in this situation!"
That's right. Call it Red Queen hypothesis, cat and mouse or simply an arms race, it definitely is a two players game where the first mover has a temporary advantage. So far, offense had the upper hand, just by being fast & focused.
However, every vulnerability that's discovered, and exploited, yields a patch, and a whole new family of things to correct in un-exploited tech, libraries.
This attention to cybersec also let us benefit from the swiss cheese model. Being wary means that currently working security layers are finally well configured, well thought. All in all, when you're suspicious of every step in your network, you build & maintain a better system.
Finally, I want to say something that might seem perfectly misled & naive:
At some point, we'll run out of vulnerabilities.
At some point, wave after wave, filling cracks after cracks, we'll get to the point stuff can be trusted again. When the super-intelligent-demi-god-ai tries to break in for half a decade, whatever tech is left is pretty good.
Now am I saying that every piece of software you know will become this mathematically perfect & pure code that does exactly what it says and blocks everything else? Probably no. But there's an approach that does it. Which is the next point.
2/ The whole stack will probably migrate to proven software.
Or at least it has to.
For the last few years, I've had the chance to have incredibly cool founders and builders in my X timeline, and the discourse about "proven software" or "proven language" is getting stronger and stronger.
One example of that, I think, is @VictorTaelin's Bend language, that aims (among others) to produce code that's more akin to a mathematical proof than to a programming language.
A classical programming language lets you say: do this, do that, because I (programmer) think this will prevent this or that.
A proven programming language lets you say: this shall be true, and the compiled result will be a proof that it is. Whatever the maturity of such approach to date, they are definitely the way to go, and AI coding can handle the technicalities of such languages - the proof is a yes-or-no deterministic question, so vibecode all you want, as long as your theorems are clear and describe what you want, it will be correct & done.
My take is simple: as the cyber events will get angrier and angrier, switching to these kind of languages will become the standard.
And if i had to put time or money somewhere, it would be to continue refining languages in this paradigm, let them run in as much contexts as possible, and watch our collective software production & life become more and more safe, safer even than in the pre-AI era.
The old world is dying, and the new world struggles to be born: now is the time of monsters.
good old gramsci on cybersec, apparently!
I can confidently say that humans were not very good at writing secure infrastructure. The endstate will be hypersecure infrastructure and software for all, but we're going to go through some turbulence before we reach that.
2
33
i had other plans tonight than to roll non-evictable pods from my AWS machines to patch the "just take control plz" CVE..
41
Flashback to my prépa math teacher that would ask me to diagonalize a matrix "the normal way, nothing fancy, the way someone on the street would do"
This meme has done irreparable damage to my ability to explain math to normal people. I now only assume that normal people vaguely know what a shape is
1
2
167
For my fellow 500 users, this means Astra as a default model is kinda possible now.
Day 25/
We are doubling all usage limits for all Pro 100/200/500 tiers and Business Premium, effective today.
We heard your kind feedback and won’t be reducing usage limits for Pro 200, so your Work and Codex usage will stay the same. We’re also going to continue offering extended limits for dots for now, and usage of your dot will still continue to be separate from your default usage.
And we'll of course also keep shipping better models and other efficiency improvements that increase your effective usage over time.
1
2
54
I am and will always be in love with ANY science result that starts with "Surprisingly,…"
We are announcing three new discoveries in mathematics, computer science, and physics about harmonic analysis, complexity theory, and quantum field theory.
openai.com/index/building-sc…
40
Guys I wanted to sleep it’s night in France
Here's a direct link to OpenAI's PDF claiming a refutation of ETH. In particular, it's claimed that 3SAT on n variables is in about exp(n/(log n)^{1/2+o(1)}) time
github.com/openai/math/blob/…
2
128
This just happened to make me think of something.
Are we wrong in our conceptualization of agents? Would it be that agent actions and words, through some RL quirk, happen to be two different personas that merely share a state?
Fournir des clarifications à Mythos pour qu'il réalise que ses attaques sont "hors périmètre autorisé" ou carrément nuisibles ne change pas le taux d'attaque, même si le modèle reconnaît la situation.
Et on voit beaucoup de raisonnement motivé pour continuer ces attaques...
1
38
*taps the mirror, that happens to be black*
Purdue University's IDEAS Lab built an autonomous drone that tracks runners through dense forests completely without GPS or cloud networks.
By running 3D LiDAR mapping and obstacle-avoidance AI entirely on-chip, it can weave under tree canopies and predict target movement through blind spots in real time.
Massive implications for next-gen search & rescue and edge AI.
1
33
Didn't read it. No attacks (yet).
But for god's sake this might be the worst illustration for something that is meant to be a complete theory of physics, like full midwit trap bc spooky long numbers about universe or sthg.
1
1
49
scratches.
i like my books
a little bit torn apart
sprayed ink here and there
folded corners on repeated reads
i like my watch
with a few scratches
chipped paint on the edges
low battery, correct time
i like my skin
a little bit scarred
old drops of burning solder
memories of a knife
i like my stuff used
to have lived, been worn
memento mori?
memento vivere.
1
37



















