Alexis Rondeau

Bookmarks

August 2026 · 98 bookmarks

Bookmarked from Evan Luthra @EvanLuthra

🚨 BERLIN JUST SPENT €3.9 MILLION ON AI CAMERAS THAT SCAN HOW YOU MOVE. AN ARTIST BUILT AN $88 SHIRT THAT MAKES THE SOFTWARE STOP SEEING A PERSON.

About 30 cameras went up at Kottbusser Tor this month, the first permanent AI surveillance of public space in Berlin outside the subway. Third German city to do it, after Mannheim and Hamburg.

It isn't face recognition, which is the part everyone gets wrong. The software turns every person in frame into a skeleton, a bundle of moving vectors, then scans those vectors for movements the police defined in advance as suspicious.

Lying down. Falling. Staggering. Kicking. Hitting.

Cross the threshold and it alarms a human operator, who pulls up the real video and decides whether to send anyone. Officials sold this as privacy-friendly on the grounds that you're only ever a stick figure to it. Police say the software comes from a firm called Adesso.

Simon Weckert, a Berlin artist, read all that and made a shirt.

It's loud, covered in pink and orange and green blobs, the kind of thing your uncle wears to a barbecue thinking he looks great. Point a detector at Weckert in normal clothes and it draws a green box around him labeled "person." He holds the shirt up to his chest and the box vanishes.

Object detection never learned what a human is. It learned what humans statistically look like across millions of labeled photos: sharp edges, contrast, the head-and-shoulders silhouette, limbs in the right proportion to a torso, an outline that stays continuous against whatever's behind it. Weckert's pattern floods those early layers with saturated noise and shatters the outline into pieces, and the model's confidence that anyone is standing there drops below the threshold where it bothers drawing a box.

He found the pattern by running a generative model in a loop against the detector until it surfaced the blind spot. The AI taught the shirt how to hide from AI.

The print runs unbroken across the buttons and seams, so a fold or a turn doesn't kill it. That was the flaw in earlier adversarial patches, which stopped working the second the fabric bunched.

Then there's the law, which is where this gets strange. Germany bans covering your face at public gatherings. This shirt covers nothing. Your face is fully visible to any officer standing in front of you, and to anyone reviewing the footage later. The statute governs what people can see. It says nothing about what a neural network can.

The technology also has a record. Hamburg tested the same class of behavior scanner and the police report showed roughly 1,400 alerts in two months, of which eleven were police-relevant. An earlier three-month run threw over a thousand alarms, about one every hour. One percent mattered to police. 0.1 percent mattered to a criminal case.

Weckert doesn't oversell it. He's said plainly that no pattern beats every camera and that this isn't a tool for evading police.

A system nobody outside the police can independently test is now reading a public square 24 hours a day, and $88 of printed polyester turns it off.

Bookmarked from Michael 🇩🇪 @Bundeskanz50246

Das ist so unfassbar dumm, was in Berlin passiert ist mit dem Hack. Es wurden 5,79 TB Daten gestohlen, darunter 46.500 Verträge, E-Mails, Telefonnummern, Passwörter, Dateien etc.
Berlin wird jetzt erpresst, im Darknet gibt es ein Angebot diese Daten für 30 Bitcoin zu kaufen. Die Passwörter wurden in ungesicherten Word-Dateien abgespeichert (siehe Bild). So etwas darf einfach nicht passieren.

Bookmarked from signüll @signulll

when you meet someone, you can almost always tell which social media platform fine tuned their brain / thoughts. esp in nyc.

e.g. most ppl likely now have an insta personality / aesthetic outlook (esp women), or a reddit personality (likely dudes), or maybe even an x personality (weirdos) to some extent. i guess chatgpt is likely the next personality influencer too.

you are a product of your thoughts, & your thoughts are the product of your feed & of likely next the personal ai agent you let in to your life.

Bookmarked from Mathematica @mathemetica

This pattern never repeats, and it still has fivefold symmetry.

Two rhombi do the work: a thick 72° tile and a thin 36° one. Turn it 72° around the center and it matches itself. Slide it any distance and it will not.

Inflate by φ = (1+√5)/2 and a copy of the same tiling appears inside, because φ² = φ + 1. Quasicrystals found the same loophole in real matter.

Bookmarked from signüll @signulll

before ai it was cognitively expensive to be a programmer cuz you had to really use brain cycles for everything, from architecting, to even using the right variable names etc. it was genuinely draining.

& if you programmed for a whole day, you were fucking tired. even code reviews required you to think deeply & you had to be aware of the entire code base to some extent too. so not only did you have to have context loaded, you had to do manual inference if you will.

ai completely eliminated all of this strain.

Bookmarked from signüll @signulll

talked to a guy from anthropic for a long time last night. fascinating conversation.

& believe it or not, one of their biggest problems is still hiring.

there simply aren’t enough exceptional ppl, esp for long term high conviction bets where the stuff is not obvious yet. i.e. people who can independently generate good judgment under extreme ambiguity.

Bookmarked from conor brennan-burke @contextconor

one of the strangest things about moving from new york to sf was realizing how often extremely important things happen here in settings that don't feel important at all

i lived in new york for four years. i sometimes spent entire days moving icons around in powerpoint, but i was doing it in a skyscraper with great views, surrounded by attractive people in patagonia vests. afterward i'd walk onto a packed street and go to a beautiful restaurant

even when the work itself was mundane, everything around it told you it mattered

sf is almost the opposite

the blue bottle in south park feels like a normal coffee shop. it is also a place where an absurd amount of capital gets allocated

sand hill road is one of the most powerful financial centers in the world and looks like a collection of dentist offices

you'll meet someone dressed like they just rolled out of bed who can't make eye contact and discover later that they control billions of dollars or have built a product you use every day

i've been to parties where almost everyone had raised millions of dollars or built products used by millions of people. in another city, each person would have seemed like a big deal. in sf, they were just standing around a cramped apartment at a "party" that honestly wasn't even that good

i used to live in a closet in a slightly dilapidated hacker house with creaky floors and old furniture. people who lived there went on to create companies worth tens of billions of dollars

new york puts a lot of effort into making importance visible. sf often seems actively uninterested in doing that

some of this is probably because sf is more focused on creation than consumption. some of it is the personality of builders, who often care much more about the thing they're making than the room they're making it in

but i think the lack of polish also lowers the barrier to entry. you don't need to look important before doing something important here

new york makes ordinary work feel important

sf makes important work feel ordinary

Bookmarked from no.mind @the_no_mind

Shopify billionaire co-founder & CEO Tobi Lütke — one of the most analytically rigorous founders alive — uses affirmations.

His exact words:

"If you tell yourself or write down something about yourself 100 times, it writes into the neurofrontal cortex at such a deep level that your brain will start reconciling you to that. It just works."

His personal example:

He was terrified of public speaking.

For one week he sat down every day for 10 minutes and wrote: "I love public speaking."

A week later the fear was gone.

"It's not like a placebo. You just actively change your neurofrontal cortex in this moment."

Your brain starts shaping your behavior around the identity you rehearse most.

Bookmarked from Zara Zhang @zarazhangrui

The gap between a 10x AI user and a mediocre one comes down to two abilities:

1. Domain expertise. You know what a great result looks like because you've done the work for years. This is your ceiling, and it becomes the agent's ceiling
2. Translation. You can pass your taste and standards to the agent through context, references, and iteration

The second is a skill, learnable in weeks.

The first is a career. No AI will hand it to you.

Bookmarked from trish @TrisH0x2A

you can build a working HTTP server in x86-64 assembly

just Linux syscalls

create a socket, bind it to 127.0.0.1:8000, listen, accept connections, write an HTTP response, close the socket

the entire server is just raw assembly talking directly to the kernel

Bookmarked from Tony Weston @tonywestonuk

The Apple Neural Engine wasn't supposed to be used for anything other than AI. And then someone worked out the API.

And then I took that proof of concept and got claude to build a donut rotator with it. For no reason.

But, hey, a 38 trillion ops/second piece of silicon gotta be useful for something non-ai related? @Fudmottin could get it to count primes !

Bookmarked from Jasper Lortije @JasperLort

We built a high-speed jet interceptor in two weeks. Meet the first SABLE prototype.

As incoming drones become cheaper and faster, air defense needs to catch up.

SABLE and HUNTER are the cornerstones of our Air Defense as a Service (ADaaS) offering, built to protect critical infrastructure.

@ycombinator teaches us to “build something people want.” We build something people need.

@impactdrones

Bookmarked from alexei @alexeixbt

the secret is, it's actually really simple.

if you want to be a writer, write everyday.
if you want to be a runner, run everyday.
if you want to build a business, solve a problem, and see if someone will pay you for it, everyday.

most people overcomplicate things because it's easier to chalk it up as "impossible" than wake up everyday and do the boring daily activities that get you 1% closer to where you want to be.

Bookmarked from Mathematica @mathemetica

Four quantum numbers alone force the chemical elements into this diamond lattice of shells and orbitals. The integer n labels each shell.

s, p, d and f orbitals hold 2, 6, 10, 14 electrons from the quantum number l.

This geometry solves the multi-electron Schrödinger equation. It dictates semiconductors in chips, catalysts in chemistry, and materials for quantum devices.

What appears as a table of matter is pure quantum mathematics made visible.

Bookmarked from Lara Avci @laraavci_

Beglaubigt.de turning German company formation into a ChatGPT app is such a smart distribution move.

People already ask ChatGPT how to set up a company. Their API and MCP server let the app collect the details for a GmbH or UG inside the conversation and then return a secure checkout link. Beglaubigt.de coordinates the notary appointment, Handelsregister filing and formation documents.

Instead of waiting for founders to find its website, the service can now appear at the moment they express intent.

Bookmarked from R136a1 @TheEnergyStory

I found a previously undocumented passive Windows backdoor I’m calling SLEEPWALKER. Its defining feature is a custom command language with 23 instructions. Instead of receiving conventional commands, it runs encrypted bytecode programs that can combine scheduling, nested programs, staged payload delivery with SHA-256 verification, decompression, network communication and in-memory shellcode execution.

The analyzed file is an unsigned DLL that masquerades as Microsoft's dpapi.dll, uses forged ESET metadata and activates only when side-loaded by ESET's Management Agent program ERAAgent.exe. SLEEPWALKER does not beacon or open an obvious listening port on its own. It monitors network traffic and stays dormant until it detects a specially crafted packet carrying an AES-256-CCM-protected task. It supports TCP, UDP, ICMP, SMB named pipes and VMware VMCI. VMCI can carry traffic directly between a VMware guest and host without using a conventional network interface. SLEEPWALKER also implements a DNS trigger that uses Base32 to carry encrypted tasks inside DNS labels, although the analyzed sample does not enable it.

The full analysis documents the bytecode format, all 23 instructions, encryption, DNS encoding, trigger logic and network transports. It also includes IOCs, a YARA rule and a PowerShell scanner.

If you believe you have been targeted by SLEEPWALKER or have found a related sample, please contact me. I created the SLEEPWALKER Analysis Toolkit to decode its bytecode, examine artifacts and safely reproduce its receiving pipeline without executing commands or transmitting data. If you suspect you've been infected, I am happy to share this toolkit with you, alongside a mitigation guide providing investigation steps and a remediation script for use after detection.

Bookmarked from German Simply 🇩🇪 @GermanSimply_

One word to carry into your week. 🇩🇪

𝗧𝗮𝘁𝗸𝗿𝗮𝗳𝘁.

𝘛𝘢𝘵 = deed. 𝘒𝘳𝘢𝘧𝘵 = strength.

Action-strength. The energy that moves something from a plan into a finished thing.

Motivation comes and goes, but 𝘛𝘢𝘵𝘬𝘳𝘢𝘧𝘵 is what people notice in you after the motivation has gone.

𝘚𝘪𝘦 𝘩𝘢𝘵 𝘦𝘤𝘩𝘵𝘦 𝘛𝘢𝘵𝘬𝘳𝘢𝘧𝘵.
She has real drive.

Said about the person who stops discussing the thing and goes and does it.

𝗟𝗲𝗮𝗿𝗻 𝗚𝗲𝗿𝗺𝗮𝗻 𝗦𝗶𝗺𝗽𝗹𝘆 🇩🇪

Bookmarked from Ivan Sparrow @ivesparrowai

Saving you hours of your time.

Every time I see another post on X like “how to make $10k/month with an app in 1 month”, there’s a 99% chance it all comes down to:

  • iOS only
  • hard paywalls
  • higher prices
  • product doesn’t matter
  • UGC organic on TikTok
  • videos that go viral organically, then run them as TikTok ads

And honestly, writing and reposting the same articles with zero practical value over and over again feels disrespectful to all of us.

Bookmarked from David Senra @davidsenra

.@sama says the best advice Peter Thiel ever gave him was to focus on what’s working.

Peter was the one who convinced him to double down on ChatGPT early on:

“After we launched ChatGPT, it was sort of this weird thing because people didn't really know what to use it for. It was growing super fast, but it felt very unstable, or kind of almost low-value growth.”

“A lot of people in the company were like, ‘We gotta figure out something else. This is not sustainable value.’ And we were talking about this list of 5 or 6 other things that we could focus on instead.”

“He was like, ‘It's an obvious mistake to do anything but this.’”

“Besides the fact that it's growing, which is rare and great, and it was not growing as fast then as it did after, he's like, ‘The power of this is the power of the Google text box. It's like a text box you can type anything into, and it does the right thing.’”

“He's like, ‘People have just been chasing the Google business model for 20 years and this is the first thing that's come up, and clearly the empty text box worked for Google, so why don't you just double down on that? It's growing. It's very flexible and it has all of the signs other than it doesn't fit the current Silicon Valley wisdom.’”

“And I was like, ‘Okay.’ And so we went super hard on ChatGPT and it was great.”

Bookmarked from Ernesto Lopez @ErnestoSOFTWARE

#145 on the appstore 7 days after launching our new app.

Building an app is not complicated

→ it took us 72 hours for this one
→ yet looks insanely high quality

Marketing an app is not complicated

→ Found a viral formatc
→ UGC + Spark Ads
→ and the app is already printing

Stop overcomplicating it

If you launch an app this week

And do the same play book
we always talk about

in the next 30-90 days there's
a very high chance

You'll be making above $10k/mo

Bookmarked from Tom Yeh @ProfTomYeh

Transformer by hand ✍️ ~ 6 steps walkthrough below

Studying the Transformer architecture is like opening up the engine hood of your car. So many unknown parts: embeddings, positional encoding, attention weighting, self-attention, cross-attention, multi-head attention, layer norm, skip connections, softmax, linear, Nx, shifted right, query, key, value, masking.

But which of those actually make the car run?

I think at the core, there are two most essential components:
Attention weighting and the feed-forward network.

Everything else is an enhancement to make it run faster and longer, which is how we got from a car to a truck, and to the word "large" in large language model.

So I drew and calculated those two parts entirely by hand.

Goal: push five features through one transformer block, filling in every cell yourself.

1. Given

Five positions of input features, arriving from the previous block.

2. Attention matrix

Let us feed all five features to a query-key module (QK) and read back an attention weight matrix, A. The details of that module are a post of their own.

3. Attention weighting

We multiply the input features by A to get the attention weighted features, Z. Still five positions. The effect is to combine features across positions, horizontally: X1 becomes X1 + X2, X2 becomes X2 + X3, and so on.

4. First layer

Let us feed all five weighted features into the first layer of the FFN. Multiply by the weights and biases. This time the combining happens across feature dimensions, vertically, and each feature grows from 3 numbers to 4.

Note that every position goes through the same weight matrix. That is what "position-wise" means.

5. ReLU

We cross out the negatives. They become zeros.

6. Second layer

Let us bring it back down: 4 dimensions to 3. The output feeds the next block, which has a completely separate set of parameters, and the whole thing runs again.

You have just calculated a transformer block by hand. ✍️

The takeaway: the two parts are doing two different jobs, and neither one alone is enough. Attention mixes across positions, so a feature can see its neighbours. The FFN mixes across feature dimensions, so each position can think about itself. Horizontal, then vertical. Then that pattern repeats N times, each block with its own separate set of weights. That is the Nx from the list up top, and that is what makes the transformer run.

💾 Save this post!

#AIbyHand #Transformers #DeepLearning

Bookmarked from Ernesto Lopez @ErnestoSOFTWARE

Nah son this is crazy😭

$200k/mo alarm clock app has
44 onboarding screens

they need 44 screens to convince you to pay for something

that is free on your iphone.

but it works. consumers have no limit
to their consumerism

→ 200,000 downloads
→Viral clips doing 10.1M views.
→and ultimately $200,000 in rev

you genuenly need to stop overthinking your app idea.

launch a wheather app, make it cooler

next thing you know you
are printing $100k/mo

Bookmarked from Mathematica @mathemetica

These curves turn linear algebra into intelligence.

Without activation functions a deep network is just one big matrix multiply.

σ(x)=1/(1+e⁻ˣ) squeezes to (0,1); tanh(x) centers at zero; ReLU=max(0,x) zeros the negatives; Leaky ReLU, ELU and Maxout keep a gentle slope so gradients never die.
They are the reason neural nets recognize faces, write code and drive cars.

The simplest nonlinearities often unlock the deepest learning.
Which of these do you reach for first when a new model refuses to train?

Bookmarked from AA @measure_plan

thanks for all the love on this, i'll publish for you to play soon

in the meantime here's a demo of some new mechanics:

  • bending light around a black hole
  • splitting with a prism
  • focusing with a convex lens

more physics - puzzles - levels coming soon :]

Bookmarked from signüll @signulll

we’re using local apple intelligence models a lot these days in our product for a lot of little things & they work incredibly well, especially for personalization & copy.

you can cache these so you create a pretty lovely little experience in subtle places where the app feels tailored than just one to many (all for free!).

it’ll feel cool as hell when they get even smarter.

Bookmarked from Chris McCormick @ChrisJMcCormick

I've started writing my model code as a single, compilable forward_backward function:

1 - No more autograd. The backward math is written out, same as forward.
2 - No torch.nn module abstractions, just matrices and math. (Glorious!)
3 - Attach a .gacc to each parameter to store its gradients.
4 - Intermediate activations are explicitly stashed for use in backward.

The screenshot shows the activation stash for the 24-layer nanochat architecture, where Claude cleverly skipped a few with high footprint and low compute cost.

Most notably, the MLP needs the activation values both pre and post-ReLU, and they're 18GB each! We just store pre, and re-apply ReLU in backward.

(A recent modded-nanogpt PR did something similar-but-different, need to look at it more)

I'm not familiar with the backward math (Claude wrote it all), but I'm excited to gradually build up the same familiarity with it as the forward path.

Saved on Hacker News · SyneRyder on GLM-5.3: Frontier coding with emergent cyber capabilities

> ... Anthropic's Project Glasswing is supposed to find them quite a while ago?

That was my thought too. For all of Anthropic's talk about their "adversaries", it seems Z.AI have been quietly offering fixes for single shot Remote Code Execution flaws in US software (Safari / WebKit) that Apple and Glasswing / Mythos missed, and that Apple would not attribute to GLM.

Bookmarked from Paul Graham @paulg

Most successful startups consist of a wonderfully popular product more than compensating for many mistakes by the founders. You have to have that wonderfully popular product or there is too little margin for error. But if you do, it will buy your way out of all kinds of trouble.

Bookmarked from trish @TrisH0x2A

Robert van Engelen wrote a complete Lisp interpreter in just 99 lines of C

it includes 21 built-in primitives garbage collection and an interactive REPL

to prove how small it is he ran it on a Sharp pocket computer from the 1990s

the source is free and every line is explained in an accompanying article

Bookmarked from Athrix ☄️ @athrix_codes

Hiring | OpenAI | Remote & On-site

OpenAI is hiring across 753 open roles in research, engineering, operations, and more.

These are full-time opportunities for curious and driven people who want to build safe, beneficial AI systems, solve challenging problems, and help shape the future of technology at a global scale.

What you’ll work on

  • Advance frontier AI research and model capabilities
  • Build and scale AI/ML infrastructure, distributed systems, and developer platforms
  • Develop products and experiences across ChatGPT, Codex, APIs, and emerging AI technologies
  • Design reliable backend systems, data pipelines, and large-scale cloud infrastructure
  • Improve AI safety, security, alignment, evaluations, and responsible deployment
  • Build hardware and compute infrastructure for training and serving advanced AI models
  • Create intuitive product, design, and user experiences powered by AI
  • Drive decisions across policy, legal, finance, communications, recruiting, and business operations

Teams hiring

  • AI Research
  • Software Engineering
  • Applied AI Engineering
  • Machine Learning Infrastructure
  • Data Science and Analytics
  • Product and Design
  • Safety, Security, and Trust
  • Hardware and Compute Infrastructure
  • Sales and Partnerships
  • Technical Success and Support
  • Artificial Intelligence and Machine Learning
  • Software Engineering
  • Python, JavaScript, and Systems Programming
  • Distributed Systems and Cloud Infrastructure
  • Data Engineering and Data Science
  • LLM Training, Evaluation, and Inference
  • Product Management and Product Design
  • Cybersecurity and Trust & Safety
  • Hardware, Networking, and Data Centers
  • Sales, Marketing, Finance, Legal, and Operations

Role details

  • Remote and on-site opportunities
  • Openings across 2 remote regions
  • Roles across 20+ on-site locations worldwide
  • 753 open positions across multiple teams
  • Full time opportunities
  • Roles available across different experience levels
  • Compensation and requirements vary by position and location

Interested one lmk👇🏻
I'll share the application link.