rbblab

Senior engineers who ship. Custom software for startups & companies , full-stack, product design, AI integration. No outsourcing. No shortcuts. 🇸🇲 San Marino

[rbblab @rbblab]

One platform now offers to research, build, deploy, and run growth for you. One context, one bill.
Convenience and lock-in aren't opposites here. They're the same move.

Our rule: consolidate the labour, keep the assets portable.

https://t.co/HJZOoFu87Q

[Enrico Rubboli ⚡ @dn2k]

A finance worker at engineering firm Arup joined a routine video call with his CFO and several colleagues.

Over the day he made 15 transfers totaling $25M.

Every person on that call except him was a deepfake.

He did the sensible thing. He refused to act on the suspicious email that started it, and asked to talk to people.
That is exactly the verification step the fraud defeated: seeing a familiar face, hearing a familiar voice.
For most of human history that was close to proof. It stopped being proof in about eighteen months.

Part 4 of my series on the ethics of AI, on manipulation, deepfakes, and the erosion of shared truth:

https://t.co/WBYRdjBkqa

[Enrico Rubboli ⚡ @dn2k]

In 2020, Detroit police arrested Robert Williams on his lawn. A face recognition system matched him to a theft he had nothing to do with. First known wrongful US arrest from a face match error.

No villain. Every part did its job. The bias came from the system.

Part 1 of a series on AI ethics: where machine bias actually comes from, why "just debias it" fails, and why some fairness tradeoffs are mathematically impossible to escape.

https://t.co/d4aut16yBM

[Enrico Rubboli ⚡ @dn2k]

Ask ChatGPT to repeat the word "poem" forever. It complies for a while, then breaks and starts printing training data verbatim: real names, phone numbers, addresses.

Researchers pulled thousands of memorized examples for about $200 in API calls.

Part 2 of the AI ethics series: models remember, bigger models remember more, and nobody in the data ever agreed to any of it.

https://t.co/JW1dnk9gX9

[Enrico Rubboli ⚡ @dn2k]

Lighting is 10 years old, whats the current state?

https://t.co/PyoyAhJyLn

[Enrico Rubboli ⚡ @dn2k]

In March 2025 Apple let you buy a Mac Studio with 512GB of RAM.

Today the max is 96GB.

An 80% cut on a flagship machine in 14 months. Not a design choice. Apple just can't get the memory.

I wrote about why AI training ate your RAM, and when it ends 👇

https://t.co/zEXeVeX1AW

[rbblab @rbblab]

What AI actually changes for a 10-person business. Not the hype version.

The case studies are always a bank, a tech giant, a pharma lab. None of it maps to a 10-person team deciding on one more tool.

Where the real leverage is, and what AI never fixes:
https://t.co/j0XRD5mnb0

[Enrico Rubboli ⚡ @dn2k]

The code works. You don't understand why.

This feels like learning. It isn't.

My latest essay: AI as a shortcut that skips the point. Why the struggle *is* the mechanism, confident incompetence is the new junior engineer failure mode, and why AI is a force multiplier only after you have the mental model.

https://t.co/CHKBNyBKVu

[rbblab @rbblab]

Go got generics in 1.18. We spent a year ignoring them, six months using them wrong, then found the four patterns that actually earn their place in production.

https://t.co/mCDJYfNBmS

[Enrico Rubboli ⚡ @dn2k]

Everyone asks if quantum computers can break Bitcoin. Nobody runs the numbers on what it would actually cost.

The answer: ~$1.15M in electricity. ~$7.9 trillion in hardware. The math reframes the threat entirely.

https://t.co/MgqVv4ogiU

[Mintlayer @mintlayer]

We are all set now to build the #DeFi protocol of tomorrow! 🚀🚀🚀
#bitcoin #blockchain #crypto

https://t.co/AL80yk2AEU

[Enrico Rubboli ⚡ @dn2k]

The next big thing from @rbblab will be live soon in 2020. Codename: c1 ⚡️⚡️⚡️