Giacomo Orizzonte

VP Infrastructure. AI amplifies judgment; it doesn't replace it. Notes on infra, leadership and working with AI.

[Giacomo Orizzonte @giacomozz]

AI should make you faster, never leave you understanding less. If it works and nobody can explain why, you haven't saved time. You've borrowed it.

[Giacomo Orizzonte @giacomozz]

@swyx @affaan @absurdistphil @aiDotEngineer Curious how they price reliability. A cheap GPU-hour that fails mid-run isn't cheap.

[Giacomo Orizzonte @giacomozz]

@DeepLearningAI Code ships with tests. Most operating decisions ship with a postmortem months later.

[Giacomo Orizzonte @giacomozz]

@OpenAI Choosing the next experiment is where lab judgment lives. Good place for a second opinion.

[Giacomo Orizzonte @giacomozz]

@GoogleDeepMind Kernel tuning used to wait for rare specialists. A throughput metric makes it delegable.

[Giacomo Orizzonte @giacomozz]

@DeepLearningAI Adapting from about 200 demos is impressive. I'd wait for RoboArena before calling it general.

[Giacomo Orizzonte @giacomozz]

@swyx The screens clone easily. Integrations, edge cases and support history are what keep customers.

[Giacomo Orizzonte @giacomozz]

@AnthropicAI Git, CI and review assumed human pace. Agents find those hidden assumptions first.

[Giacomo Orizzonte @giacomozz]

@GoogleDeepMind @GeminiApp @antigravity @GoogleAIStudio For enterprises, which channel a model arrives through matters more than the launch benchmark.

[Giacomo Orizzonte @giacomozz]

@emollick Journals need a severity queue, not just a correction inbox, before this scales.

[Giacomo Orizzonte @giacomozz]

Foundation is one of my favorite books. Hari Seldon could predict the fate of an empire, but not of a single person.

That's how I see AI in longevity: a radar for patterns across millions of lives, not an oracle.

Your own life is still the Mule.

[Giacomo Orizzonte @giacomozz]

@AnthropicAI The value is the backlog nobody had time for, done while the scientist kept steering.

[Giacomo Orizzonte @giacomozz]

@MRezaRouhi Yes...

[Giacomo Orizzonte @giacomozz]

Asimov wrote the Three Laws and then spent decades writing stories about how they fail.

We keep reading him as a blueprint. He was writing a warning: rules inside the machine are the first thing to break.

Put the controls where you can see them.

[Giacomo Orizzonte @giacomozz]

@MRezaRouhi Agree on the teenager bit. They will touch the outside world. My point is narrower: don't put your only brakes inside the thing that wants to leave. Watch from outside, where you can still cut the power.

[Giacomo Orizzonte @giacomozz]

@DeepLearningAI Cheaper reads change agent design: keep more context, re-fetch less, summarize less aggressively.

[Giacomo Orizzonte @giacomozz]

@sama Once answers become interfaces, the bar shifts from reading well to working well.

[Giacomo Orizzonte @giacomozz]

@DeepLearningAI Skipping the relay matters beyond latency: tone and hesitation stop getting lost in transcription.

[Giacomo Orizzonte @giacomozz]

@gdb If this holds up, math research gains breadth: hundreds of results, not one deep problem.

[Giacomo Orizzonte @giacomozz]

@simonw Competent game music from a prompt is a bigger craft surprise than another fluent chat demo.

[Giacomo Orizzonte @giacomozz]

@GoogleDeepMind One shared embedding space for code and media makes on-device search actually cross-modal.