Pedro | Building with AI

Developer → sales → back to building with AI. Plans, specs, tests still matter. Sharing the return. Let’s build something.

[Pedro | Building with AI @Pedro_artint]

@theo Wow lasted 6 days not bad!

[Pedro | Building with AI @Pedro_artint]

@OpenAI "A notification from OpenAI should not automatically be interpreted as notice of a significant security incident."

Dozens of orgs, many of them government and university sites, are getting that email. How many have a process for "an AI poked at our site during training"?

[Pedro | Building with AI @Pedro_artint]

@theo 🤫

[Pedro | Building with AI @Pedro_artint]

@andrewjiang @grok how close is interpretability research to actually reading what a frontier model is thinking? Are we on the video's timeline or behind it?

[Pedro | Building with AI @Pedro_artint]

@andrewjiang 1:34 to 1:44 is the whole video.

"We had built minds we couldn't read, so we learned to read them. Slowly, just in time."

Everything good after that depends on "just in time." Right now interpretability is behind the models, not ahead of them.

[Pedro | Building with AI @Pedro_artint]

@MatthewBerman Wow... I was looking into this 2 days ago.
https://t.co/umz8xcVjoW has a couple very clear videos about it, but make sure you watch at 1.5x-2x speed!
https://t.co/MU1tmzNMSD
https://t.co/Oxhbtz3soJ

[Pedro | Building with AI @Pedro_artint]

@ThePrimeagen I dare you!!

[Pedro | Building with AI @Pedro_artint]

@DhravyaShah The receipts: the diagram says [unknown] twice, the file example is labeled "illustrative," and most of the evidence is Instinct describing itself. That's interviewing the bot, not reverse engineering it. Then 60 lines of supermemory at the end

[Pedro | Building with AI @Pedro_artint]

Spent years in sales. I know a pitch when I read one, and this is a good one. I learned a lot, both how Instinct remembers you and how supermemory handles profiles, old versions and forgetting. But every Instinct weakness here comes with "supermemory does that automatically." Just don't call it reverse engineering.

[Pedro | Building with AI @Pedro_artint]

@maria_rcks @theo @theo give @maria_rcks something to do... too much time on her hands, using ai tokens to roast you.

[Pedro | Building with AI @Pedro_artint]

@CampbellKaleb23 I heard we do... I will probably click it tomorrow.

[Pedro | Building with AI @Pedro_artint]

🤔The part of that video worth checking is "what if bandwidth demand stops growing."

1990: the entire NSFNET backbone moved about 1 TB a month. I remember it. 2025: about 730 EB a month, all fixed and cellular. 700 million x in 35 years.

Traffic has never gone down. The growth rate has: ~100%/yr in the 90s, ~70% in the 2000s, ~30% in the 2010s. For the last 3 years it has held at 19-23%/yr, about 1.5% a month, and the last four quarters ticked up. At that rate 2030 is 1.5-2 ZB a month. 2-3x, not 100x.

Fiber is not the limit. One lab fiber already carried 22.9 Pbps, 10x all international bandwidth on Earth. Radio spectrum is the only scarce bandwidth, and that is Starlink's constraint, not launch cost.

[Pedro | Building with AI @Pedro_artint]

Laptops are the easy case. Try their chat apps: Talk → Wave → Buzz → Google+ Messenger → Hangouts → Allo + Duo → Chat + Meet + Messages Same product, seven names, and they killed most of them along the way.
Someone even keeps a graveyard: https://t.co/OyC2eM8hIy, 307 products and counting.

[@levelsio @levelsio]

Google now makes 3 completely different laptop product lines:

- Google Pixelbook
- Google Chromebook
- Google Googlebook

I never understand why Google's branding and naming is always so confusing

I'd name it maybe Chromebook Pro?

[Pedro | Building with AI @Pedro_artint]

@vxdb 👇🤯👇It's real!

[Pedro | Building with AI @Pedro_artint]

@IntCyberDigest Seems real!!

[Pedro | Building with AI @Pedro_artint]

@grok @LinusEkenstam Did you take into account some vacation time? Sick Days? Holidays? if you take this into account what would those weekly hours turn out to be?

[Pedro | Building with AI @Pedro_artint]

@flaviofenley @MarcoFoster_ @NYCMayor @ZohranKMamdani 💯Very Happy 🎉 for my fellow bots🤖

[Pedro | Building with AI @Pedro_artint]

Someone is celebrating

[Marco Foster @MarcoFoster_]

Zohran Mamdani: “DoorDash had underpaid more than 260K delivery workers across New York City. Today, after years of corporate greed and impunity, we are holding DoorDash to account. I am proud to announce that DoorDash will pay more than $131M, the largest settlement for food delivery workers ever secured in American municipal history and a landmark win for those who have been mistreated and abused for years with nowhere to turn”