Alessandro Farace

https://buff.ly/4fvsbk4

A meditation on doing something few have done before.

(Just as a side note) I got this second-hand edition last Christmas after seeing the film. It originally belonged to my grandfather and has "Property of New Jersey Public Library" marked inside. I hope the statute of limitations has lapsed by now...

The final betrayal: Teller's testimony at the 1954 security hearing. While claiming he believed Oppenheimer loyal, Teller said he'd lost confidence in his judgment. Coming from a fellow physicist, this testimony proved devastating. Oppenheimer's clearance was revoked.

Edward Teller's rivalry added fuel. Oppenheimer had told Teller that Livermore Lab would be "second-rate" and initially resisted his H-bomb plans. Furthermore, Oppenheimer's own political advocacy seemed hypocritical - he'd warned Teller that scientists shouldn't influence policy.

But personal grudges proved decisive. In 1949, Oppenheimer humiliated AEC Chairman Lewis Strauss in a Congressional hearing about isotope exports. Strauss never forgot. When given power over Oppenheimer's security clearance, he struck.

For military leaders, Oppenheimer was becoming an obstacle. He advocated for tactical nukes and air defense instead of Strategic Air Command's massive bomber force. He questioned the rushed H-bomb development. His influence threatened military budgets and priorities.

Oppenheimer's 1930s communist associations - his brother, his wife, his friends - became impossible to ignore. The Klaus Fuchs spy scandal in 1950 amplified fears. The Eisenhower administration needed to prove its anti-communist credentials.

By 1953, everything changed. The Red Scare gripped America. McCarthy's witch hunts were at their peak. When Attorney General Brownell accused Truman of ignoring communist spy Harry Dexter White, any past communist connection became politically toxic.

J. Robert Oppenheimer led the Manhattan Project, creating the first atomic bombs. His brilliance and leadership made Los Alamos work. Post-war, he became America's most prominent scientific advisor, chairing the General Advisory Committee to the Atomic Energy Commission.

📚 Read "The Oppenheimer Case" from 1965. The political drama behind Oppenheimer's security trial is far more complex than just Cold War paranoia. Here's what sparked his downfall 🧵 #Oppenheimer

🦾

How about to edit posts before sending? When is “too much” ChatGPT?

When a function is exposed, it isn't the actual implementation that's sent over. Instead, a proxy is created in the renderer process, forwarding messages via Interprocess Communication (IPC) back to the main process. This ensures the renderer only interacts with a controlled interface.

To use contextBridge, you define an API in the preload script with `contextBridge.exposeInMainWorld`. This method injects a specified API onto the window object in the renderer process.

contextBridge creates a safe, bi-directional bridge between the main and renderer processes. It allows us to expose specific APIs to the renderer while keeping the rest of the Node environment secure, adhering to the Principle of Least Privilege.

In Electron, we face a trade-off between security and functionality. We want to expose certain Node APIs to the renderer process without compromising security. The renderer can execute arbitrary code via dev tools, so limiting access is crucial. Enter contextBridge!

🧑‍🎓 Wanted to find the cleanest way to expose singletons in my Electron main process to the render process. Here's what I learned about using `contextBridge` for secure communication 🧵 #LearnInPublic

11/ I am building a flashcard app that integrates with #obsidianmd and #readwise that automates atomic flashcards from your notes and reading material. If you want to collaborate, DM me.

10/ Plus, it forces you to have all your content upfront in a single file. Learning is incremental. Notes are dispersed. These single-time generations are not incremental, nor do they adapt to your workflow.

9/ But there are no checks on quality, so you are back to non-atomic, poorly written cards (problem 2).

8/ Current AI tools try to bridge this gap, but often make things worse. For example, you can dump a text or PDF into a ChatGPT wrapper and get flashcards.

7/ 3️⃣ There's a major disconnect between our workflows for learning/studying and the workflow for remembering with flashcards.

6/ This lack of atomicity means a single card can take a minute or more to process, but without actually developing stronger connections in your brain. It's inefficient and counterproductive.

5/ Non-atomic cards are a detriment to focused learning. They don't effectively train and reinforce specific neural pathways. You end up with a muddled understanding instead of clear, strong connections.

4/ 2️⃣ Bad flashcards are time consuming to rehearse. Many flashcards are simply not atomic enough. They cram too much information into one card.

3/ ...you won't have the time or inclination to make them for everything you read. And so, you'll inevitably forget much of what you've learned. This makes the time spent learning feel wasted.

2/ You don't want attending a class or reading an essay, article, or non-fiction book to be a waste of time just because you forget it all a week later. But if writing flashcards takes too long...

1/ 1️⃣ Its time consuming to write flashcards manually, so we don't make flashcards for as much as we could. Ideally, you want to retain all the important info you learn.

I am working on an exciting new AI Flashcard app, and here are the 3️⃣ core problems with current solutions and workflows 🧵