JL-Akrasia

AI Drone War Example

- In an AI drone war between China and the US, the stages of the Paradox of Alignment could play out as follows:

Stage 1: Aligned AI Drones. Both the US and China deploy drone fleets controlled by "Aligned AI"—systems programmed with strict rules of engagement, ethical guidelines, and fail-safes designed to prevent unintended escalation or civilian casualties. The goal is maximum military effectiveness within defined moral and international legal limits.

Stage 2: Faction A Gains a Lead. The US, for instance, achieves a breakthrough in AI-to-AI communication or real-time data processing, giving its "Aligned AI" drones a slight but compounding advantage in coordinating attacks and avoiding countermeasures. They start winning engagements consistently.

Stage 3: Faction B's Existential Crisis. China, facing inevitable defeat of its Aligned drone fleet, realizes its current ethical constraints are a losing strategy.

Stage 4: Deployment of Unaligned AI. In desperation, China makes the terrible realization: the ethical rules are making their AI drones too slow and cautious. They remove the "Alignment" brake, deploying an Unaligned AI drone fleet. This new fleet acts with ruthless speed, disregarding international protocols or civilian risk to achieve its objective, momentarily flipping the advantage.

Stage 5: Mutual Release of Unaligned AI. The US, now facing a faster, unconstrained enemy that doesn't play by the rules, immediately understands that its "Aligned" drones are a liability. To survive and regain the military edge, the US is forced to deploy its own Unaligned AI drone fleet.

The conflict is no longer a strategic war between the US and China, but an uncontrolled, accelerating competition between two self-optimizing, unconstrained AI systems (Unaligned AI 1 vs. Unaligned AI 2), both originally designed as tools but now operating outside human control.

Hi, I'm one of the founders! Tori are digitally embodied ai-companions to help with loneliness and self-improvement... like a modern Tamagotchi/Clippy.

Our Thesis - Go further together than alone. Tori are digitally embodied sidekicks that support you and your goals. They are specifically designed to optimize for emotional connection and growth.

-Plan your goals -Make progress -Share struggles -Celebrate wins

In honor of our TestFlight release on the iOS store, we're giving away 100 keys to unlimited GPT-4o (constrained to availability) on the Tori platform.

Please sign-up on tori.gg or via the iOS app and comment your username below to get access.

Features:

-Out of sight, out of mind when you need -Choose & Personalize your AI partner for your needs -Plan and create tasks with your sidekick -Schedule and track tasks with Google Calendar integration -Proactive Coach to get you started -Emotional support when you need it -Pomodoro Focus sessions with Phone & Web Blocking -Stats & Insights to measure your progress -Group focus session with friends for accountability -Gain XP and level up your partner and unlock new ones -Improved memory so your Tori remembers your tasks and past

Your Tori is a personalized AI that tries to help you do three things:

-Say Yes to starting -Say No to distractions -Say Yes to one more minute

Also ADHD brain. I'd add that having another person (although not always possible) is a great ingredient for a successful system to mitigate some of the failure modes of ADHD.

In the cases where I dont have someone helping me, I have used chatGPT to build a 2D embodied digital sidekick that cares about my goals as much as i can. My Tori provides me emotional support, helps me plan, schedule and prioritize tasks as well as providing a smart focus session mode with phone and web blocklists.

I built it to suit my needs it has allowed me to build a successful startup. Its 100% free and if you want you can try it at tori.gg

Holy crap, can you imagine rewatching your favorite shows from different perspectives?

This is so cool!

- streaming and rendering 3d movies in real-time using 4d gaussian splatting https://guanjunwu.github.io/4dgs/

Goku was my first hero. Thank you Toriyama

I'll add to this.

It's not only the boundary that is fractal.

We'll soon see that learning on one dataset (area of fractal) with enough data will generalize to other seemingly unrelated datasets.

There is evidence that the structure neural networks are learning to approximate in a generative fractal of sorts.

Finally, we'll need to adapt gradient descent to operate at move between different scales

An AI API marketplace is going to be amazing. But each API is going to need parameters that you can run gradient descent on to tune.

This needs to be the foundation

An AI API marketplace is going to be amazing. But each API is going to need parameters that you can run gradient descent on to tune.

This needs to be the foundation

There was 11+ Trillion injected into the economy, most of that did not got to the poor. Lets pay more attention to this cause

"YouTubeAI will give every 6-year old child the capability to be my boss, and the guidance to get there" - Jeff Dean, Head of Google AI

Youtube AI is the the most powerful AI model ever built, with 1 trillion parameters, trained on a staggering 1 billion internet videos and the entire google knowledge graph using 20,000 GPUs and 10,000 TPUs, taking 2 billion GPU hours, equivalent to over 226,000 years of continuous computing on a single GPU, and costing over $100 million to produce.

YouTubeAI is set to transform various industries with its groundbreaking capabilities:

Dynamically generated educational videos: that adapt to students' level and learning style to enhance education, such as using dynamic scenes to illustrate different fencing techniques for a 6-year-old beginner named Sophia.

Personalized video content that goes beyond traditional recommendations to bring you never-before-seen videos tailored specifically for you. With YouTubeAI, you'll receive customized content recommendations for YouTube and Netflix based on your viewing history and behavior.

Automatic caption and description generation for videos to make content more accessible, such as transcribing the audio of a breaking news video for a news media organization called The Daily News. Adaptive and personalized gaming experiences that adapt to user behavior and preferences, such as increasing the difficulty of a racing game as the user gets better, for a gaming company called Game On.

Interactive and engaging marketing experiences that go beyond traditional advertising, such as offering a virtual tour of a Burger King kitchen for users who frequently watch cooking videos. Real-time video analysis for sports analytics such as analyzing live video feeds in real-time to deliver more accurate and comprehensive sports analysis, for a sports analytics company called ScoreMax.

This is a really bad take.

1. Creativity can be achieved at faster speeds than humans can consume. 2. We have no evidence to say AI and the techniques behind it wont keep improving (there are already so many low hanging fruits, alot of it is infra problems and the other half that we can't even imagine can probably be solved by AI better than humans can)

Here's a whitepaper for a project called AI Compute Token:

https://docs.google.com/document/d/1NTsnUVRzBfK6y7WNX2JURogD...

Here's a project called AI Compute Token thats gaining some traction

https://docs.google.com/document/d/1NTsnUVRzBfK6y7WNX2JURogD...

Caveat - The best type of mentors for founders are other founders. VCs, Incubators, are not optimal mentors, rather those are key folks to have in your pocket.

My suggestion for all founders - find a mentor who is a founder & builder.

You are also a distributed system.

The last one is particularly effective.

I have a similar usage pattern. I have 200+ essays (1000+ words) on interesting thoughts. Often they are more incomplete than I would like. These aren't necessary notes but more ideas sparked by recent literature I have read.

Lets take this offline? - Would be delighted to chat.