🕹️ AI Powered DOOM
It’s estimated that the 1993 game Doom was installed on more computers than Windows 95 at one point.
To this day, Doom has a huge modding community. It propelled shareware style distributions, birthed video game speedrunning, and popularized network multiplayer.
The full game was just 2.39 Megabytes.
Today’s Ai5:
- 🕹️ AI Powered DOOM
- 🤖 OpenAI Are Releasing an Agent
- 🎭 Midjourney Personality Test
- 📱 Claude Artifacts on Mobile
- 📜 Everything to Know About SB 1047
Prompt of the Day 🎨
Are you old enough to remember him?
AI Powered DOOM 🕹️
A new paper from Google entitled Diffusion Models are Real-Time Game Engines might just change gaming forever.
In the paper, researchers present GameNGen. It’s a diffusion model that produces playable gameplay at 20FPS ran from a single chip.
Continuing the tradition of adapting the 1993 game DOOM to platforms it wasn’t meant for, GameNGen works by having an AI agent learn to play the game with reinforcement learning. Following this, a diffusion model is trained to produce the next frame in the game, conditioned on what it had previously learned.
They found that human tests are only slightly better at distinguishing short clips of the actual game from clips of the AI generated simulation.
Can you tell the difference? 👇
OpenAI Are Releasing an Agent 🤖
Ok so cards on the table, OpenAI haven’t explicitly said they’re releasing an AI agent. But, I’m predicting their next big release will be a powerful AI agent… here’s why. 👇
According to a recent report, OpenAI is cooking up two models codenamed Strawberry (which we’ve already heard about) and a new model, Orion.
Strawberry is said to be a math and programming powerhouse, capable of solving complex problems with enhanced logic and reasoning.
But here’s where it gets interesting: internal OpenAI documents reveal plans to use Strawberry for autonomous internet searches, allowing the AI to plan ahead and conduct in-depth research. In other words, they’re building an AI agent.
The new model Orion is being developed as GPT4’s successor. The interesting thing is that Strawberry will produce synthetic data to train Orion.
AI training AI, what could go wrong!?
Anyway, an AI that can reason through complex tasks, conduct its own research, and communicate at levels surpassing GPT4? It’s not hard to imagine this evolving into an AI assistant that will most likely be better than anything we’ve seen in the past.
Midjourney Personality Test 🎭
Midjourney are doing their second survey to understand the structure of personality and it’s relationship to our perception of beauty.
The survey is available for anyone to try. It takes about 10 minutes and gives you statistics on nearly 200 different measures of personality!
The results are actually really in depth, and you might just learn something about yourself along the way!
Midjourney are giving fast GPU hours away to anyone who takes the survey. You can take the test for yourself here!
Claude Artifacts on Mobile 📱
Anthropic just announced that Claude Artifacts are now available on iOS and Android. If you’re on the fence about how powerful this type of technology is, check this out.
I created that working snake game on the second attempt (the first one worked, but I had to ask for a d-pad.)
Considering this is the first version of the tech, image where it might be in 12 months?
In other Anthropic news, they’ve recently published the system prompts for each Claude model. It’s pretty standard, the instructions basically telling Claude to think through tasks, and reassure the user it might be wrong or hallucinating if unsure.
It also instructs Claude to be “Face Blind”, meaning that Claude will refuse to identify people in pictures, even if the AI recognizes them.
It’s an interesting move from Anthropic because knowing the system prompt can make the manipulation of the AI much easier.
Everything to Know About SB 1047 📜
The Safe and Secure Innovation for Frontier AI Models Act (SB 1047) just passed, here’s what that means for AI.
The bill aims to regulate the development and use of powerful AI models. This includes AI models trained using a significant amount of computing power (over 10^26 operations) or costing over $100 million to train (most LLMs.)
The bill requires developers of these models to implement strict safety and security protocols, including:
- The ability to enact full shutdowns
- Regular third-party audits
- Whistleblower protections
From what I can tell the bill seems reasonable. It was supported by Elon Musk and Anthropic, while OpenAI opposed the bill and suggested their own version.
Others are saying the bill will remove competitive advantage from AI companies in the US and stifle startups.
Snack Sized 5 🍪
1️⃣ This AI generated movie trailer has nearly 150,000 views in 3 weeks on YouTube.
2️⃣ Check out these Jellyfish figures made in the night sky with 1,000 drones. (YouTube)
3️⃣ Neural Frames have been shipping a bunch of updates including a new UI. If you want to make an AI powered music video, this is for you.*
4️⃣ Midjourney announce they’re “officially getting into hardware“. Keen eye spots Midjourney hired an engineer previously working on Apple Vision Pro. (X)
5️⃣ This guy asked ChatGPT, Claude, and Gemini to draw pictures of themselves, then checked to see if they could recognize each other. The results are interesting! (X)
Share Ai5 and get Rewards!
Join the conversation on X.
Connect on LinkedIn.
Read the Blog on Rareconnections.
Feedback? Yes please! matt@myai5.com
All sponsored links are marked with an asterisk*