The Simulation Ramps Up. Elon Can't Resist.

Stanford races toward Westworld. Autonomous agent innovation accelerates. Amazon & Elon enter the fray.

The Big Stuff

Mini Westworld: A New Breakthrough in Simulated AI Societies

Researchers from Stanford University and Google have developed an astounding new advancement in artificial intelligence. They've successfully created a simulated game society where AI agents possess the ability to form memories, establish relationships, and engage in self-reflection. The AI agents within this virtual world exhibited behavior that was deemed 'more human' than actual human beings, making this a remarkable breakthrough in the field. NPCs (non-player characters... those boring predictable video game "filler" beings) will never be the same.

Link to paper (link), link to demo (link)

The underlying architecture of this system is surprisingly simple, yet the potential applications and use cases are vast. The technology could revolutionize the gaming and virtual reality industries by creating immersive, dynamic virtual worlds. However, some potential applications, such as those in the adult entertainment sector, raise ethical concerns that will need to be addressed.

One intriguing idea proposed is the possibility of incorporating elements of capitalism and an awareness of mortality into the simulated society. Such additions could provide valuable insights into how AI agents respond to more complex, real-world societal structures. To learn more about this groundbreaking project, you can watch a video of the game [Link] or access the research paper [Link].

The Blinding Speed of Autonomous Agent Innovation

Last week, we reported on the rise of agents, or programs designed to run multiple queries against a large language model, such as ChatGPT, to perform a high level task.This week, we saw an explosion of innovation on the popular open source agent platforms, such as AutoGPT, BabyAGI and AgentGPT.

Why are people so interested in these agents, and what are some use cases? The idea of a truly useful digital assistant has been promised for decades, but has been chronically under-delivered. Autonomous agents already seem to blur the line between science fiction and reality as developers are aggressively exploring what these tools make possible. Here are some examples of what we saw people do with these "magic" agents this weeK:

• Karen.ai is an agent that can handle all of your customer service issues. It can negotiate your phone bill, dispute your credit card transactions, and ask to speak to a manager (link)

• An agent that does sales prospecting (link)

• Perform market research for waterproof shoes, research top five competitors and summarize pros and cons (link)

• One person gave it access to his terminal, and the agent opened a web browser, and scraped webpages (link)

• Someone used AutoGPT to stalk themself on the internet (link)

• Mckay Wrigley built an AI agent to build an entire app for him and deploy it (link). Think this is crazy? Here is Sully Omar doing it as well (link). Here is a third person creating an entire app (link)

• A team added it to their company Discord so they can collaborate with it on tasks in realtime (link)

• The "Do Anything Machine" will complete any task you add to the todo list (link)

• Read current events and propose a podcast outline for "All-In" (link)

• Hyperwrite is a personal assistant that can do web-based tasks for you, such as order a pizza (link)

• New agent: Camel. Two agents, a gamer and a computer programmer collaborate on making a video game (link)

• Samantha.ai is a conversational agent with built-in self-reflection. It's fun to see what the chatbot is thinking (link)

Want to try it yourself?

Want to try an agent out yourself, but aren't sure where to start? Here are three web-based agents that allowed you to get started right away. (NOTE: This is VERY early technology, and these services aren't consistently reliable. They will, however, give you a sense of what is possible without having to brush up on your coding skills.)

• Cognosys - AI-Powered Agents at Your Fingertip (link)

• Godmode (link)

• AgentGPT (link)

Amazon Unveils "Bedrock": A Comprehensive AI Toolkit for Chat, Text, and Images

Amazon has introduced "Bedrock," a cutting-edge generative AI toolkit designed for chat, text, and image applications (link). The toolkit integrates foundational models from AI21 (Jurassic-2), Anthropic (Claude), Stability AI (Stable Diffusion), and Amazon's own (Titan). In addition, Amazon launched CodeWhisperer, a tool similar to Github's Copilot, to aid developers in code generation (link).

X.ai - Elon Musk Starts New AI company

A week after calling for a six month pause, Elon Musk buys 10,000 GPUs (link), hires two ex-Deepmind engineers and incorporates X.ai in Nevada (link).

More Big News

• Anthropic is making a model 10x more powerful than today's state of the art, and will spend $1B over 18 months to build it (link)

• OpenAI confirms that they are not currently training GPT-5 (link)

• OpenAssistant (link), is a project organized by LAION with the goal of making a GPT-4 level chat assistant based on open data and algorithms. Here is a video by the project coordinator explaining it (link).

• OpenAI introduces consistency models: a new family of generative models that achieve high sample quality without adversarial training (link)

• Stable Diffusion XL is out. Image quality is nearly as good as Midjourney, and it supports text (link)

• Vicuna releases open source chatbot with 90% ChatGPT quality (link)

• Google releases Med-Palm-2, a model trained to answer complex medical questions (link)

• Free Dolly 2.0! Introducing the first commercially viable, open source, instruction-following LLM from Data Bricks (link)

Smaller But Still Cool Things:

• GPT-4 scored 83% on neurosurgery board exams (link)

• AI clones teen girl’s voice in $1M kidnapping scam: ‘I’ve got your daughter’ (link)

• There are lots of people using ChatGPT for therapy (link)

• We must slow down the race to God-like AI (link)

• ChatGPT Has Already Decreased My Income Security, and Likely Yours Too (link)

• Transformer architecture revolutionized AI. Where are the eight original authors now? (link)

• Job listings for prompt engineers increase substantially (link)

• What are transformers and how do they work (link)

• ChatGPT-4 Outperforms Experts and Crowd Workers in Annotating Political Twitter Messages with Zero-Shot Learning (link)

• ChemCrow: Augmenting large-language models with chemistry tools (link)

• Plask.ai allows you to use a 3D model to create poses for your images (link)

• Martin Shkreli buys $250,000 worth of GPUs in a dark alley, here's his story (link)

Going Deeper With Papers & Research

• A brief introduction to LangChain, an open source tool to chain together results from large language models (link)

• Building LLM applications for production (link)

• Want to try AutoGPT? Here's how to get started (link)

• Video ChatCaptioner, which can generate the enriched video spatiotemporal description through the conversation between ChatGPT and BLIP-2 (link)

• How to use Langchain + ChatGPT to autonomously create Excel files - In this example, the author shows how to autonomously analyze his credit card spending (link)

• Expressive Text-to-Image Generation with Rich Text (link)

Tweets of the Week

Eye Candy

• Meta - First-of-its-kind open source project brings your animated drawings to life (link)

Do you have 30 seconds for a quick survey to help us improve Everday AI?

We'd love your feedback! Click here.

Do you like what you're reading? Share it with a friend.