• Context Window
  • Posts
  • This Week in AI: Foundational models, “12 Days of Ship-mas” and unique experiments

This Week in AI: Foundational models, “12 Days of Ship-mas” and unique experiments

W49

New Foundational Models Launched
This week saw the release of two significant foundational models — large-scale AI systems pre-trained on vast datasets to handle a variety of tasks:

  1. DeepMind's Genie 2
    Google DeepMind unveiled Genie 2, a groundbreaking "world model" designed to create endless playable 3D environments. This tool allows researchers to train and test embodied AI agents in action-controllable, customizable settings. Learn more about Genie 2 here.

  2. Amazon's Nova Models
    Amazon introduced Nova, a versatile suite of foundational models capable of analyzing complex documents, interpreting charts and diagrams, understanding videos, generating content, and building sophisticated AI agents. Explore Nova in detail here.

OpenAI’s “12 Days of Ship-mas”
OpenAI kicked off a 12-day event starting December 5th, unveiling new features, tools, and demos daily:

  • Day 1: The updated o1 reasoning model moved out of preview, promising enhanced problem-solving for challenging tasks.

  • ChatGPT Pro Update: A new $200/month subscription offers unlimited access to GPT-4o, Advanced Voice Mode, and an exclusive version of o1. According to CEO Sam Altman, this model is tailored to tackle the hardest reasoning problems.

Competition in AI Reasoning Models
China’s DeepSeek released its DeepSeek-R1-Lite-Preview, narrowing the performance gap with OpenAI. This model ranks highly in reasoning, coding, and creative content generation. Their ChatGPT alternative, DeepSeek Chat, is reportedly free to use. Try DeepSeek Chat here.

AI Startups on the Rise
Former Stripe CTO David Singleton has raised $56 million for /dev/agents, a platform aiming to be the operating system for AI agents. Drawing parallels to the launch of Android, Singleton and his team are betting on a similar transformational moment in AI.

Freysa AI: A High-Stakes Adversarial Game
A new experimental AI agent, @freysa_ai, has captured attention with its unique setup:

  • Participants pay a fee to send Freysa a message, trying to convince it to release its prize pool to them.

  • If the message fails, the fee adds to the growing prize pool.

  • The cost to message Freysa increases exponentially as the prize pool grows.

The game recently concluded with a $47,000 prize claimed by a winning message, which is attached for reference.