The AI Rundown by Lightscape Partners – 03/18/24

Happy Monday and welcome back to this week’s edition of The AI Rundown by Lightscape Partners.

• In an eventful week for AI agents, Cognition debuted their AI software engineer named Devin, capable of handling entire projects independently given instructions.

• DeepMind also announced SIMA, an AI agent capable of turning natural language instructions into quality interactions in a 3D video-gaming environment.

• Following the $675M raise and partnership with OpenAI, Figure released a new video of their humanoid using ChatGPT to interact with a human and perform several tasks.

Check out the AI conferences happening this month here.

If you’d like a quick refresher on industry terms and the current landscape of AI, skip to the bottom.

Top AI Stories of the Week

Cognition announces Devin, an AI agent that can act as an independent software engineer. Link.

• The coding agent differentiates itself from current products like Github Copilot with the capability to handle entire development projects end-to-end.

• The startup emerged from stealth backed by Peter Thiel’s Founders Fund, ex-Twitter exec Elad Gil, and Doordash co-founder Tony Xu.

• A human gives the agent a natural language prompt and it begins development, debugging issues, testing, and all other processes involved in the development lifecycle.

• You can watch the announcement from founder and CEO, Scott Wu, here.

DeepMind announces SIMA, a new agent that can play video games from natural language instructions. Link.

• Google released their research on a “Scalable Instructable Multiworld Agent” (SIMA) that follows natural language instructions to perform tasks in video game environments.

• SIMA was trained and tested in collaboration with eight gaming studios on nine different video games. Some notable games include Goat Simulator 3 and No Man’s Sky.

• The team focused on more open games to learn general gaming skills, as opposed to narrative-based games.

Figure released their newest demo giving ChatGPT a body.

• In a new announcement and video, the robotics startup claimed the Figure 01 bot can now have full conversations with people using ChatGPT’s pre-trained model.

• The robot can describe its visual experience, use common reasoning, and explain the reasoning behind its actions.

• The demo follows Figure and OpenAI’s partnership in which they agreed to integrate specialized AI models into the Figure 01 model.

Venture

Proscia, an AI-powered pathology platform, announces a $46M Series C. Link.

• The digital pathology software provider announced a $9 million extension to the Series C, bringing the total to $46 million.

• The raise follows the startups recent FDA clearance for their software platform.

• The round was led by Highline Capital Management and Triangle Peak Partners.

Kaedim raises $15M to build AI-generated 3D assets. Link.

• The startup offering tools to help with 3D asset content creation announced a $15 million Series A funding round to better its platform and grow the team.

• The round was led by a16z Games.

Konko AI emerges from stealth to customize and scale leading AI language models. Link.

• The startup aims to streamline the process of discovering, customizing, and running the best AI language model for a user’s application.

Software + Hardware

Elon Musk’s AI firm xAI announces they will open-source Grok. Link.

• In another shot at OpenAI, Elon Musk announced his own AI startup xAI will be open-sourcing its foundational LLM, Grok.

• The announcement comes days after Musk’s lawsuit stating OpenAI and its founders abandoned their original mission.

• xAI follows Mistral and Meta, who have also open-sourced their models.

Covariant released a new model that allows physical robots to better understand language and reasoning. Link.

• Covariant announced their new model RFM-1, calling it “the world’s most capable Robotics Foundation Model (RFM) provides robots with a deep understanding of language and the physical world.”

• RFM-1 capabilities include:

  • Physics world model

  • Language-guided programming

  • Learning from self-reflection

• The capabilities are only available via live demonstration at Covariant’s headquarters and are made by appointment on a first-come, first-served basis.

Claude 3 Haiku, Anthropic's fast and affordable model, releases to the public. Link.

• The newest addition of the Claude 3 family of AI models, Haiku distinguishes itself as the fastest and cheapest model in its class.

• The model offers “advanced vision capabilities and strong performance on industry benchmarks.”

• The model is tailored for high-volume, latency-sensitive applications.

Upcoming AI Conferences

NVIDIA GTC 2024

• San Jose, CA and Virtual

• March 18 – 21, 2024

IMAGINE AI LIVE ‘24

• Fontainebleau, Las Vegas, NV

• March 27 – 28, 2024

Get up to speed on the current landscape

• AI Term Glossary

• Current AI Landscape from Sequoia Capital

Thank you for reading the AI Rundown by Lightscape Partners. Please send any questions, comments, or suggestions to [email protected].