The AI Rundown by Lightscape Partners – 05/06/24

Ukraine has an AI spokesperson, CoreWeave triples its valuation, and Sam Altman provides a glimpse into the future of ChatGPT and AGI.

Ted Wagner
May 06, 2024

Credit: Ukraine Ministry of Foreign Affairs

Good morning and welcome back to this week’s AI Rundown by Lightscape Partners.

Last week in AI:

Ukraine unveiled the first AI spokesperson for a nation, Victoria Shi. The spokesperson will handle easy inquiries, freeing up time for human diplomats during wartime.
CoreWeave raised $1.1 billion, tripling its valuation in five months as the demand for GPU capacity continues to skyrocket.
In a talk at Stanford University, Sam Altman heavily criticized the quality of GPT-4, claiming each new model will be drastically better. The OpenAI CEO continued to say more about the state of AI and the ongoing demand for compute power.

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

Otherwise, let’s get into it.

Watch the first video here.
The digital human is being deployed as a time-saving measure to free up human diplomats during wartime.
Ukraine claims its AI spokesperson is a historic first, noting that this is the first of many efforts to incorporate AI into Ukraine’s diplomatic efforts.
The AI’s likeness is based on a Ukrainian singer and influencer, Rosalie Nombre.

The Nvidia-backed company nearly triples its valuation to $19 billion, up from $7 billion five months ago.
The company began acquiring GPUs to mine crypto coins before pivoting into AI computing.
The fundraising and valuation highlight the excessive demand for GPU capacity as the AI race only speeds up.

Altman claimed GPT-4 will be the “dumbest model any of you will have to use again…by a lot.”
The OpenAI CEO stated the cost of AGI is irrelevant, claiming any cost to create AGI is worth the economic upside.
Watch the full talk here.

On Monday of last week, DeepMind and Google released a paper describing Med-Gemini, a group of fine-tuned AI models for healthcare.
While in a research phase, Med-Gemini is allegedly outperforming GPT-4.
The biggest breakthrough is the model’s ability to capture context and temporality, such as understanding the background and setting of symptoms and timing the sequence of their onset.
The announcement comes a week after OpenAI announced its partnership with Moderna.

Overfitting occurs when an algorithm fits too closely to its training data, resulting in a model that can’t make accurate predictions or conclusions.
The study found that among other LLMs, Mistral, the French startup open-sourcing its foundation model, had the worst overfitting among the models tested.
GPT-4, Claude, Gemini, and Llama showed little signs of overfitting.

Copilot Workspace allows developers to brainstorm, plan, build, test, and run code all in natural language.
The new platform leverages all the Copilot-powered agents built to date and works across the entire software development life cycle.

Amazon Q is a generative AI-powered assistant “for accelerating software development and leveraging companies’ internal data.
The assistant can generate, test, and debug code, and has multi-step planning and reasoning capabilities.
Besides coding, Amazon Q can answer questions across business data “such as company policies, product information, business results, code base, employees, and many other topics by connecting to enterprise data repositories.”

The eight US newspapers allege that millions of copyrighted news articles have been used without permission to train their LLMs.
The Chicago Tribune and Denver Post are two of the 8 newspapers involved in the lawsuit.
The New York Times sued OpenAI and Microsoft in December 2023 alleging illegal use of copyrighted material.