OpenAI is racing to launch a new AI product, code-named Strawberry, which aims to enhance reasoning capabilities in chatbots, particularly in solving complex problems like math and programming. This initiative is part of OpenAI’s strategy to maintain its competitive edge in conversational AI, especially with the upcoming launch of its flagship model, Orion. Strawberry could also help generate high-quality training data to reduce errors in AI outputs. The technology was demonstrated to U.S. national security officials and was previously known as Q*. This explains the strawberry emojis bouncing around the internet.
Google has expanded its Gemini 1.5 AI models with three new experimental versions, including a smaller model and improved Pro and Flash variants. These models can process extensive multimodal inputs and achieve near-perfect retrieval rates. While the reception has been mixed, users can test the models for free in Google AI Studio and the Gemini API.
Anthropic has published the system prompts for its AI models, including Claude 3 Opus, aiming for transparency and ethical practices in AI development. These prompts outline the models’ capabilities and limitations, such as avoiding facial recognition and maintaining impartiality on controversial topics. Anthropic plans to regularly update and disclose changes to these prompts, potentially pressuring competitors to follow suit.
Why do we care?
There are two significant changes here. OpenAI showed this to the US government before release, and Anthropic published its system prompts. That’s my primary focus. Assume the models themselves become commodities. How each company approaches regulation and security will be a competitive difference. OpenAI is electing to manage regulation concerns by consulting with the government, which is a new strategy. So is Anthropic bringing more transparency to their prompts.