Microsoft’s Copilot service is receiving several new features, including support for OpenAI’s GPT-4 Turbo model, an updated DALL-E 3 model for image creation, a code interpreter feature for accurate calculations and data analysis, and deep search functionality in Bing. These enhancements aim to improve Copilot’s understanding of queries, image quality, code generation, and search results.
I’ll highlight a piece in hacker news about CoPilot’s security. While security measures are in place, such as tenant isolation and training boundaries, organizations must ensure proper permissions, labels, and human review to prevent data breaches. Assessing and improving data security posture before implementing Copilot is recommended.
Microsoft Copilot has reached general availability, making it ready for use in production environments.
IBM and Meta have launched the AI Alliance, a collective of startups, organizations, and universities to advance AI innovation while ensuring safety, security, and diversity. The alliance includes over 50 members, including AMD, Dell Technologies, Hugging Face, Oracle, and Intel. The group plans to establish project standards and guidelines through a governing board and technical oversight committee. Notably absent from the alliance are Microsoft, OpenAI, Google, and AWS.
Perplexity AI has released its own AI large language models (LLMs), pplx-7b-online and pplx-70b-online, which aim to provide helpful, factual, and up-to-date information. These LLMs differentiate themselves from other leading models by offering current knowledge and combining a web index with a conversational AI chatbot interface. Perplexity’s models have been evaluated to perform better than OpenAI’s GPT-3.5 and Meta’s Llama 2 regarding freshness and factuality. The models are available through Perplexity’s API, with a subscription fee for access.
So many options! Your value to your customers goes up as complexity does, particularly as organizations must proactively manage permissions and conduct thorough reviews to mitigate data breach risks. Model differences lead to product differences, and this is rich territory for advisory services.
And the players themselves are looking to build frameworks that providers can leverage. All of this is why I continue to be excited for the AI space.