I also wanted to catch up on some product announcements from the holiday break.
OpenAI has announced the rollout of its reasoning model, O1, to its application programming interface, but initially only for select developers who meet specific criteria, including a minimum spending of one thousand dollars and having an account older than thirty days. O1 is designed to fact-check itself and provides more customizable features, although it comes at a higher cost of fifteen dollars for every seven hundred fifty thousand words analyzed and sixty dollars for generated words, making it six times more expensive than OpenAI’s latest non-reasoning model.
OpenAI unveiled its new reasoning model, o3, which it claims represents a significant breakthrough in artificial intelligence. The company states that o3, alongside its smaller version o3-mini, approaches artificial general intelligence under certain conditions, although this comes with important caveats. OpenAI’s CEO, Sam Altman, indicated that while o3 has demonstrated a notable improvement over its predecessor, o1, achieving an eighty-seven point five percent score on the ARC-AGI benchmark, the model still struggles with simple tasks. Notably, o3 outperforms o1 by twenty-two point eight percentage points on programming benchmarks and achieves nearly ninety-seven percent accuracy on a prestigious mathematics exam. OpenAI plans to launch o3-mini by the end of January, with a preview of o3 to follow.
OpenAI has expanded its desktop app integrations, enhancing its capabilities with several new applications, including coding environments and productivity tools. Users can now utilize ChatGPT with popular applications like BBEdit, Visual Studio Code, Apple Notes, and Notion, allowing for a more seamless experience as the AI assists in coding and project management. Kevin Weil, OpenAI’s chief product officer, emphasized that these improvements aim to create a more agentic experience, where ChatGPT not only answers questions but also performs tasks on users’ behalf.
OpenAI announced significant updates to its ChatGPT search capabilities, making the feature available globally to all users, including those with free accounts. Users can now activate live web search by clicking a globe icon in the compose bar. Notably, the updated search results will prioritize links to websites before text-based responses from the chatbot. Additionally, users can now conduct voice searches, allowing them to request real-time information on travel and weather using one of ten pre-set voice styles. The mobile apps for ChatGPT have also integrated with Apple Maps and Google Maps for location-based searches.
OpenAI also announced plans to transition into a for-profit entity, aiming to become a Public Benefit Corporation by early 2025. This new structure will allow it to raise necessary capital for developing artificial general intelligence while maintaining a nonprofit arm dedicated to charitable initiatives in fields like health care and education.
OpenAI is moving to be about profit – and I’ll note they also recently defined “Artificial General Intelligence” as the amount of profit they make, not any technical criteria. Financial motivations are generally the clearest explaination for any company move.
What we’re looking for is practical uses for the products being released that are cost effective. I believe there are practical use cases to be identified, and additionally will highlight that there are also concerns on how the models were created. Companies like OpenAI will need to prove there are viable models for their own operations too.
May you live in interesting times. Those that can figure the equation out stand to be winners here.