Photo credit: www.cnbc.com
During the Italian Tech Week 2024, Sam Altman, Co-founder and CEO of OpenAI, announced the launch of a groundbreaking artificial intelligence model capable of “thinking with images.” This new model enables it to comprehend and analyze user-created sketches and diagrams, even if they are of lower quality.
The newly introduced reasoning model, dubbed o3, was accompanied by a smaller variant known as o4-mini. This release comes on the heels of OpenAI’s earlier model, o1, which was unveiled in September and designed to tackle complex problems through multi-step reasoning.
With the capabilities of o3, users can now upload a range of visual materials, such as whiteboards and sketches, allowing the AI to analyze and elaborate on these images. Notably, the models also support various image-editing functionalities, including rotation and zooming.
Since the debut of its widely popular ChatGPT chatbot in late 2022, OpenAI has been on an aggressive path to enhance its AI models, expanding their capabilities beyond text to encompass images, voice, and video functionalities. As competition intensifies from industry players such as Google, Anthropic, and Elon Musk’s xAI, the urgency for innovative advancements in generative AI has escalated.
OpenAI highlighted that this is the first instance where its reasoning models are equipped to independently utilize the full suite of ChatGPT tools, including web browsing, Python execution, and both image understanding and generation. This integration allows for better handling of complex, multi-faceted problems and represents significant progress toward greater AI autonomy.
The company, recently valued at $300 billion during a funding round, emphasized that o3 and o4-mini are its inaugural models capable of integrating visual information into reasoning processes. This unique feature signifies a shift from mere image recognition to a deeper level of conceptual understanding.
Last month, OpenAI also unveiled a feature for native image generation, which garnered widespread attention for its ability to create artwork reminiscent of Studio Ghibli’s anime style.
The o3 model is particularly tailored for applications in mathematics, programming, and scientific analysis, while the o4-mini is designed for improved speed and efficiency at a reduced cost. Both models were made available to ChatGPT Plus, Pro, and Team subscribers starting Wednesday.
OpenAI’s community has often humorously critiqued the unusual names of its AI models. Aligned with this playful spirit, CEO Sam Altman took to X to suggest that the organization should refine its model naming system, inviting lighthearted teasing until this is addressed.
The company also assured users that both o3 and o4-mini underwent extensive stress testing through its most rigorous safety protocols to date. In conjunction with this, OpenAI updated its “Preparedness framework,” which offers guidance on safety measures.
Recently, OpenAI faced scrutiny concerning modifications to its safety procedures. In a statement, the company reserved the right to adjust its safety protocols in response to developments from “another frontier AI developer” that may pose similar risks.
In a notable policy change, OpenAI declared it would no longer mandate safety tests for certain specialized models. It also refrained from issuing a “model card” for its GPT-4.1 model, a report that typically outlines safety evaluations before a model’s public release. This followed a previous mid-February introduction of its AI agent tool, Deep Research, which preceded the publication of its system card.
As of now, OpenAI has not provided additional comments regarding these developments.
WATCH: OpenAI considering building social network
Source
www.cnbc.com