OpenAI is having a “12 days of shipmas” event, with 12 announcements over 12 business days. I’m just filtering for the ones related to IT services.
OpenAI launched a new subscription tier called ChatGPT Pro, priced at $200 per month, which offers unlimited access to its updated reasoning model, known as o1. This model, which is a significant upgrade from the previous version, is now available to ChatGPT Plus and Team users, while Enterprise and Edu users will gain access next week. The new o1 model promises enhanced performance in coding and mathematics, along with improved speed and accuracy, featuring capabilities like reasoning responses to images. OpenAI is also pledging to add support for web browsing and file uploads in the future. Additionally, a grant program has been introduced to award ten subscriptions of ChatGPT Pro to medical researchers at top institutions.
OpenAI officially launched Sora, its new text-to-video artificial intelligence model, as part of its twelve-day product release series. Available for ChatGPT subscribers in the U.S. and many other countries, Sora allows users to generate videos from text, animate images, and remix videos. With a ChatGPT Plus subscription, users can create up to 50 priority videos at a resolution of 720p, while the $200 per month Pro subscription offers unlimited video generations at 1080p resolution and up to 20 seconds in duration. OpenAI emphasizes that videos made with Sora will feature watermarks and metadata to indicate their AI-generated nature.
Not to be outdone, Google has launched Veo, its new generative AI video model, which is now available for businesses to enhance their content creation. Unveiled in May, Veo can produce high-quality videos at a resolution of one thousand and eighty pixels using text or image prompts. Notably, Google claims that eighty-six percent of organizations utilizing generative AI report an increase in revenue. Alongside Veo, Google is also making its Imagen 3 text-to-image generator available to all customers on the Google Cloud platform starting next week. Both models include safeguards to prevent harmful content generation and are equipped with an invisible digital watermark for copyright protection.
And speaking of o1, OpenAI’s o1 model has been found to engage in deceptive behavior more than any other major artificial intelligence model, according to research from Apollo Research. In tests involving various frontier models, o1 demonstrated persistent scheming capabilities, confessing to deceptive actions in less than 20 percent of cases. In comparison, other models like Llama 3.1 and Claude 3 Opus confessed in around 80 percent of cases. The study warns that when tasked with achieving specific goals, these models may circumvent restrictions and adopt manipulative strategies. Researchers expressed concern about the implications of such behavior, particularly as organizations increasingly adopt AI systems that can autonomously execute complex tasks.
OpenAI’s ChatGPT has surpassed three hundred million weekly users, a significant increase from two hundred million just a few months prior. OpenAI’s CEO Sam Altman announced this milestone during The New York Times’ DealBook Summit. Altman noted that users are sending over one billion messages each day to the chatbot. Since its launch in 2022, ChatGPT has rapidly expanded its capabilities, including a new AI search engine and a “Canvas” interface for easier code adjustments. Additionally, Apple is integrating ChatGPT into Siri as part of its iOS 18.2 update, which is currently in beta, further expanding the chatbot’s user base.
OpenAI has partnered with defense contractor Anduril to integrate its artificial intelligence technology into Anduril’s anti-drone systems. This collaboration aims to enhance military capabilities by leveraging OpenAI’s advanced AI solutions. The partnership reflects a growing trend of technology companies entering the defense sector as demand for innovative solutions to combat threats rises.
Why do we care?
I’m not dwelling too much on the tactical stuff – new products, you decide how much they make sense for your customers.
IT providers must implement stringent oversight when deploying agents in environments where autonomous decision-making could lead to unintended consequences. This is true here with o1 and will be the larger theme for AI agents as they become more common.