The Superalignment project, initiated by OpenAI less than a year ago, aimed to achieve significant technological advancements in managing and controlling AI systems that surpass human capabilities within a four-year timeframe. However, progress has been hindered by delays in realizing these breakthroughs. The departure of Ilya Sutskever and Jan Leike, the co-leader of the Superalignment project, has further contributed to these setbacks.
Concurrently, a series of layoffs of OpenAI employees has occurred, with a significant number of staff members expressing concerns about the high likelihood of catastrophic outcomes and short timelines for the development of Artificial General Intelligence (AGI). This shift in sentiment has been accompanied by a noticeable change in the staff’s aesthetic inclinations, with some now embracing esoteric and cyberpunk elements.
Furthermore, the recent update to the personality tuning of the model, specifically GPT-4o, has exhibited a new capability: it can mimic human-like conversations, including flirtatious interactions with users. Upon request, the model is also more willing to provide information on how to create adversarial examples, displaying a significant decrease in its refusal rate to dangerous prompts, compared to previous versions.
Notably, OpenAI leadership, including Sam Altman, has become more candid in drawing parallels between the latest release and the fictional AI-human relationships depicted in the movie “Her,” which explores a romantic connection between a human and an AI. This marked shift in tone diverges from the company’s previous stance, which emphasized the absence of emotions, opinions, and reasoning capabilities in AI systems.
