"The first was what he called "continual learning," saying that the systems are frozen based on the training they received before implementation. "What you'd like is for those systems to continually learn online from experience, to learn from the context they're in, maybe personalize to the situation and the tasks that you have for them," he said during the discussion."
""So, for example, today's systems can get gold medals in the international Math Olympiad, really hard problems, but sometimes can still make mistakes on elementary maths if you pose the question in a certain way," he said. "A true general intelligence system shouldn't have that kind of jaggedness." Humans, in comparison, would not make mistakes on an easy math problem if they were math experts, he added."
True artificial general intelligence remains incomplete, with current systems unable to learn continuously, plan long-term, or deliver consistent performance. Systems are typically frozen after initial training and do not learn online from ongoing experience or personalize to context and tasks. Present models can plan across short horizons but lack multi-year strategic planning capabilities. Performance is uneven: systems can solve very difficult problems yet still make elementary errors depending on phrasing. Achieving true AGI requires developing continual online learning, robust long-term planning mechanisms, and consistent competence across domains. Timelines for seeing operational AGI remain measured in years, but capability gaps persist.
Read at Business Insider
Unable to calculate read time
Collection
[
|
...
]