In 2025, the path to AGI (artificial general intelligence) seems more complicated than ever, partly due to the cost implication and vast resources, including the computing power required for developing the coveted benchmark.
In 2024, several AGI predictions were made, including OpenAI CEO Sam Altman's version, which indicated that AGI would be achieved within the next 5 years with current hardware, further citing that it would whoosh by with little societal impact. As it turns out, Google AI Studio's product manager, Logan Kilpatrick, recently shared some interesting insights about AGI and how it could be achieved (via Business Insider).
Straight shot to ASI is looking more and more probable by the month… this is what Ilya sawDecember 30, 2024
The Google executive indicated that a "straight shot to ASI is looking more and more probable by the month." He further added that Artificial Super Intelligence is achievable without a keen focus on intermediate products or model releases while talking about former OpenAI chief scientist Ilya Sutskever's vision for Safe Superintelligence Inc. Kilpatrick further claimed that the chief scientist aims to hit the coveted milestone with a straight shot.
Kilpatrick admits Sutskever's straight-shot vision for ASI seemed like a long shot and "unlikely to work since if you get the flywheel spinning on models/products, you can build a real moat." However, it's beginning to show great promise, especially with the success of scaling test time computing. As such, the straight shot approach and scaling greater heights while building toward ASI increasingly look more plausible.
We are still going to get AGI, but unlike the consensus from 4 years ago that it would be this inflection point moment in history, it’s likely going to just look a lot like a product release, with many iterations and similar options in the market within a short period of time (which fwiw is likely the best outcome for humanity, so personally happy about this).
Google AI Studio PM, Logan Kilpatrick
Towards the end of last year, OpenAI shipped its o1 reasoning model to broad availability, prompting a technical employee to allude that the ChatGPT maker had achieved the coveted AGI benchmark. "In my opinion, we have already achieved AGI, and it’s even clearer with O1," the employee added. "We have not achieved 'better than any human at any task,' but what we have is “better than most humans at most tasks.”
Kilpatrick shared some insights about AGI, stating that it will still be achieved. However, "it's likely going to just look a lot like a product release." The executive statement builds on Sam Altman's previous remarks about AGI simply whooshing by with little societal impact. Altman further indicated that the safety concerns expressed by users about the milestone won't be experienced during the AGI moment. Kilpatrick comparing the AGI benchmark to a simple product release suggests it might not be as iconic as previously thought.
This news comes as reports suggest top AI labs, including OpenAI, Google, and Anthropic, are struggling to develop sophisticated AI models due to a lack of high-quality content for model training. However, Sam Altman dismissed the claims, indicating "there's no wall."
Former Google CEO Eric Schmidt reiterated Altman's sentiments, highlighting that no evidence suggests scaling laws have begun to stop. Perhaps this could be why AI companies are shipping models with sophisticated reasoning capabilities, allowing them to think through complex scenarios and tasks like humans rather than relying on information from the internet.
