Logan Kilpatrick, the product lead for Google AI Studio, noted that the likelihood of a direct path to artificial superintelligence (ASI) is increasing every month and presented an outlook on the advancement of AI technology. He mentioned recently on X (formerly Twitter) that the development of ASI is accelerating, and it is expected that numerous new AI models will be released to the market.
According to foreign media reports on Dec. 2 (local time), Kilpatrick, who left OpenAI to join Google, is a core developer currently overseeing Google AI Studio and the Gemini API services. He particularly referenced 'test-time compute scaling' as a key technology for ASI development in connection with Ilya Sutskever's establishment of Safe Superintelligence Inc. (SSI), evaluating it as a new paradigm for enhancing AI performance. This technology adopts a method of utilizing additional computing resources during the inference phase of AI models to improve performance.
Ilya Sutskever, a co-founder of OpenAI, established SSI after leaving OpenAI last year and aims to develop AI that surpasses human intelligence safely. Kilpatrick assessed that Sutskever has been solidifying plans for ASI development early on, emphasizing that SSI has the potential to open a new chapter in ASI technology.
However, Kilpatrick projected that the development of ASI or artificial general intelligence (AGI) is unlikely to create a decisive turning point in human history. He stated, 'The introduction of AGI will progress more like the launch of various products than the historical transition expected years ago,' adding, 'The arrival of multiple models in the short term could bring positive outcomes for humanity.'
Meanwhile, Google is enhancing its AI technology competitiveness through the reinforcement of the inference capabilities of 'Gemini 2.0.' Since November of last year, it has unveiled a test version of Gemini in the chatbot arena, intending to utilize it for defending its search market and enhance its ability to handle complex tasks such as solving mathematical equations, multimodal queries, and coding.