Latest Interview with Google CEO: Google Excels in Multimodal Abilities, Yet Falls Short in Coding
8 hour ago / Read about 0 minute
Author:小编   

In an exclusive interview, Sundar Pichai, the CEO of Google, disclosed that Google's models stand at the forefront of the industry in intelligent domains, including text processing, multimodal input handling, speech and audio processing, as well as general reasoning. Nevertheless, there remain certain deficiencies in areas such as agentic programming, instruction adherence, and long-horizon task management, which the team is diligently working to enhance. The recently launched Gemini 3.5 Flash model exhibits issues related to artifacts and performance deterioration, issues that Google pledges to promptly resolve through subsequent training sessions and will progressively ease usage constraints. Google will not hastily convert its search function entirely to an AI model; the sources and links featured in search results will be preserved for the foreseeable future, and the business model will persist in amalgamating subscriptions and advertising. The agent-based product, Spark, is slated for launch this summer, and Google intends to adopt a phased promotional strategy to garner user trust and guarantee safety. Furthermore, Google is making its TPU computing power accessible to competitors to sustain its hardware supremacy and achieve economies of scale. Pichai is of the opinion that Artificial General Intelligence (AGI) represents an inevitable trajectory, with technological advancements progressing at a pace surpassing initial expectations and the realization node drawing near, thus necessitating proactive societal preparation.