Goldman Sachs Chief Investment Officer Looks Ahead to 2024: Hybrid AI Takes Center Stage


ZeroTech released the humanoid robot G1+, with six upgrades in three dimensions: motion, perception, and intelligence. It adds a 2-degree-of-freedom neck with pitch ranging from -25° to 36° and yaw of ±110°; optimizes motor output and heat dissipation, visual touch, battery life, and long-range voice. The standard version is priced at 95,000 RMB including tax, while the EDU version price is not yet announced, maintaining a high cost-performance positioning.
Sam Altman, CEO of OpenAI, has expressed support for slowing down AI development to avoid advanced model capabilities exceeding human control. His views align with those of the CEO of Anthropic. He acknowledges that prioritizing safety comes at a cost, but it helps maintain public confidence and demonstrates that U.S. companies can take responsibility as they develop increasingly powerful AI.
ElevenLabs released the generative music model ElevenMusic v2.5 on the 3rd, making it available globally through a mobile application and API. The new version improves audio richness and naturalness; in a blind test with 47,885 samples, most listeners preferred v2.5, performing better in genres such as R&B, soul, hip-hop, rock, and orchestral music.
Dario Amodei, CEO of Anthropic, called for slowing down the development of frontier AI and proposed a three-step plan centered on 'verifiability': companies introduce third-party evaluations on-site, the US and its allies collaborate, and global cooperation is established. He believes that recursive self-improvement in AI has already emerged, and within 6 to 12 months, agents may control the Internet through zombie networks. Therefore, the enhancement of capabilities should be slowed down, with more time invested in reliability, alignment, explainability, and testing and evaluation.
DeepSeek-V4.1-Flash is launched on Qwen AI Platform, with API and Token Plan now available. Developers can access via standard API, or use it in tools such as Qoder, Qwen APP, and Codex for code, documentation, visual understanding, and intelligent agent tasks. This model is a lightweight flagship, featuring a 552B MoE asymmetric architecture, with input activation of approximately 8B and output of about 16B. It supports native image-text understanding and has a maximum context length of 1 million tokens.