Source URL: https://www.wired.com/story/how-do-you-get-to-artificial-general-intelligence-think-lighter/
Source: Wired
Title: How Do You Get to Artificial General Intelligence? Think Lighter
Feedly Summary: Billions of dollars in hardware and exorbitant use costs are squashing AI innovation. LLMs need to get leaner and cheaper if progress is to be made.
AI Summary and Description: Yes
Summary: The text discusses the anticipated surge of AI-powered applications by 2025, highlighting the competitive landscape among companies like OpenAI, Google, and xAI. It emphasizes the high cost of training and inference for large language models (LLMs) and the expected decline in inference costs that could democratize access to AI technology for developers and consumers.
Detailed Description:
The provided text outlines significant trends and future expectations in the domain of artificial intelligence, with particular focus on generative AI and large language models (LLMs). Here are the critical points:
– **Competitive Landscape**:
– Major players like OpenAI, Google, and xAI are engaged in a “gladiatorial battle” to dominate the field of generative AI and eventually achieve artificial general intelligence (AGI).
– The text mentions significant investments, specifically Elon Musk’s $6 billion funding for xAI and the acquisition of Nvidia H100 GPUs for training LLMs.
– **Ecosystem Imbalance**:
– The current AI ecosystem is described as “bottom heavy and top light,” where few wealthy corporations benefit at the expense of smaller developers.
– High costs associated with inference make it challenging for developers to produce applications that can compete effectively, leading to a dependency on either low-performing models or risking bankruptcy through high operational costs.
– **Inference Cost Dynamics**:
– Inference costs are highlighted as a critical barrier, with current estimates showing a stark contrast in costs between generative AI search queries and non-gen AI queries (e.g., $10 per query for generative AI vs. $0.01 for traditional search).
– The expectation is set that a new law of AI inference could drastically lower these costs by a factor of 10 each year due to advancements in AI algorithms, inference technologies, and chip pricing.
– **Future Projections**:
– By 2025, it is suggested that a more favorable economic model will emerge, similar to past tech revolutions.
– As inference costs decline significantly, developers are projected to gain access to high-quality AI models, facilitating a rapid increase in innovative AI applications.
Key Implications for Security and Compliance Professionals:
– **Opportunities in Innovative Applications**: The anticipated proliferation of AI apps can provide opportunities for security and compliance professionals to address the related challenges, such as securing sensitive data processed by these applications.
– **Cost Efficiency and Security Investments**: As costs for implementing AI reduce, organizations might allocate resources more effectively towards robust security frameworks and compliance measures tailored for AI applications.
– **Governance and Regulatory Monitoring**: The rise of AI applications necessitates increased attention to governance, regulations, and compliance frameworks to manage the risks associated with deploying large-scale AI systems.
Overall, the text presents a forward-looking view on generative AI that is crucial for understanding shifts in the technology landscape and implications for various stakeholders, including those focused on security, privacy, and compliance.