Source URL: https://www.theregister.com/2024/09/10/brute_force_ai_era_gartner/
Source: The Register
Title: We’re in the brute force phase of AI – once it ends, demand for GPUs will too
Feedly Summary: Gartner thinks generative AI is right for only five percent of workloads
AI techniques that require specialist hardware are “doomed," according to analyst firm Gartner’s chief of research for AI Erick Brethenoux – who included GPUs in his definition of endangered kit.…
AI Summary and Description: Yes
Summary: The insights from Gartner’s analysts emphasize a critical reassessment of AI hardware needs and the limitations of generative AI. They suggest a shift towards composite AI, combining generative methods with traditional AI techniques to enhance reliability and utility in business applications.
Detailed Description: Gartner’s discussions highlight the evolving landscape of AI, particularly in its reliance on hardware and the practical applications of generative AI. Key points include:
– **Specialist Hardware Considerations:**
– Analyst Erick Brethenoux argues that AI techniques necessitating specialized hardware are at risk of obsolescence. He notes that over time, more generalized hardware solutions (often termed “vanilla machines”) have continuously evolved to handle AI workloads effectively.
– This phenomenon suggests that the current reliance on GPUs and other dedicated hardware is emblematic of an immature “brute force” phase of AI, where software techniques have yet to be fully optimized.
– **Generative AI Limitations:**
– Brethenoux and fellow analyst Bern Elliot discuss the overhyped nature of generative AI, likening its prevalence in discussions to being “90 percent of the airwaves but only five percent of practical use cases.” This indicates a discrepancy between enthusiasm for generative AI and its real utility in current business scenarios.
– Both analysts recommend cautious usage of generative AI, highlighting that it is not suited for all tasks and stressing the importance of traditional AI techniques that have proven efficacy.
– **Composite AI as a Solution:**
– The discussion posits that organizations can still leverage existing AI capabilities effectively without solely relying on generative methods. The combination of generative AI with established techniques—termed “composite AI”—can lead to improved outcomes.
– For example, using generative AI to create user-friendly descriptions of predictive maintenance applications can enhance interpretability without requiring excessive reliance on potentially flawed generative outputs.
– **Guardrails and Reliability:**
– Analysts call for the implementation of safeguard mechanisms to validate the outputs produced by generative AI, given its inherent unreliability. They liken its functioning to Swiss cheese—indicating unknown pitfalls associated with its deployment.
– Despite recent improvements in mitigating hallucinations—misleading or inaccurate outputs—it remains crucial for users to apply checks to ensure the robustness of generative solutions in practical applications.
In conclusion, the insights from Gartner’s discussions both caution against an unchecked rush towards generative AI and highlight the need to combine this new technology with conventional methods to achieve reliable, effective outcomes in AI deployments. This perspective is particularly vital for security and compliance professionals as they navigate the complexities and challenges associated with AI integration into existing systems.