And let’s think some Friday Big Ideas.
From IEEE Spectrum, Prompt engineering, the practice of optimizing prompts for language models, may be best done by the models themselves rather than human engineers, according to new research. Automated prompt-generation tools have been developed to outperform human-engineered prompts in performance and speed. Rick Battle and Teja Gollapudi from VMware conducted a systematic test on how different prompt engineering strategies impact language models’ ability to solve math questions. They found a surprising lack of consistency in the performance, concluding that the best strategy is likely specific to the particular model, dataset, and prompting strategy combination. They suggest that humans should not manually optimize prompts. Similar findings have been observed in image generation algorithms. While prompt engineering jobs may still exist in some form, the nature of the job is likely to evolve as AI models continue to advance.
A Bloomberg Law column by Roy Strom offers that law firms are not lagging in adopting generative AI tools compared to other industries…. Yet. While only a tiny percentage of companies currently use AI to produce goods or services, professional services firms, including law firms, are more likely to have implemented AI. Examples of law firms using AI to assist clients include Allen & Overy, DLA Piper, and Dechert. However, ethical considerations, privacy concerns, and the cost of AI tools may limit broader adoption. Law firms can stay ahead by investing in employee training programs and offering billable hour credits for learning about AI.
And in the Washington Post, Josh Tyrangiel offers the potential for artificial intelligence (AI) to reorganize the federal government and improve its efficiency. He highlights the success of Palantir, an AI company, in the corporate world and its struggle to secure government contracts. The article emphasizes the need to fully fix the bureaucratic system and adopt software practices to leverage AI in government operations fully. It suggests that AI can help streamline processes, enhance decision-making, and improve services. The author calls for political cooperation and a focus on specific areas, such as the Veterans Affairs Department, to demonstrate the transformative power of AI in government.
Finally, John Edwards in Information Week focuses on Cyber resilience, which combines IT cybersecurity with business continuity and is considered more important than cybersecurity alone. It focuses on preparing for and recovering from cyber incidents, ensuring business continuity. Cyber resilience complements cybersecurity by emphasizing response and recovery, providing a comprehensive approach to enterprise protection. Organizations must prioritize risk assessment, implement strong cybersecurity controls, develop incident response and business continuity plans, and engage in regular employee training to build a successful cyber resilience strategy.
Why do we care?
I’m sure I’ll dive in more on the Wednesday live show on some of these, but I did want to spend a moment on prompt engineering. That does look like it may turn out to be a fad, and I appreciate a bit of healthy skepticism in a realm that seems to move fast. The good news is that it was exposed as a fad reasonably quickly.
