Вештачка интелигенција

The Current State of Generative AI Technology: An Experimental Phase with Limitations

Summary

Generative AI technology continues to make strides in various industries, but it remains largely in the experimental phase with some limitations. The debut of ChatGPT, which captured global attention, showcased its ability to generate everything from Shakespearean sonnets to student […]

The Current State of Generative AI Technology: An Experimental Phase with Limitations

Generative AI technology continues to make strides in various industries, but it remains largely in the experimental phase with some limitations. The debut of ChatGPT, which captured global attention, showcased its ability to generate everything from Shakespearean sonnets to student essays. However, its tendency to produce “hallucinations” of incorrect information has hindered its potential to revolutionize most industries.

Anthony Aguirre, the founder and CEO of the Future of Life Institute, a nonprofit organization focused on mitigating catastrophic risks associated with advanced artificial intelligence, explained the key difference between being capable of doing something and being capable of doing it well enough for a specific purpose. He highlighted the example of self-driving cars, which are functional to some extent but not reliable enough to replace human drivers.

Sherry Marcus, the Director of Applied Science at Amazon Web Services (AWS), noted that AWS customers are at different stages of progress with generative AI technology. Some have already integrated it into production, while others are just beginning their journey.

One prevalent use of generative AI technology, as highlighted by speakers from various industries, is in writing computer code. Microsoft’s GitHub, an online code repository platform, revealed that approximately half of the programming on their platform is written using an AI-powered tool called Copilot, which automatically suggests lines of code. Lili Cheng, Corporate Vice President at Microsoft, acknowledged that developers using Copilot feel more productive, emphasizing the potential for greater efficiency and accessibility in programming.

Cheng also cited AI-generated meeting transcript summaries as another useful application of the technology. Financial institutions have also confirmed active utilization of AI models for tasks such as coding, document generation, and capital allocation.

Despite the benefits, executives emphasized the importance of approaching the integration of generative AI technology cautiously, particularly in sectors where accuracy is crucial. They advised companies to proceed slowly and carefully to ensure the technology’s accurate implementation.

Gary Marcus, a professor at New York University, acknowledged the potential for errors in coding and other areas with generative AI but highlighted the tech industry’s ability to address and rectify these issues promptly. However, in nearly any other type of business, these hallucinations present a serious problem.

The executives stressed the need for a gradual and deliberate approach to the integration of generative AI technology, especially in critical use cases within legal and security domains. They emphasized the importance of developing technologies, guidelines, and frameworks to protect against unintended consequences.

Vijoy Pandey from Cisco expressed his belief in the usefulness of artificial intelligence for “easier tasks” with relatively minor consequences for errors. The challenge lies in moving the technology forward to address more sensitive and critical use cases, such as those in legal and security fields. Pandey suggested focusing on building technologies, guidelines, and frameworks to safeguard against mistakes made by humans.

Overall, while generative AI technology is still in its experimental phase with limitations, it holds significant potential across industries. As the technology evolves and addresses its current challenges, it will be crucial to strike the right balance between innovation and caution to ensure safe and effective integration.

FAQs