Deep Dive into Fed's GenAI adoption report

This paper, published by the Federal Reserve Bank of St. Louis, investigates the rapid adoption of generative AI in the United States. The authors conducted a nationwide survey to measure the prevalence of generative AI use at work and at home. The study found that generative AI has been adopted faster than technologies such as personal computers and the internet, and is used in a wide range of occupations and job tasks. The authors also analyze the potential impact of generative AI on labor productivity based on their survey results and existing research on the technology's impact on task productivity.

Main Themes:

• Rapid Adoption of Generative AI: Generative AI is being adopted faster than previous transformative technologies like personal computers and the internet.
• Widespread Usage: Generative AI is a general-purpose technology used across a diverse range of occupations and tasks, both at work and at home.
• Potential for Productivity Gains: While still early, there is evidence suggesting that generative AI can lead to substantial increases in worker productivity.

Key Findings:

• Adoption Rate: As of August 2024, 39.4% of the U.S. population aged 18-64 reported using generative AI.
• This is higher than the adoption rates of the internet (20% after two years) and personal computers (20% after three years) at similar points in their lifecycles.
• Workplace Usage: 28% of employed respondents reported using generative AI at work.
• Over 24% used it at least once in the week prior to the survey, and nearly 1 in 9 used it every workday.
• Occupations: While adoption varies across occupations, generative AI usage is evident in both high-skilled and lower-skilled jobs.
• The highest usage is reported in computer and math occupations, followed by management and business/finance occupations.
• Tasks: The most common tasks generative AI is used for are writing, administrative tasks, and interpreting/translating/summarizing text or data. However, usage is spread across a wide range of tasks.
• Potential Impact on Productivity:The authors estimate that 0.5% to 3.5% of all work hours in the U.S. are currently being assisted by generative AI.
• "If we assume that generative AI increases task productivity by 25 percent - the median estimate across five randomized studies - this would translate to increase in labor productivity of between 0.125 and 0.875 percentage points at current levels of usage." (Bick, Blandin, Deming 2024)

Join our community: getcoai.com
Follow us on Twitter or watch us on Youtube
Get our newsletter!

OpenAI has recently launched a number of new features to its API. The Realtime API enables developers to build speech-to-speech experiences within their applications. The Vision Fine-tuning API enables developers to fine-tune GPT-4o with images and text to improve its visual understanding capabilities. Model Distillation lets developers create cost-effective models by using the outputs of more powerful models like GPT-4o to train smaller models. Prompt Caching helps developers reduce costs and latency by automatically caching input tokens, thereby reducing the amount of computation needed for frequently repeated inputs.

OpenAI's new Realtime API:

• Low-latency, multimodal experiences: The Realtime API enables developers to build applications with fast speech-to-speech conversations, similar to ChatGPT’s Advanced Voice Mode.
• Natural conversational experiences with a single API call: Developers no longer need to use multiple models for speech recognition, text processing, and text-to-speech. The Realtime API handles the entire process with one call.
• Streaming audio inputs and outputs: This allows for more natural conversations compared to previous approaches that resulted in noticeable latency and loss of emotion and emphasis.
• Automatic interruption handling: The Realtime API, much like Advanced Voice Mode in ChatGPT, can manage interruptions smoothly.
• Persistent WebSocket connection to exchange messages with GPT-4o: This underlies the Realtime API's functionality.
• Function calling: Voice assistants built with the Realtime API can respond to user requests by triggering actions or accessing new information.
• Six preset voices: The Realtime API utilizes the same six preset voices already available in the API.

The sources also discuss new features and capabilities in the Chat Completions API:

• Audio input and output in the Chat Completions API: This will allow developers to build applications that use audio without needing the low-latency of the Realtime API.
• Input and receive text or audio: Developers can choose to have GPT-4o respond with text, audio, or both.

Join our community: getcoai.com
Follow us on Twitter or watch us on Youtube
Get our newsletter!