Higher wage occupations are generally more exposed to LLMs than lower wage jobs. workforce could have at least 10% of their work tasks affected by the introduction of GPTs, while around 19% of workers may see at least 50% of their tasks impacted.” Further, the level of exposure is not evenly distributed across occupations. They define exposure as whether an LLM would “reduce the time required for a human to … complete a task by at least 50 percent.” As an alternative, they also had OpenAI’s GPT4 provide similar annotations of whether ChatGPT can reduce the time for task completion.īased on their analysis of over 1,000 occupations, the authors conclude that “approximately 80% of the U.S. For each task, they had human annotators assess the level of exposure to tools like ChatGPT. They coded whether those tasks will be impacted by LLMs. The authors look at over 1,000 occupations and the underlying tasks associated with these occupations (e.g., a task for a kindergarten teacher might be to “involve parents and older students in children’s activities”). We already know AI in general has properties of a general purpose technology. Recognizing general purpose technologies early and building the right strategy for them can separate winners from losers. And at the micro level, it's known to impact competitive dynamics. At the macro level, they are known to stimulate innovation and economic growth. General purpose technology is a term for a technology that has the potential for pervasive use by a large number of industries. GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models (by Sam Manning, Daniel Rock and coauthors) How will models like ChatGPT impact the labor market? Kartik’s Substack | Kartik Hosanagar | Substack
#Jumpcut startup software
If you are working on tools for software developers, I'd love to see a great solution to these two problems.Ĭheck out my Gen AI newsletter at
#Jumpcut startup code
Even if you protect your code and have guarantees that your code will not be used to train the AI, there is a risk that copyrighted code (or GPL code) can enter your repo. First, developer understanding of the code is generally lower when using tools like Copilot. So should firms embrace AI code suggestions? It has to be weighed against some serious drawbacks as well. This paper is a nice complement to Sida Peng's paper which shows that AI can improve developer productivity but this one shows that it can also improve code popularity and activity in a community like GitHub. No notable change in code quality as measured by the proportion of open issues that were solved (but not sure this measure comprehensively captures code quality). Significant increase in code popularity measured by number of watchers and forks (unclear why this happens but one hypothesis is that CoPilot code has better comments/documentation making it easier to read).ģ. Significant increase in code activity measured by number of create, delete, commit & pull requests.Ģ. They find that use of CoPilot is associated with:ġ. For Visual studio code, they have 1000+ treated and control repos (fewer repos for other 2 IDEs). Authors are able to identify which repos used CoPilot and which ones did not and construct treated and control repos using this. they analyze 2 years worth of data from Github code repositories related to 3 IDEs - Visual studio Code, Jetbrains, and Visual Studio. Authors study how does AI pair programming affect code popularity and quality. Gen AI can improve developer productivity but theere can be challenges including the difficulty in debugging or integrating the suggested code.
The Impact of Generative AI on Software Development ( by Fangchen Song and Ashish Agarwal) Another interesting paper on the impact of Gen AI on software development at our workshop on Business & Gen AI ( ).
0 kommentar(er)
