Others point out that LLMs are still relatively new and their full potential may not yet be realized. They suggest that LLMs could lower the entry barriers for new programmers and even create a new solopreneur economy. However, some participants express skepticism, noting that the promised productivity gains from LLMs have not yet materialized in a noticeable way in the industry. They also highlight that LLMs are not very effective at dealing with complex or poorly defined problems, which are common in programming.
Key takeaways:
- Despite the hype around large language models (LLMs) like GPT-3 and GitHub Copilot, their impact on productivity in the tech industry is not yet clearly visible in terms of an increase in open source projects or a decrease in bug fixing time.
- Some users find LLMs useful for generating boilerplate code or assisting with specific tasks, but they are not seen as a game-changer for the entire coding process. The bottleneck in software development often lies in areas like requirement gathering, stakeholder management, and problem-solving, rather than the actual coding.
- There are also concerns about the accuracy of LLMs' suggestions, with some users reporting that they often suggest incorrect solutions. The usefulness of LLMs may also be limited by the user's ability to clearly define the problem and ask the right questions.
- Despite these limitations, some users believe that LLMs have the potential to significantly change the way coding is done in the future, with one predicting that in ten years, most keyboard inputs will be directed towards guiding an AI coding assistant rather than writing code directly.