A certain kind of talk around LLMs that I find increasingly puzzling (twitter.com)

The author expresses confusion over the negative perceptions surrounding large language models (LLMs) like ChatGPT when it comes to coding assistance. Contrary to the reports of poor performance and hallucinations in code generation, the author shares a positive experience using recent models (ChatGPT 5.4 and 5.5) for various programming tasks across multiple languages. They emphasize the LLMs' ability to accurately generate and audit code, improving their workflow and reducing debugging time by identifying all necessary code changes rather than missing crucial parts. This discussion raises important implications for the AI/ML community, particularly regarding user education and expectations. The author suggests that many criticisms may stem from users’ lack of understanding or the use of outdated models rather than genuine deficiencies in LLMs. They advocate for clear communication with AI tools, positing that precision in instructions leads to better results. This perspective invites a reevaluation of the effectiveness of LLMs and underscores the potential for substantial improvements in software development processes, highlighting the need for a deeper understanding of how to leverage these tools effectively.