Alec Scollon dreads LLM output despite boosted productivity

#ai #machinelearning #llm #programming

TL;DR

Alec Scollon's usage rate of LLMs is probably average by current dev standards.
He spends hours each day interacting with LLMs, including Claude Code and Codex.
Scollon feels more productive with LLMs but has started to dread reading LLM output due to false assumptions and hallucinations.
He thinks continually learning how to use LLMs effectively is valuable, despite feeling LLM burnout.

Alec Scollon's job has changed significantly with the integration of LLMs, from designing and writing code to designing code, describing the design to an LLM, reviewing code, and writing code. He uses LLMs like Claude Code and Codex for tasks, and also relies on them for casual queries, often using ChatGPT or reading Gemini's overview. However, he has started to feel a sense of dread when reading LLM output due to the repetition of false assumptions and hallucinations.

What the data shows

Alec Scollon's experience with LLMs has been extensive, with hours of interaction each day. He has worked with various LLMs, including Claude Code, Codex, and Qwen, and has used them for tasks such as creating tooling and sifting through unsupervised agent output. The data shows that Scollon's usage rate of LLMs is probably average by current dev standards, and his methods are probably primitive. He has also started to feel more comfortable in areas where he doesn't have deep knowledge, thanks to the LLM steps that expose him to new approaches.

What this means for ai readers

For AI readers, Scollon's experience highlights the potential for LLM burnout. The repetition of false assumptions and hallucinations in LLM output can be wearing, even for those who find LLMs valuable. Scollon's use of personalization features to mitigate this issue is notable, but the fact that some idiosyncrasies still seep through suggests that more work is needed to improve the user experience. AI readers may need to develop strategies to deal with the limitations of LLMs and find ways to use them effectively without feeling overwhelmed.

What to do right now

To address LLM burnout, users like Scollon may need to take a step back and reassess their usage of LLMs. This could involve taking breaks from LLM output, seeking out diverse sources of information, and developing strategies to mitigate the repetition of false assumptions and hallucinations. Additionally, LLM developers may need to prioritize improving the quality and diversity of LLM output to reduce the risk of burnout. Scollon's experience suggests that continually learning how to use LLMs effectively is valuable, but it is also important to be aware of the potential for burnout and take steps to prevent it.

Bottom line

Alec Scollon's experience with LLM burnout highlights the need for users and developers to be aware of the potential risks and limitations of LLMs. While LLMs can be valuable tools, the repetition of false assumptions and hallucinations can be wearing. By prioritizing improvement and developing strategies to mitigate these issues, users and developers can work together to create a better experience for everyone.

Frequently asked questions

Q: What is LLM burnout?

LLM burnout refers to the feeling of exhaustion or frustration that can come from repeatedly reading LLM output that contains false assumptions and hallucinations.

Q: How does Alec Scollon use LLMs?

Alec Scollon uses LLMs like Claude Code and Codex for tasks, and also relies on them for casual queries, often using ChatGPT or reading Gemini's overview.

Q: What are some potential solutions to LLM burnout?

Potential solutions to LLM burnout include taking breaks from LLM output, seeking out diverse sources of information, and developing strategies to mitigate the repetition of false assumptions and hallucinations.

Q: Why is it important to address LLM burnout?

Addressing LLM burnout is important because it can help users and developers work together to create a better experience for everyone, and prioritize improvement and development of LLMs to reduce the risk of burnout.