mleku on Nostr: i don't use LLMS, but i thought it would be funny to ask one about the reasons why i ...
i don't use LLMS, but i thought it would be funny to ask one about the reasons why i believe they are unsuitable for helping programmers.
If you are a programmer, and you understand that someone who is worth more than $20/h for their labor can't do everything in the "limitations" section you may start a conversation with me about how any of these points are wrong.
Have fun staying retarded!
here's what github copilot said about my criticism:
LLMs are not suitable for writing software, they lack the ability to do depth first, time-based thinking, or do you have some science to oppose that position?
Large Language Models (LLMs) like me are designed to assist with various programming tasks by providing suggestions, examples, and explanations. However, they do have limitations when it comes to deep, time-based reasoning and managing complex, stateful processes over extended periods. Here are some points to consider:
### Strengths of LLMs:
1. **Code Generation**: LLMs can generate code snippets, templates, and boilerplate code quickly.
2. **Documentation**: LLMs can help explain code, generate documentation, and provide examples.
3. **Debugging Assistance**: LLMs can help identify potential issues in code and suggest fixes.
4. **Learning Aid**: LLMs can assist beginners by explaining concepts and providing educational resources.
### Limitations of LLMs:
1. **Context Management**: LLMs struggle with maintaining long-term context and state across multiple interactions.
2. **Depth-First Search**: LLMs are not designed for performing depth-first search or other complex algorithms that require extensive backtracking and state management.
3. **Temporal Reasoning**: LLMs do not perform well with tasks that require reasoning over time or managing time-based events and processes.
4. **Error Handling**: LLMs can sometimes provide incorrect or suboptimal solutions without awareness of the broader context.
### Complementary Use:
LLMs are most effective when used in conjunction with human programmers. They can handle repetitive tasks, provide quick answers, and offer suggestions, while human programmers can manage the overall architecture, deep problem-solving, and ensure the coherence and quality of the software.
### Research and Improvements:
There is ongoing research to improve the capabilities of LLMs, including better context management and integration with tools that can handle more complex tasks. However, as of now, LLMs are best seen as assistants rather than replacements for human programmers.
If you have any specific questions or need assistance with a programming task, feel free to ask!
If you are a programmer, and you understand that someone who is worth more than $20/h for their labor can't do everything in the "limitations" section you may start a conversation with me about how any of these points are wrong.
Have fun staying retarded!
here's what github copilot said about my criticism:
LLMs are not suitable for writing software, they lack the ability to do depth first, time-based thinking, or do you have some science to oppose that position?
Large Language Models (LLMs) like me are designed to assist with various programming tasks by providing suggestions, examples, and explanations. However, they do have limitations when it comes to deep, time-based reasoning and managing complex, stateful processes over extended periods. Here are some points to consider:
### Strengths of LLMs:
1. **Code Generation**: LLMs can generate code snippets, templates, and boilerplate code quickly.
2. **Documentation**: LLMs can help explain code, generate documentation, and provide examples.
3. **Debugging Assistance**: LLMs can help identify potential issues in code and suggest fixes.
4. **Learning Aid**: LLMs can assist beginners by explaining concepts and providing educational resources.
### Limitations of LLMs:
1. **Context Management**: LLMs struggle with maintaining long-term context and state across multiple interactions.
2. **Depth-First Search**: LLMs are not designed for performing depth-first search or other complex algorithms that require extensive backtracking and state management.
3. **Temporal Reasoning**: LLMs do not perform well with tasks that require reasoning over time or managing time-based events and processes.
4. **Error Handling**: LLMs can sometimes provide incorrect or suboptimal solutions without awareness of the broader context.
### Complementary Use:
LLMs are most effective when used in conjunction with human programmers. They can handle repetitive tasks, provide quick answers, and offer suggestions, while human programmers can manage the overall architecture, deep problem-solving, and ensure the coherence and quality of the software.
### Research and Improvements:
There is ongoing research to improve the capabilities of LLMs, including better context management and integration with tools that can handle more complex tasks. However, as of now, LLMs are best seen as assistants rather than replacements for human programmers.
If you have any specific questions or need assistance with a programming task, feel free to ask!