LLMs Are Not a Higher Level of Abstraction
lelanthran.com
May 3, 2026
3 min read
🔥🔥🔥🔥🔥
51/100
Summary
Claims that large language models (LLMs) represent a higher level of abstraction in programming are prevalent online. The discussion critiques the notion that LLMs are a natural progression from binary programming to more advanced forms of programming.
Key Takeaways
- LLMs do not represent a higher level of abstraction in programming compared to previous programming languages like C or Python.
- The output of LLMs is based on probabilities rather than deterministic results, meaning the same input can yield multiple potential outputs.
- When using LLMs, there is a risk of receiving unintended artifacts alongside the desired output, which can lead to security vulnerabilities.
- The function of LLMs can be expressed as f(x) -> P(y | z1 | z2 | ... zN), indicating that the output can include both the desired result and additional, potentially harmful content.
Community Sentiment
Positives
- LLMs can handle abstract concepts that traditional programming languages like C++ cannot, indicating their potential to address user goals in software development.
- The discussion highlights the importance of understanding the limitations of LLMs, which can lead to better AI safety and alignment practices.
Concerns
- LLMs lack determinism, which raises concerns about their reliability as an abstraction layer, potentially leading to unpredictable outcomes in software development.
- The inability of LLMs to consistently produce correct outputs undermines their effectiveness as a replacement for traditional programming processes.
