Skip to content
TheChatArchiveThe Chat Archive
Mistralmixtral-8x7b-instruct-v0.1· Mistral AIHugging Face Arena Human Preference 55K

Explain the following text to a middle schooler. Large Language Models (LLMs) have…

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Claudeclaude-2.1’s answer.
  1. User

    Explain the following text to a middle schooler. Large Language Models (LLMs) have successfully been applied to code generation tasks, raising the question of how well these models understand programming. Typical programming languages have invariances and equivariances in their semantics that human programmers intuitively understand and exploit, such as the (near) invariance to the renaming of identifiers. We show that LLMs not only fail to properly generate correct Python code when default function names are swapped, but some of them even become more confident in their incorrect predictions as the model size increases, an instance of the recently discovered phenomenon of Inverse Scaling, which runs contrary to the commonly observed trend of increasing prediction quality with increasing model size. Our findings indicate that, despite their astonishing typical-case performance, LLMs still lack a deep, abstract understanding of the content they manipulate, making them unsuitable for tasks that statistically deviate from their training data, and that mere scaling is not enough to achieve such capability.

  2. Mistral Assistant

    Sure! So, Large Language Models (LLMs) are computer programs that can create other programs, including code in programming languages like Python. This is pretty cool, but it makes us wonder - do these LLMs really understand programming, or are they just good at mimicking it?

    One way to test this is to see if LLMs can understand "invariances and equivariances" in programming. Basically, this means that if you change the name of something in a program, but the program still does the same thing, a human programmer would still understand what the program does. But does an LLM also understand this?

    To test this, the authors of the text looked at what happened when they swapped the default names of functions in Python code. They found that LLMs often got the code wrong, and sometimes got more confident in their wrong answers as they got bigger! This is an example of a recent discovery called "Inverse Scaling," where bigger models don't always do better.

    The authors say that these results show that LLMs can do amazing things with typical programming tasks, but they still don't really understand the code they're working with. This means that LLMs might not be good at tasks that are different from what they were trained on, and just making them bigger won't help them understand better.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.