# Eight Things to Know about Large Language Models

**Authors:** Samuel R. Bowman  
**arXiv:** [2304.00612](https://arxiv.org/abs/2304.00612)  
**Date:** 2023-04-02  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.00612v1/

## Key points

- Large language models (LLMs) have gained widespread public deployment
- LLMs become more capable with increasing investment even without targeted innovation
- Important LLM behaviors emerge unpredictably as a byproduct of increasing investment
- LLMs often learn and use representations of the outside world
- There are no reliable techniques for steering the behavior of LLMs
- Experts are not yet able to interpret the inner workings of LLMs
- Human performance on a task does not necessarily set an upper bound on LLM performance
- Brief interactions with LLMs can often be misleading

## Summary

In recent months, large language models (LLMs) have gained widespread public deployment, prompting increased attention and engagement from advocates, policymakers, and scholars across various fields. While this technology raises many urgent questions, it is important to consider some potentially surprising points that may be overlooked. This paper provides evidence for eight such points: Firstly, LLMs predictably become more capable with increasing investment even without targeted innovation. Secondly, many important LLM behaviors emerge unpredictably as a byproduct of increasing investment. Thirdly, LLMs often appear to learn and use representations of the outside world. Fourthly, there are no reliable techniques for steering the behavior of LLMs. Fifthly, experts are not yet able to interpret the inner workings of LLMs. Sixthly, human performance on a task does not necessarily set an upper bound on LLM performance. Seventhly, LLMs need not express the values of their creators nor the values encoded in web text. Lastly, brief interactions with LLMs can often be misleading. It is crucial to note that these considerations may have significant implications for how we approach the development and deployment of LLMs in various contexts. Understanding these implications is essential in order to ensure that appropriate measures are taken to mitigate any negative consequences while maximizing its benefits when engaging with or deploying large language models. Overall, this paper highlights some key aspects that should be taken into account when engaging with or deploying large language models as this technology continues to evolve rapidly and gain wider adoption across different industries and sectors.