# GPT-4 Technical Report

**Authors:**  OpenAI  
**arXiv:** [2303.08774](https://arxiv.org/abs/2303.08774)  
**Date:** 2023-03-15  
**Summary page:** https://www.summarizepaper.com/en/paper/2303.08774v2/

## Key points

- OpenAI has released a technical report on the development of GPT-4, a large-scale, multimodal model that can accept both image and text inputs and produce text outputs.
- GPT-4 is based on the Transformer architecture and is pre-trained to predict the next token in a document.
- The post-training alignment process results in improved performance on measures of factuality and adherence to desired behavior.
- GPT-4 was able to pass a simulated bar exam with a score around the top 10% of test takers.
- GPT-4's ability to handle multimodal inputs sets it apart from its predecessors, allowing it to process both images and text simultaneously.
- This opens up new possibilities for natural language processing applications that require visual context.
- OpenAI was able to accurately predict some aspects of GPT-4's performance based on models trained with no more than 1/1,000th the compute of GPT-4.
- The development of GPT-4 represents another significant step forward in natural language processing research, particularly in terms of its ability to handle multimodal inputs.
- While the model is less capable than humans in many real-world scenarios, it exhibits human-level performance on various professional and academic benchmarks.

## Summary

OpenAI has released a technical report on the development of GPT-4, a large-scale, multimodal model that can accept both image and text inputs and produce text outputs. GPT-4 is based on the Transformer architecture and is pre-trained to predict the next token in a document. The post-training alignment process results in improved performance on measures of factuality and adherence to desired behavior. Notably, GPT-4 was able to pass a simulated bar exam with a score around the top 10% of test takers. What sets GPT-4 apart from its predecessors is its ability to handle multimodal inputs, allowing it to process both images and text simultaneously. This opens up new possibilities for natural language processing applications that require visual context. The report also highlights how OpenAI was able to accurately predict some aspects of GPT-4's performance based on models trained with no more than 1/1,000th the compute of GPT-4. This suggests that their infrastructure and optimization methods are highly effective at scaling up models while maintaining predictable behavior. Overall, the development of GPT-4 represents another significant step forward in natural language processing research, particularly in terms of its ability to handle multimodal inputs. While the model is less capable than humans in many real-world scenarios, it exhibits human-level performance on various professional and academic benchmarks.