Skip to content

GPT-4 System Card

Authors: OpenAI
Year: 2023
Venue: OpenAI technical report

type: other collection: misfiled status: reviewed

Links: link

Summary

Summary

OpenAI's system card for GPT-4 describes the safety challenges posed by the model's limitations and capabilities, the qualitative and quantitative evaluations run with external experts, and the model- and system-level mitigations (RLHF-based fine-tuning for helpfulness and harmlessness, monitoring, usage policies) adopted before deployment. It compares an early instruction-tuned version with the launch version and concludes that the mitigations reduce but do not eliminate misuse, calling for anticipatory planning and governance.

Abstract

Large language models (LLMs) are being deployed in many domains of our lives ranging from browsing, to voice assistants, to coding assistance tools, and have potential for vast societal impacts.[1, 2, 3, 4, 5, 6, 7] This system card analyzes GPT-4, the latest LLM in the GPT family of models.[8, 9, 10] First, we highlight safety challenges presented by the model’s limitations (e.g., producing convincing text that is subtly false) and capabilities (e.g., increased adeptness at providing illicit advice, performance in dual-use capabilities, and risky emergent behaviors). Second, we give a high-level overview of the safety processes OpenAI adopted to prepare GPT-4 for deployment. This spans our work across measurements, model-level changes, product- and system-level interventions (such as monitoring and policies), and external expert engagement. Finally, we demonstrate that while our mitigations and processes alter GPT-4’s behavior and prevent certain kinds of misuses, they are limited and remain brittle in some cases. This points to the need for anticipatory planning and governance.[11] Content Warning: This document contains content that some may find disturbing or offensive, including content that is sexual, hateful, or violent in nature. 41
Source: pdf

Extracted details

Field Value Source Confidence
contribution analysis of safety risks of GPT-4 (hallucination, harmful content, disinformation, privacy, cybersecurity, dual-use capabilities, emergent behaviours) llm 0.85
contribution description of OpenAI's evaluation approach with external red-teaming and quantitative evaluations llm 0.85
contribution overview of model-level and system-level mitigations before deployment llm 0.85
contribution comparison of GPT-4-early and GPT-4-launch behaviour llm 0.85
models GPT-4 llm 0.85
models GPT-4-early llm 0.85
models GPT-4-launch llm 0.85
tasks language modeling llm 0.85
tasks safety evaluation llm 0.85
tasks red teaming llm 0.85
interaction other llm 0.85
limitations mitigations are limited and remain brittle in some cases llm 0.85
limitations custom fine-tuning and image capabilities are out of scope llm 0.85
limitations architecture, training data and compute are not disclosed llm 0.85

Keyword mentions

Counts of vocabulary terms in the full text (first page of occurrence in brackets).

  • model: GPT 253 [p.1], Transformer 1 [p.31]
  • dataset: ARC 11 [p.3], Phishing 6 [p.13]
  • technique: Fine-tuning 2 [p.2]
  • task: Medical / healthcare 13 [p.5]
  • topic: Large language models 17 [p.1], Open source 2 [p.12]

Files

Path (under papers/) Collection Pages Size SHA-256 Note
references/misfiled/openai-2023_gpt-4-system-card.pdf misfiled 60 990.8 KB ca3677e1b83e

Database key: gpt-4-system-card · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set gpt-4-system-card COLUMN VALUE