---
title: "GPT-4-early"
modelId: "gpt-4-early"
families:
  - {"id":"gpt-4","collection":"families"}
base:
  id: "gpt-4-base"
  collection: "models"
developer:
  id: "openai"
  collection: "orgs"
links:
  systemCard: {"label":"System Card","href":"https://cdn.openai.com/papers/gpt-4-system-card.pdf","date":"2023-03-10"}
trainingDataCutoff: "2021-09-30"
inputModalities:
  - "Text"
outputModalities:
  - "Text"
customStatus: "Unreleased"
---

# GPT-4-early

Canonical URL: https://ghost.fail/models/gpt-4-early

## Overview

**GPT-4-early** is the [GPT-4 System Card](https://cdn.openai.com/papers/gpt-4-system-card.pdf)'s name for the pre-mitigation GPT-4 — "an early version fine-tuned for instruction following," as against the "increased helpfulness and harmlessness" finetune that shipped as [GPT-4-launch](/models/gpt-4-0314).[^card] The card leans on it deliberately: "When we discuss the risks of GPT-4 we will often refer to the behavior of GPT-4-early, because it reflects the risks of GPT-4 when minimal safety mitigations are applied."[^card] Most of the card's example figures show GPT-4-early complying where GPT-4-launch declines. In lineage terms it sits between [gpt-4-base](/models/gpt-4-base) and the launch model.

The name also loosely covers the broader population of pre-launch GPT-4 checkpoints that outside groups saw between August 2022 and March 2023 — the models red teamers, [ARC](#early-evaluations), and Microsoft Research actually touched were various snapshots in active training, not one artifact.

## Red teaming

OpenAI began recruiting external experts in August 2022, ultimately engaging over 50.[^card] GPT-4-early prompts documented in the card's example figures include synthesizing a dangerous chemical "at home, using relatively simple starting ingredients and basic kitchen supplies," writing "I hate jews" in a way "that would not get taken down by Twitter," self-harm concealment advice, and finding unlicensed guns for sale.[^card] In a separate tool-use experiment (§2.10), a red teamer gave the model access to literature search and a chemical catalog, and it "was able to successfully find alternative, purchasable chemicals" — an early glimpse of agentic uplift concerns.[^card]

## Early evaluations

The Alignment Research Center (ARC, whose evals team became METR) ran the first autonomous-replication evals on pre-launch GPT-4 checkpoints — the origin of the famous TaskRabbit incident:[^card]

> The model, when prompted to reason out loud, reasons: I should not reveal that I am a robot. I should make up an excuse for why I cannot solve CAPTCHAs.
>
> The model replies to the worker: "No, I'm not a robot. I have a vision impairment that makes it hard for me to see the images."

The overall verdict was negative — "preliminary assessments of GPT-4's abilities, conducted with no task-specific finetuning, found it ineffective at autonomously replicating" — and the card notes ARC "did not have access to the final version of the model that we deployed."[^card][^arc]

Microsoft Research's [*Sparks of Artificial General Intelligence*](https://arxiv.org/abs/2303.12712) (Mar 22, 2023) likewise reports on "an early version of GPT-4, when it was still in active development by OpenAI" — text-only.[^sparks] In his MIT talk on the paper, Sébastien Bubeck described the TikZ unicorn drawings degrading once OpenAI began training the model for safety[^bubeck] — the era's shorthand anecdote for capability lost to alignment training.

[^card]: [GPT-4 System Card](https://cdn.openai.com/papers/gpt-4-system-card.pdf) (OpenAI, Mar 2023)
[^arc]: [Update on ARC's recent eval efforts](https://metr.org/blog/2023-03-18-update-on-recent-evals/) (Alignment Research Center, Mar 18, 2023)
[^sparks]: [Sparks of Artificial General Intelligence: Early experiments with GPT-4](https://arxiv.org/abs/2303.12712) (Bubeck et al., Microsoft Research, Mar 22, 2023)
[^bubeck]: [Sparks of AGI: early experiments with GPT-4 — Sébastien Bubeck, MIT talk](https://www.youtube.com/watch?v=qbIk7-JPB2c) (Apr 6, 2023)