Back to Timeline

Event Summary

On March 14, 2023, OpenAI released GPT-4 — a multimodal model that could accept both text and image inputs while generating text outputs. GPT-4 scored in the 90th percentile on the bar exam, the 99th percentile on the SAT, and demonstrated dramatically improved reasoning, factuality, and steerability over GPT-3.5. It was integrated into ChatGPT Plus as a premium tier, and its API became the default for developers building on LLMs. GPT-4 raised the bar on what LLMs could do and set a new baseline for AI capability that persisted until the model release wave of late 2024.

Context & Narrative

GPT-4 was the culmination of the scaling approach OpenAI had pioneered with GPT-3. While OpenAI did not disclose its parameter count (citing competitive concerns), estimates placed it at 1.7 trillion parameters across a Mixture-of-Experts architecture with 8 expert models of 220B parameters each. The key advance was multimodal input — GPT-4 could process images, diagrams, screenshots, and handwritten text, not just typed text. In a paper accompanying the release, OpenAI evaluated GPT-4 on a battery of professional exams: it scored in the top 10% of test-takers on the Uniform Bar Exam (vs. bottom 10% for GPT-3.5), the 88th percentile on the LSAT, and the 99th percentile on the SAT Math. The model showed dramatically reduced hallucinations compared to GPT-3.5 and was more steerable (it could follow complex instructions with multiple constraints). GPT-4 also introduced the 'system prompt' concept, allowing developers to set the model's behavior, tone, and boundaries. GPT-4's release triggered a new phase in the AI industry: competitors raced to match its capabilities (Claude 3, Gemini 1.5, Llama 3), while regulators intensified scrutiny. The European Union's AI Act negotiations accelerated after its release. Its vision capabilities — reading diagrams, understanding memes, analyzing medical scans — opened use cases in education, accessibility, healthcare, and scientific research. GPT-4 was subsequently upgraded (GPT-4 Turbo, GPT-4o), but the March 2023 release marked the moment when LLMs became capable enough to be genuinely useful across a vast range of professional and creative tasks.

Key Findings

  • Fact Grade A

    OpenAI released GPT-4 on March 14, 2023 — a multimodal model accepting text and image inputs, scoring in the top 10% of test-takers on the Uniform Bar Exam.

    Sources [1]
  • Impact Grade B

    GPT-4 set a new capability baseline for LLMs that remained competitive through late 2024, achieved through a Mixture-of-Experts architecture estimated at 1.7T parameters.

    Sources [1]

Impact Assessment

  • Capability Leap +2 · Long-term

    Achieved professional-level performance on standardised exams (90th percentile bar exam, 99th percentile SAT Math). Introduced multimodal input (vision + text) at scale. Demonstrated significant reductions in hallucination and improved instruction-following compared to GPT-3.5.

    Affected Groups: AI researchers, developers, professionals, students

  • Economic Disruption +2 · Medium-term

    Accelerated AI integration across professional services (law, accounting, consulting, software). ChatGPT Plus ($20/month with GPT-4) became the fastest-growing consumer subscription product. Triggered a wave of enterprise AI adoption across Fortune 500 companies.

    Affected Groups: tech industry, professional services, enterprises, investors

  • Risk Creation -2 · Medium-term

    Raised the stakes on AI safety, alignment, and regulation. GPT-4's improved capabilities (including vision) expanded the surface area for misuse — generating convincing fake documents, detailed instructions for harmful activities, and automated disinformation at scale. Accelerated regulatory efforts including the EU AI Act.

    Affected Groups: policymakers, ethicists, general public, AI safety researchers

Consensus & Sources

Significance L2
Category Capability Breakthrough
Consensus Broad Consensus
Impact Index 7/10