返回时间轴
2019-02-14

OpenAI says GPT-2 is 'too dangerous to release' — the first mainstream AI safety debate goes global

Safety & Ethics

事件摘要

On February 14, 2019, OpenAI announced GPT-2, a language model capable of generating remarkably coherent text. But instead of releasing it fully, OpenAI withheld the complete model, citing 'concerns about malicious applications.' The decision sparked an intense global debate: was OpenAI responsibly cautious or grandstanding for attention? The staged rollout — releasing progressively larger versions over months — became a template for AI safety disclosure. It also marked the first time ordinary people heard the argument that AI could be too dangerous to release, permanently changing how the public thought about AI risk.

影响评估

  • Paradigm Shift +1 · Long-term

    Established the staged disclosure model for advanced AI models. First time the 'too dangerous to release' debate reached mainstream audiences. Shifted public perception of AI from purely exciting to potentially risky.

    Affected Groups: general public, AI companies, policymakers, researchers

共识度与来源

重要度 L1
分类 Safety & Ethics
共识度 Broad Consensus
影响指数 5/10
  • 1
    OpenAI says GPT-2 is too dangerous to release — Slate

    URL: https://slate.com/technology/2019/02/openai-gpt2-text-generating-algorithm-ai-dangerous.html

    OpenAI says its text-generating algorithm GPT-2 is too dangerous to release.
    News Report Citation logged Live source
  • 2
    GPT-2 — Wikipedia

    URL: https://en.wikipedia.org/wiki/GPT-2

    GPT-2 was announced in February 2019.
    Reference Evidence Citation logged Live source

关联事件

← Origins
2014-09-01 L2

Attention mechanism proposed — the foundation of modern AI architectures

Bahdanau, Cho, and Bengio publish 'Neural Machine Translation by Jointly Learning to Align and Translate', introducing the attention mechanism that allows neural networks to dynamically focus on relevant parts of input when generating output. This innovation directly leads to the Transformer architecture.

→ Successors
2020-06-11 L3

OpenAI releases GPT-3, proving that scaling language models unlocks emergent capabilities

In June 2020, OpenAI published GPT-3 (Generative Pre-trained Transformer 3), a 175-billion-parameter language model that demonstrated remarkable few-shot learning—the ability to perform novel tasks from just a handful of examples without fine-tuning. GPT-3 could write essays, generate code, translate languages, and answer questions at a quality that often blurred the line between human and machine output. The paper 'Language Models are Few-Shot Learners' (Brown et al.) validated the scaling hypothesis: larger models trained on more data get qualitatively better, not just incrementally better.

2023-03-14 L2

OpenAI releases GPT-4, the first multimodal large language model

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.

Local Lineage OpenAI says GPT-2 is... 2019-02-14 · L1 Attention mechanis... 2014-09-01 · L2 OpenAI releases GP... 2020-06-11 · L3 OpenAI releases GP... 2023-03-14 · L2