The open-model debate: benefits versus safety risks
What happened
On October 13, MIT Technology Review weighed the benefits and risks of open versus closed models. Open models support research, customization and broader cyber defense, but once downloaded, access can't be revoked, and safety training can be stripped out by retraining. Closed models face misuse too, and open weights don't mean the training data or the hardware to run them are available. Researchers are exploring tamper-resistant training, and some experts argue for staged releases; current methods still can't withstand every tampering attack.
Written by AI from the coverage · updated 2 hours ago
Coverage
Follow the reports to see the story from different sides.
- MIT Technology Review · AIOpen models have real benefits. Do they outweigh the risks?
开放模型支持研究、定制和更广泛的网络防御，但下载后难以撤销访问，安全训练也可能被重新训练移除。MIT Technology Review 对比其与封闭模型的收益和风险，指出封闭模型同样存在滥用问题，开放权重也不保证训练数据和运行硬件可及。研究者正在探索抗篡改训练，部分专家主张分阶段开放模型，但现有抗篡改方法尚不能抵御所有篡改攻击。
Heat over time
Not enough continuous observations to draw a trend yet.