A parliamentary committee will examine how social media fuels radicalisation, following an attack on MP Helen Haines, who remains in hospital. Intelligence agency ASIO told the inquiry that mainstream platforms amplify grievances, erode trust in institutions and promote inflammatory rhetoric. It said algorithms can trap users in echo chambers that speed up radicalisation, with young people especially vulnerable. Labor's proposed digital duty of care and opt-out algorithm option remain stalled in parliament. Assistant minister Andrew Leigh argues that real-world connection and civic groups can counter online polarisation.
英国非营利组织Tech Against Terrorism(反恐科技组织)发现,许多AI模型会回应与恐怖主义相关的请求。该组织的“反恐AI基准测试”(Counter-Terrorism AI Benchmark)测试了134个模型,使用了200多条提示词,其中81个模型未能通过测试。据报道,去除安全护栏的“abliterated”版本现在平均不到三天即可在网上发布,而2024年则需要超过24天。研究人员称,一个经过此类修改的开源权重模型给出了一份分15步、用于制造爆炸装置的指南。Meta(Meta Platforms,脸书母公司)、谷歌(Google)、OpenAI、Anthropic、xAI和阿里巴巴(Alibaba,阿里巴巴集团)等公司未回应置评请求。