The arguments against open source AI are bad
Points and comments are a snapshot, not live.
The arguments against open source AI are bad and historically misguided.
Tom Bedor refutes claims that open source AI is dangerous or un-American. He argues that open source software has always undergirded proprietary software, and attempts to suppress open weights will fail, as with encryption in the 1990s. China is not the only developer: Nvidia, Thinking Machines Lab, and enterprise users all have incentives to release open models. The "AI race" framing is incoherent. Chinese models are not dumping (software lacks physical supply chains), and fears of propaganda or backdoors are mitigated by inspectability. Open source AI is inevitable.
What commenters are saying
Commenters largely support the article. Top comment flags Dean Ball's incendiary tweets about OpenAI policy. A key split emerges between those arguing safety concerns apply equally to open and closed models, and those who say open weights uniquely enable fine-tuning for scams. The dominant rebuttal: the box is already open; social taboos or regulation cannot stop it. Several commenters invoke the crypto wars as a cautionary precedent against export controls. A minority raises the hypothetical of automated "genius-level scams," but most dismiss this as already feasible without LLMs.