This article in The Register about "Poison Fountain" looks to be crithype, and the Poison Fountain project looks to be misdirection, scam, art project, or some other thing, but almost surely not a serious data poisoning proposal.
AI industry insiders launch site to poison the data that feeds them: https://www.theregister.com/2026/01/11/industry_insiders_seek_to_poison/
Poison Fountain starts with "We agree with Geoffrey Hinton: machine intelligence is a threat to the human species". This is a tarball of wrong. (1)
The rest of the website is absurd, and the "Poison Fountain Usage" list doesn't make any sense. There are far more efficient and safer ways to poison data that don't require you to proxy content for an unknown third party. Some of these are implemented in software, as opposed to <ul> in HTML. That bullet list reads like an amateur riffing on what they read about AI web scrapers, not like industry insiders with detailed information about how training works.
Recommend viewing the top level https://rnsaffn.com , which I suspect The Register may not have done.
The Register:
None of this passes a smell test. Crithype (and poor fact checking, it seems) from The Register it is.
(1) Hinton stands to gain professionally and financially from people believing this. Hinton personally bears a large amount of responsibility for setting off this so-called species level danger. Hinton, like all of us, cannot possibly know whether "machine intelligence" is even possible, let alone dangerous to people; that's a fanciful notion that serves the agendas of the wealthy and powerful quite well. In other words, crithype. Etc.
#AI #GenAI #GenerativeAI #Anthropic #PoisonFountain #UncriticalReporting #crithype #TheRegister
AI industry insiders launch site to poison the data that feeds them: https://www.theregister.com/2026/01/11/industry_insiders_seek_to_poison/
Poison Fountain starts with "We agree with Geoffrey Hinton: machine intelligence is a threat to the human species". This is a tarball of wrong. (1)
The rest of the website is absurd, and the "Poison Fountain Usage" list doesn't make any sense. There are far more efficient and safer ways to poison data that don't require you to proxy content for an unknown third party. Some of these are implemented in software, as opposed to <ul> in HTML. That bullet list reads like an amateur riffing on what they read about AI web scrapers, not like industry insiders with detailed information about how training works.
Recommend viewing the top level https://rnsaffn.com , which I suspect The Register may not have done.
The Register:
Our source said that the goal of the project is to make people aware of AI's Achilles' Heel – the ease with which models can be poisoned – and to encourage people to construct information weapons of their own.
None of this passes a smell test. Crithype (and poor fact checking, it seems) from The Register it is.
(1) Hinton stands to gain professionally and financially from people believing this. Hinton personally bears a large amount of responsibility for setting off this so-called species level danger. Hinton, like all of us, cannot possibly know whether "machine intelligence" is even possible, let alone dangerous to people; that's a fanciful notion that serves the agendas of the wealthy and powerful quite well. In other words, crithype. Etc.
#AI #GenAI #GenerativeAI #Anthropic #PoisonFountain #UncriticalReporting #crithype #TheRegister