Prompts are unsafe, and that means language models are not fit for purpose
Over on the newsletter, I wrote about how the insecurity and poor design of prompts mean that language models are not safe.
Prompts are wholly insecure. They are holding back every attempt at turning language models into safe and reliable tools. Language models are not fit for purpose and should not be integrated into public-facing products unless the industry finds a new way to control them.
The truth about the EU AI Act and foundation models, or why you should not rely on ChatGPT summaries for important texts
I recently also published a deep dive into what the EU AI Act means for language models.
Developers (not deployers) of foundation models need to register their models, with documentation, prior to making it available on the market or as a service.
AI links
- “The Senate’s hearing on AI regulation was dangerously friendly”. “A number of experts and industry figures say the hearing suggests we may be headed into an era of industry capture in AI.”
- “The Urgency of Moving from Bias to Power”
- “Why Timnit Gebru Isn’t Waiting for Big Tech to Fix AI’s Problems”
- “Data Statements: From Technical Concept to Community Practice - ACM Journal on Responsible Computing”
- “Drag Your GAN: Interactive Point-based Manipulation on the Generative Image Manifold”. This is interesting: 1. seems to work with smaller training data sets, 2. faster than diffusion models, 3. an actual user interface.
- “It continues… – Neil Clarke”. Every paying outlet—even many that only ‘pay’ in attention—is going to be hit hard by generative AI.
- “The Fanfic Sex Trope That Caught a Plundering AI Red-Handed”. If, like me, you get the login wall, apparently the reason why the AI writing tool Sudowrite gets the results it gets is because OpenAI trained theirm models on on Archive Of Our Own, which is something AO3 is trying to prevent.
- “Debt Collectors Want To Use AI Chatbots To Hustle People For Money”. So, this sort of stuff is exactly the reason why the “high-risk” category exists and is more heavily regulated in the proposed EU AI Act.
- “This Is Catfishing on an Industrial Scale - WIRED”. Generative AI is amazing for fraud. It’s quickly become the scammers go-to toolset.
The rest
- “Pluralistic: Venture predation”
- “The web’s most important decision - The History of the Web”
- “Adactio: Links—The Industrial Hammer Complex”
- “The Industrial Hammer Complex”
- “Patience is its own reward — Chocolate and Vodka”
The best way to support this blog or my newsletter is to buy one of my books, The Intelligence Illusion: a practical guide to the business risks of Generative AI or Out of the Software Crisis.