UL NO. 405: My AI Bill Deep-Dive, AI Poisoning, an IR Prep Checklist, and Discovery++

You can safely assume you've created God in your own image when it turns out he hates the same people you do.

Anne Lamott

ChatGPT majorly upleveled last week with basically a Voltron upgrade. The latest update allows users to analyze documents, browse the web, and generate images using text prompts all in one session. Previously, users had to toggle each feature on independently and could only use one at a time. The fact that you can now uplodad PDFs and interact with them on ChatGPT is going to take out hundreds of companies that were based on that feature alone.

We essentially have features that are companies. And guess what? When someone else does that feature better, or in a trusted platform, you often no longer have a company.

Artists are fighting back against AI with Nightshade, a new tool that 'poisons' AI models with corrupted training data. Developed by researchers at the University of Chicago, Nightshade alters pixels in images in a way that's invisible to the human eye but confuses AI models. This means that an AI model trained on these 'poisoned' images will learn incorrect information, for instance, seeing a dog as a cat. MORE | MORE | MORE

In a surprise to absolutely nobody, US Immigration and Customs Enforcement (ICE) has been using an AI tool to sift through social media posts of visa applicants. The tool, Giant Oak Search Technology (GOST), assigns a score from 1-100 to determine the person's risk level. Stop being surprised by this kind of thing; the only question is how safe and fair this tech is, not whether or not it'll be used. MORE