• AI Tangle
  • Posts
  • ☕️ Vigilante Justice from Nightshade

☕️ Vigilante Justice from Nightshade

AI Tangle Newsletter

Good morning from the team at AI Tangle, and happy Tuesday! As the week gets going, so does the news, as Google slips and gives the world a peek of how its new Assistant with Bard app will look like. OpenAI and nonprofit CSM are hard at work, too, making up AI guidelines for the industry to follow. Meanwhile, Nightshade emerges from the shadows to throw a wrench into big AI's plans of using unethical model training data. Dive into the tech tales with us as we untangle the latest happenings this Tuesday morning.

THE BIG AI STORY

Hailing from the University of Chicago, project Nightshade aims to give artists a fighting chance against generative AI by tricking AI models into interpreting a completely different image than what a human would see. Ben Zhao, a computer science professor who had a hand in Nightshade's development, says that the project isn't by any means a be-all-end-all weapon for artists but rather a means to force AI industry leaders to start paying for licensed work, not just scraping it off the internet - "The real issue here is about consent, is about compensation."

How does it work?

Nightshade targets the associations that models make - for example, a human might describe a cow as a four-legged animal with a large body, two horns, and black-and-white coloring. But what if a model thought of a cow as actually being a jet plane with frying pans for wheels and sticks for wings? Nightshade breaks the understanding that models have formed of everyday concepts and things, and it only takes as little as 100 "poisoned" samples, which look completely unaltered compared to the image fed into it, to corrupt a prompt entirely.

​IN PARTNERSHIP WITH INCOGNI

Every day, big data brokers on the internet profit from your private sensitive information. Be it your phone number, DOB, SSN, or anything else, the result is the same - it gets sold to the highest bidder.

But you can deny them the chance to do it with Incogni. It scrubs your personal data from the web, confronting the world’s data brokers on your behalf. Incogni helps remove your sensitive information from all broker types, including People Search Sites.

​6 QUICK HITS

To gain the trust of policymakers and parents alike, OpenAI announced on Monday that it would be collaborating with nonprofit Common Sense Media (CSM) to minimize tech and media harm to kids and adolescents. The move most likely stems from OpenAI's quest to show regulators that their work is an overall net positive to society, not the opposite, as regulatory bodies in recent years have begun investigating OpenAI’s private information handling. As part of the partnership, the two will work together to create "family-friendly" GPT chatbots based on CSM's standards and ratings, says OpenAI CEO Sam Altman.

Made by The Browser Company, which launched Arc's first companion app for iPhones last year, Arc Search is the latest product to leave the company's doors with a unique take on browsing the web. Dubbed "Browse for me," Arc Search's powerful AI-based feature returns a neatly built webpage with information about whatever the user searched. Powered by different models, including OpenAI's, the "Browse for me" feature reads a minimum of six web pages and uses them as building blocks for a new page.

Leaks of Google Assistant with Bard have started circulating the internet as an X/Twitter user spotted something interesting - a new native Bard app to replace the Assistant app. In the demo video, which showed up in the Pixels Tips app that is available to most Google Pixel phones, a user uploads a picture of a plant and asks the app to help them take care of it, which prompts Bard to follow up with videos and detailed tips on plant care. Additionally, it still has similar functionality to how Google Assistant currently works, meaning the new app retains its previous abilities while now being infused with the power of Bard.

After the insurgence of vile, AI-generated deepfake porn of Taylor Swift last week on the platform, X/Twitter blocks any searches related to the American singer as a "temporary action" to prioritize safety, said head of business operations Joe Benarroch in a statement. The incident spread like wildfire, including to the White House, which on Friday called the spread of the AI-generated photos "alarming." Though there are currently no federal laws to combat the creation or sharing of deepfake images, many US politicians are starting to rise to the occasion.

Despite a disappointing earnings report for investors and analysts, Tesla recently announced that the company would be upping its capital expenditures to exceed $10bn in 2024, but warned shareholders that it expects to grow more slowly this year as it focuses on its next-generation vehicle which has been dubbed the Model 2. Among the new Model 2, Tesla will also be investing in AI, autonomy-like products and expanding its Supercharger network.

Rebellions, a South Korean-based fabless AI chip startup, recently announced that the company had bagged an impressive $124m in its Series B round of funding from big names, such as telecom giant KT, to develop its third AI chip - Rebel. With the leftover capital, as the original goal was $90m, Rebellions will invest it into new hires and its data center-focused chip Atom. The focal point of this story, however, is the partnership with Samsung, as the two South Korean companies will develop the Rebel chip together, backed by the previous relationship of the successful Atom chip.

4 AI TOOLS

​Durable - For young businesses needing a website, Durable is the AI-powered website-building tool for the job, allowing business owners to get sites up and running quickly with the help of a powerful editor.

Curiosity - Curiosity is a powerful AI-based search app bundled with an AI Assistant designed to neatly help pack all your files, emails, apps, and others into one place, eliminating the need to spend time on endless searching.

Streak - AI-powered data entry, precise insights, and tailored suggestions - Streak AI is your team's CRM co-pilot designed to help your team make informed decisions.

Jellypod - Calling itself the "better approach to news," Jellypod is an AI-powered tool that turns busy inboxes full of email subscriptions into comprehensive, personalized daily podcasts.

​AI READ & WATCH

The Reality of AI Spam (3-min read)

With the rise of AI, it isn't surprising to many that bad actors have begun flooding the web with AI-generated spam and misinformation, but how bad is it in reality?

A Fully-autonomous AI Restaurant in California (2-min watch)

A restaurant in Pasadena, California, has gone futuristic by adopting robot cooks and AI-powered kiosks. Join CBS news anchor Joy Benedict as she goes for a tour around CaliExpress by Flippy.