Welcome back to Designer Input,

This Week’s Topics:

• Can We Poison AI?: Nightshade

• 200 Best Innovations of 2023: Authentic AI

• 3 AI Tools for Designer2.0’s

• AI-Meme Tool of the Week: Generate Research Papers

• Stories Worth Reading: How to generate 10 images per second?

Main Insight

🧪Can We Poison AI?

“Prompt-Specific Poisoning Attacks on Text-to-Image Generative Models”

Artists may have a new weapon to fight against Generative AI models.

A new technique called Nightshade is giving artists a way to protect their work against AI image generators that are using their artwork without permission.

Nightshade works by allowing artists to add invisible poisoned pixels to their images before uploading them online.

When AI models like DALL-E, Stable Diffusion, or Midjourney are trained on datasets containing these "poisoned" images, the models get confused about what the images actually represent.

Study shows corrupting just 100 images can break a model trained on hundreds of millions of examples. The impacts spread to related concepts too.

If "dog" images are poisoned, associated concepts like "puppy" and "husky" get confused because of the bleed-through effect, too.

If models are extensively contaminated by wrong data using Nightshade, AI companies may eventually need to start providing royalties and credit to artists.

What do you think could happen if subtle art poisoning becomes common among digital artists?

Interesting

💡200 Best Innovations of 2023

This week, Time Magazine has released its list of the 200 Best Inventions of 2023.

The list highlights some of the most impactful and exciting new products and ideas of the year.

The 200 inventions were selected from 21 different categories.

Here are some of my top picks showcasing the most exciting new inventions from a few categories.

1) AI _{(of course} 😃₎

• Authentic AI: Can tell you whether an item is fake or genuine with a photo from your phone.

• Adobe Photoshop Generative Fill

• OpenAI GPT-4

• Dall-E 3

• Runway Gen-2: Film Editing

• Eyes in the Skies: Dedrone City-Wide Drone Detection

• The Project Gutenberg Open Audiobook Collection: Thousands of free audiobooks using new neural text-to-speech technology

2) Design - Las Vegas Sphere

3) Household - Hunter Douglas Aura Illuminated Shades

The Aura Illuminated Shades provide shading, helping keep your home energy efficient, but they also create light of their own.

4) Experimental - 3D Modeling, NVIDIA Neuralangelo

The AI model, built by NVIDIA Research, turns 2D video into realistic 3D Models, capturing fine details of complex materials like roof shingles, panes of glass, and smooth marble.

5) Accessibility - Music Not Impossible

The special device translates audio from a concert venue’s mixers and placed microphones into vibrations that allow people who are deaf to feel the music on their skin.

SSD-1B: A filtered SDXL model that is 50% smaller and 60% faster without much quality loss. _(Free)

Tome: Tome is more engaging than a slide deck, and easier to build than a webpage. Use AI to quickly create multimedia presentations, microsites, and more. _(Freemium)

Image App: Generate with the latest models including Stable Diffusion SDXL 1.0, OpenAI DALL-E, Kandinsky 2.2, and more. _(Freemium)

AI-Meme Tool of the Week

Why Search For Scientific Sources When You Can Generate Them?

Invite your friends to Designer Input 👇🏻

Or copy the link here to share: https://newsletter.designinputstudio.com/subscribe?ref=PLACEHOLDER

Our Picks for You

Stories Worth Reading

• The new Latent Consistency Diffusion Model can generate 10 images per second with only 2-4 steps. Try it here.

• Lenovo introduced a new bendable phone model that can be worn like a wristwatch too.

• Twelvelabs.io launched their new Pegasus-1 video-to-language model. It can analyze videos, create summaries, and answer questions from the content.

• The employees secretly using AI at work.

• Midjourney has rolled out the beta version of its new website. While image generation is not yet available on the site, it does feature a "coming soon" banner indicating that capability will be added in the future.