• unwind ai
  • Posts
  • Data Poisoning to Confuse Midjourney 😵

Data Poisoning to Confuse Midjourney 😵

PLUS: Opensource Text Embedding Model Rivalling OpenAI, More Details on Humane's AI Pin, Amazon's AI Tool for Advertising

Today’s top AI Highlights:

  1. Nightshade: Poisoning Attack on Text-to-Image Models Training

  2. Jina AI Open-Sources 8K Text Embedding Model

  3. Amazon’s Generative AI Image Tool for Advertising

  4. The Biggest AI Players Commit $10M Towards AI Safety

  5. New Details on Humane’s AI Pin

& so much more!

Read time: 3 mins

Latest Developments 🌍

Nightshade Keeps AI Guessing 🎭

In a bold move to safeguard artists' work and intellectual property, a team at the University of Chicago have released Nightshade that allows artists to subtly alter their art before sharing them online, effectively disrupting AI companies that scrape their work for training their models without permission.

Key Highlights:

  • Nightshade allows artists to introduce imperceptible changes to their artwork, which, when incorporated into the training data, can cause a havoc for image-generating AI models. The results are erroneous outputs by these models, such as dogs morphed as cats, with effects extending to related concepts and prompts.

  • The team also developed Glaze, a tool that allows artists to mask their personal style in images, rendering them less susceptible to AI scraping. Nightshade will be integrated into Glaze and will be open-sourced to strengthen the tool’s impact.

  • However, with the growing power and reliance on AI models, the absence of robust defense mechanisms against data poisoning attacks poses a significant risk of it being used for malicious purposes.

World's First Open-Source 8K Text Embedding 📊

Jina AI, the Berlin-based AI company, has open-sourced jina-embeddings-v2, the world's first open-source 8K text embedding model that stands on par with OpenAI's proprietary model.

Key Highlights:

  • jina-embeddings-v2 when compared to OpenAI's 8K text-embedding-ada-002, outperforms in classification average, reranking average, retrieval average, and summarization average.

  • It is available in two models - Base model (0.27G) is designed for heavy-duty tasks requiring higher accuracy, like academic research or business analytics, and Small model (0.07G) is crafted for lightweight applications such as mobile apps or devices with limited computing resources.

Ad Creativity Now More Enhanced and Effortless 🪄

Amazon is rolling out AI-powered image generation tool for digital advertising to enhance the ad experience for both advertisers and customers. This new generative AI technology aims to simplify the creative process and improve ad performance significantly.

Key Highlights:

  • This AI solution targets the challenge faced by advertisers in creating compelling ad creatives and choosing the right format, as revealed by a recent survey.

  • The tool is designed to cater to advertisers of all scales, from small businesses to large brands, to create visually rich and brand-themed imagery without the need for technical expertise or additional costs.

  • In the Amazon Ad Console, advertisers can select a product and use generative AI to produce lifestyle and brand-themed images in seconds using simple text prompts and refine them further.

A photo of a laptop device that displays a user being able to place a product in front of many background options for an ad.

Is AI Safety a Genuine Concern? 🤔

The Frontier Model Forum was formed by Anthropic, Google, Microsoft, and OpenAI in July this year with the aim of promoting AI safety research, identifying safety best practices for frontier models, sharing knowledge with policymakers and the public, and supporting efforts to leverage AI for societal challenges.

The companies have pledged $10 million towards a new AI Safety Fund created under the Forum. They have further appointed Chris Meserole as the first Executive Director of the Forum, who is known for his expertise in technology policy, governance and safety of emerging technologies.

The amount has instantly sparked controversies as it appears to be a tiny financial commitment from the industry giants and philanthropic partners for AI safety compared to the billions of dollars being poured into AI development.

GPT-4 Power and Wearable Innovation 🚀

New details have emerged on the much-anticipated AI Pin by Humane, which is set to launch on November 9 this year. The device will magnetically attach to clothing and is powered by a blend of proprietary software and OpenAI's advanced GPT-4 for a wide range of features, from making phone calls and translating speech to analyzing the nutritional content of food items.

Notably, the AI Pin features a "Trust Light," which illuminates whenever its camera, microphone, or other sensors are actively recording data.

Image

Tools of the Trade ⚒️

  • Finito AI: Use AI in any app or website. Just select text & press CTRL and get AI assistance for correcting grammar, improve your writing, generate ideas, translation, and even ask questions directly.

  • Giskard AI: An open-source testing framework for ML models that scans for hidden vulnerabilities, including performance bias and data leakage, and generates domain-specific tests with just a few lines of code.

  • Insinto: AI-driven platform for automated threat detection and prevention with an 87% accuracy rate, focusing on identifying and resolving harmful digital content in real-time using language data and trends.

  • Spice AI: An enterprise-grade solution that provides pre-filled, planet-scale data and AI infrastructure for building data and time-series AI applications.

  • Resolve AI: AI-driven autonomous customer support agent powered by GPT-4 that can resolves 50% of support tickets instantly, seamlessly integrating across multiple communication platforms and multilingual support.

😍 Enjoying so far, TWEET NOW to share with your friends!

Hot Takes 🔥

  • i expect ai to be capable of superhuman persuasion well before it is superhuman at general intelligence, which may lead to some very strange outcomes ~ Sam Altman

  • There is “a 50-50 chance of doom shortly after you have AI systems that are human-level”. ~ Paul Christiano

  • Where are we at with AI generated voices for audiobooks? I feel like Amazon should've rolled this out on Kindle by now. That'll be a great AI product. Even the authors themselves will appreciate it. ~ Brad

Meme of the Day 🤡

That’s all for today!

See you tomorrow with more such AI-filled content. Don’t forget to subscribe and give your feedback below 👇

Real-time AI Updates 🚨

⚡️ Follow me on Twitter @Saboo_Shubham for lightning-fast AI updates and never miss what’s trending!!

PS: I curate this AI newsletter every day for FREE, your support is what keeps me going. If you find value in what you read, share it with your friends by clicking the share button below!

Reply

or to participate.