- The Rundown AI
- Posts
- Google's AI ethics crisis
Welcome, AI enthusiasts.
Despite shipping at a rapid pace, Google is having a frustrating month.
Last week’s Gemini 1.5 breakthrough? Overshadowed by OpenAI’s Sora.
Now, the company faces major backlash after its AI image generator went viral — for all the wrong reasons. Let’s investigate…
In today’s AI rundown:
Google pauses AI image generation
Stability AI debuts Stable Diffusion 3
Animate flat vector illustrations
AI tackles nuclear fusion instability
6 new AI tools & 4 new AI jobs
More AI & tech news
Read time: 4 minutes
LATEST DEVELOPMENTS
OPENAI
Image source: The Financial Times
The Rundown: Google announced that the company is pausing Gemini’s ability to generate images of people, following heavy criticism after the model continuously overcorrected for diversity in its outputs.
The details:
Gemini’s attempts to ‘diversify’ its image outputs of people went viral across X, with users posting inaccurate results and finding difficulty generating Caucasian characters.
Google acknowledged the issues with a statement, highlighting the importance of diversity but saying it ‘missed the mark’.
The incident sparked broader discussion over bias and safety in AI models.
Google paused Gemini’s image generation abilities of people for now, saying it will “re-release an improved version soon.”
Why it matters: Ethical balance is going to be a very difficult task for AI companies, and this situation highlights why many believe open, less restrictive models are imperative. As AI creates more of the world’s content, its ability to shape world views grows — putting even more importance on models remaining as impartial and neutral as possible.
TOGETHER WITH TÜV SÜD
The Rundown: TÜV SÜD’s experts provide deep insights into AI regulations, helping you navigate the complex landscape of compliance and governance.
Here’s what TÜV SÜD can help you with:
Compliance Clarity: Demystify the EU's legal landscape for AI.
Ethical Alignment: Ensure your AI meets ethical and transparency standards.
Strategic Advantage: Turn regulation challenges into competitive strengths.
Discover TÜV SÜD’s approach to making your AI ready today.
STABILITY AI
Image source: Stability AI
The Rundown: Stability AI just unveiled Stable Diffusion 3, a new iteration of the company’s image generator that promises significant improvements in quality, multi-subject generation, and enhanced spelling within images.
The details:
The model is backed by a new transformer architecture similar to OpenAI’s Sora, improving quality and performance.
The release will include several different model sizes, ranging from 800M to 8B parameters.
Improved typography enables more accurate text generation and full sentences within images.
The upgraded model is currently available in ‘early preview’ via a waitlist.
Why it matters: While there was a major gap in the past between open models and closed AI leaders, it’s becoming clear that open-source can compete with the best. This upgrade continues the AI image acceleration — with impressive text skills that even Midjourney and OpenAI have yet to crack completely.
AI TRAINING
The Rundown: With Remix, you can easily create animated flat vector illustrations using a straightforward prompt and the smooth loop feature.
Hat tip to Chase Lean for this creative idea.
Download the Remix app, sign up for free, and click the “+” option to start the image generation process.
Select the "text" option and start your initial prompt with "A simple flat vector illustration of a [insert description]."
If you need assistance, the "prompt enhancer" option is quite helpful, so don't hesitate to use it.
Finally, choose your preferred model (e.g. Juggernaut XL) and image size.
Once you've selected your ideal image, choose the loop option to generate an animated gif version!
PRESENTED BY BRILLIANT
The Rundown: Build the skills to thrive in the world of AI with Brilliant. Interactive, bite-sized lessons help you:
Build serious problem-solving skills
Expand your knowledge into math, logic, programming, data science, and more
Make sharpening your mind an everyday habit
Join 10M+ people around the world and start your 30-day free trial. Plus, get 20% off a premium annual membership when you’re ready to level up.
AI RESEARCH
Image source: Midjourney
The Rundown: Scientists at Princeton just made a breakthrough in the pursuit of near-limitless clean energy, using AI to solve a critical instability challenge within fusion reactors.
The details:
Researchers developed an AI system capable of predicting disruptive "tearing" events in fusion plasma up to 300 milliseconds in advance.
In tests, the AI demonstrated the ability to adjust reactor power and plasma shape to prevent disruptions.
Unlike previous approaches, the model learns from real-world experimental data rather than theoretical physics.
Why it matters: This AI solution could make fusion reactors far more reliable and stable, bringing us a step closer to the long-held dream of clean, abundant fusion energy.
NEW TOOLS & JOBS
🦙 Alpaca- AI-powered Photoshop plugin (link)
✍️ Recaster- Improve your e-commerce SEO by up to 80% with seamless content generation (link)
💭 Dreamwriter- Create premium content that captivates your audience effortlessly (link)
🎉 Greeting Cards GPT- Customizable themed greeting cards for any occasion (link)
🎯 Dart- Project management tool that automates tasks (link)
💼 Talently Mock Interview- AI-powered interview coach (link)
QUICK HITS
Reddit’s previously reported sale of its data for AI training had its mystery buyer revealed, with Google striking the deal for around $60M per year.
OpenAI released new updates to GPTs for both creators and users, including expanded profile options and new ratings and feedback forms to provide to builders.
Google launched its ‘Help Me Write’ tool in Chrome (powered by Gemini), allowing users to generate AI writing or revisions with a single click.
Disney announced its investment into five startups as part of its Accelerator program, with three of them — ElevenLabs, PrometheanAI, and AudioShake — operating in the AI sector.
Poe announced new image understanding capabilities, integrating GPT-4V and LLaVa for use across the company’s range of chatbots.
Jasper acquired AI image editing platform Clipdrop from Stability AI, expending its AI-powered marketing capabilities and establishing a presence in Europe.
DatologyAI introduced a platform to automatically curate AI training data, optimizing model performance, efficiency, and cost-effectiveness.
THAT’S A WRAP
SPONSOR US
Get your product in front of over 450k+ AI enthusiasts
Our newsletter is read by thousands of tech professionals, investors, engineers, managers, and business owners around the world. Get in touch today.
FEEDBACK
How would you rate today's newsletter?Vote below to help us improve the newsletter for you. |
If you have specific feedback or anything interesting you’d like to share, please let us know by replying to this email.
Join the conversation