Welcome back, AI Enthusiasts!

Subscribe | Try Thareja.AI

In today’s AI summary rundown:

1. OpenAI Set to Launch New Search Engine, Domain Evidence Suggests: OpenAI is rumored to be launching a search engine soon, potentially on May 9th. Recent server logs revealed the domain search.chatgpt.com, and a screenshot shared by Ashutosh Shrivastava shows OpenAI's new SSL certificate logs for this domain. This information suggests OpenAI is working on a new search feature. Stay tuned as OpenAI could soon provide a new way to search for information using advanced AI technology.

2. Microsoft Bans U.S. Police Departments from Using Azure OpenAI for Facial Recognition: Microsoft is reaffirming its ban on U.S. police departments from using its Azure OpenAI Service for facial recognition. New language added to the service's terms prohibits police departments from using OpenAI's AI tools for this purpose, including current and future image-analyzing models. Microsoft aims to ensure the responsible use of its AI tools, restricting their application in sensitive law enforcement contexts.

3. Paul Trillo Unveils First Commissioned Music Video Made with OpenAI’s Sora AI Generator: Artist Paul Trillo created the first commissioned music video using OpenAI's Sora AI video generator for the band Washed Out. Trillo had the idea for this project nearly a decade ago but could only bring it to life now with AI technology. The video showcases the creative potential of AI in music production, marking an exciting milestone in AI-assisted art.

   Read time is 4 min..

Insight: New Study Exposes 'Overfitting' in Popular AI Benchmarks, Shaking Confidence in Current Leaderboards

Recent research conducted by Scale AI reveals that certain large language models (LLMs) are significantly overfitting popular AI benchmarks. This raises serious questions about the validity of current leaderboards and evaluation methods.

Details:

• Understanding Overfitting: Overfitting refers to models memorizing specific questions from benchmarks instead of developing genuine problem-solving skills. Essentially, the models become too familiar with test data, performing well only because they've seen similar problems before.

• The New Dataset: Scale AI introduced a new dataset named GSM1k, which is closely modeled after the popular GSM8k benchmark used to evaluate math skills in AI models.

• The Experiment: When tested on GSM1k, several leading AI models performed significantly worse than they did on GSM8k. This discrepancy indicates that these models might be memorizing answers from GSM8k rather than demonstrating true understanding.

Notable Results:

- Underperformers: 

Models like Mistral and Phi notably struggled on GSM1k compared to GSM8k, revealing their overreliance on memorization.

- Consistent Performers: 

Models such as GPT-4, Claude, Gemini, and Llama maintained similar scores across both benchmarks, suggesting a more genuine grasp of problem-solving.

Why It Matters:

This research challenges the reliability of current AI benchmarks and leaderboards, revealing that overfitting may be inflating perceived model performance. As AI technology advances, these findings highlight the urgent need for evaluation methods that can't be gamed so easily. Datasets like GSM1k represent a step in the right direction, promoting more accurate assessments of AI capabilities.

AI Training: Virtually Try on Clothes with 'idm-vton' on Replicate

Step-by-Step Rundown:

1. Access the "idm-vton" Model: 

Head to the "idm-vton" model on Replicate. It's completely free to use, so dive in!

2. Pick Your Images: 

- Garment Image: Choose a clear photo of the clothing you want to 'try on.'

- Model Image: Find a clear photo of the person who's going to rock that outfit.

3. Select the Clothing Category: 

Depending on your clothing image, select the appropriate category:

- Upper Body (for tops)

- Lower Body (for pants, skirts, etc.)

- Dress (if it's a full dress)

4. Customize Your Settings: 

Play around with the various settings to get the best result:

- Seed: Changes the random seed for unique results

- Mask: Adjusts the masking of the garment

- Steps: Increases or decreases the detail level

5. Run and Review: 

Click the "Run" button and see your new look come to life! 🎉

Quirky AI Tools

1. 🚀 Waxaing AI: Take your project management to new heights with Waxaing AI, designed specifically for growth marketers. 📈 This tool streamlines campaign tracking, optimizes workflows, and provides real-time insights, giving you everything needed to unlock your team's full potential. 🌟 

2. 🤖 Amazon Q: Meet Amazon Q, the ultimate generative AI-powered assistant from AWS. 💡 It simplifies complex tasks, automates workflows, and helps you access valuable insights faster than ever. 🚀 Elevate your productivity with this intuitive AI companion! 💪 

3. 🥁 Soundry AI: Rock your music creation with Soundry AI! 🎵 This VST plugin delivers a treasure trove of AI-generated sound samples, perfect for music producers and DJs. 🥁 Mix, match, and unleash your creativity with endless possibilities. 🎶 

4. 💬 NVIDIA ChatRTX: Say hello to seamless conversations with NVIDIA ChatRTX! 🌍 Featuring multi-language voice support and an expanded lineup of LLMs, it's perfect for smooth global communication. 💪 Level up your chat experience and connect like never before! 🌟 

5. 👩‍🎤 Synthesia Expressive AI Avatars: Create ultra-realistic AI avatars with Synthesia's Expressive AI Avatars. 🎭 Add human-like expressions to your digital characters and bring storytelling to life like never before. 🌟 Unleash your creativity with this groundbreaking tool!

6. 💻 Relume: Design wireframes and craft UX copy in seconds with Relume's AI magic. ✨ This tool simplifies web design, giving you the power to create professional layouts effortlessly. 🚀 Get your creative juices flowing and build something brilliant in no time. 🌟 
   

Trending AI Startups

1. Hadean: 🌍 Bridging the gap between physical and virtual worlds, Hadean harnesses the transformative power of spatial computing and AI to tackle the globe's toughest challenges. Their AI-powered platform is the backbone of immersive worlds, offering digital twins, synthetic environments, and creator-friendly community experiences. 🛠️ Whether reshaping cities or redefining gaming, Hadean is rewriting the future of spatial computing. 🚀 

2. Medallia: 📈 Medallia DXA gives you 100% visibility into every digital customer experience, scoring them automatically to uncover the best opportunities for improvement. With DXA, strengthen customer relationships and supercharge your digital presence while boosting revenue. 💪✨ Say goodbye to guessing and hello to flawless customer satisfaction. 🏆

3. Skyways: 🚁 Frustrated by bumper-to-bumper traffic, Skyways was born to pioneer a new era in aerial transportation. With cutting-edge advances in unmanned aviation, electric propulsion, and energy storage, they are redefining travel. ✈️ Supported by YCombinator and a stellar engineering team, Skyways is soaring toward an electrified, traffic-free future. 🌟 

These startups are at the forefront of AI innovation, offering a range of services and technologies that are shaping the future of various industries.

AI-Generated Image of the Day

Prompt: Cityscape, an artistic illustration in the style of an Asian contemporary illustrator, abstract, against a dark blue background, forest, nature, magical animals, with white edging, dark black shadows, white and red accents, purple details, pink highlights, terror.

Tip: the more specific the better

Question of the Day

Today’s Meme 

That’s a Wrap 

That’s it for today’s news in the world of AI!

If you have anything interesting to share, please reach out to us by sending us a DM on Twitter: @dthareja or email me at [email protected]

How was today's newsletter?

Thanks for reading.

Until next time!

Dhiraj

p.s. if you want to sign up for this newsletter or share it with a friend or colleague, you can find us here

Try Thareja.AI: Time Tracking Software with Screenshots Capture

Sign up for Free

Automate: So you can focus on GROWTH! 

• Project Management

• Employee Tracking

• Desktop Screenshot Capture

• Productivity Coaching

• Proof of Work

• Timesheets

• Approvals

• Invoicing