- The Brainyacts
- Posts
- 220 | 🚨🚨 HUGE SECURITY VULNERABILITY IN MICROSOFT COPILOT!
220 | 🚨🚨 HUGE SECURITY VULNERABILITY IN MICROSOFT COPILOT!
Brainyacts #220
It’s Friday. If you haven’t yet taken 5 seconds to take the poll below, would you? Recently we hit over 5200 subscribers! It is time to learn more about each of you. It helps me improve the newsletter.
What is your primary role? |
Here we go.
In today’s Brainyacts:
Who is reading Brainyacts?
A legal-specific frontier AI model
🚨🚨 HUGE SECURITY VULNERABILITY IN MICROSOFT COPILOT! and other AI model news
Google vs. Gartner on GenAI ROI and more news you can use
👋 to all subscribers!
To read previous editions, click here.
Lead Memo
🤓✋ Who is reading Brainyacts?
Yes, I am double-dipping on this today. It is really important. I know some of you have taken the poll at the top already and I want to share how it looks so far:
👋 🙏 For those that haven’t taken the poll - will you? It takes one click. It helps me so much in making sure Brainyacts remains helpful and useful to you.
What is your primary role? |
Spotlight
🆕 🧑🔬 New Legal-specific AI model in development
Equall is a new player driving cutting-edge AI technology in legal. Founded to push the boundaries of what AI can achieve in law, Equall is making waves by introducing specialized AI systems that tackle complex legal challenges with remarkable efficiency.
The highlight of Equall’s work is the Saul family of language models, designed specifically for legal reasoning. These models, ranging from 7 billion to a staggering 141 billion parameters, have already set new benchmarks in the industry. Notably, Saul-141b has outperformed much larger commercial models, including GPT-4, in legal reasoning tasks—a remarkable feat for a specialized model.
So, why is having a legal-specific AI model like Saul so important? The answer lies in the complexities and nuances of the legal field. Law isn’t just about language; it’s about context, precedent, jurisdictional differences, and a deep understanding of legal principles that can vary dramatically across cases. General AI models, while powerful, often fall short in these areas because they’re trained on broad datasets that lack the depth required for precise legal reasoning. As we know, many lawyers and judges have learned this the hard way.
In contrast, models like Saul are trained on extensive legal data, allowing them to better grasp the subtleties of legal language and reason far better than their general-purpose counterparts.
Here is a glimpse of the data they use to train it:
But developing these niche AI models is no easy task. The challenge lies in the need for highly specialized data, sophisticated adaptation processes, and the computational power to fine-tune models to meet the rigorous demands of legal tasks. Equall’s work with the Saul models showcases the difficulty and the reward: each step of their development, from data curation to domain-specific optimization, is a testament to the complexities involved in creating AI that can truly understand and perform legal work. The result? Saul models outperform even the most advanced general models, such as GPT-4, in legal reasoning benchmarks.
Feel free to dig into their research papers here. They explain the training methods more than the use case applications. For use cases/evaluation criteria check out the LegalBench project.
What makes Equall’s approach particularly exciting is its commitment to open-source development. The entire Saul family of models is now available on Equall’s HuggingFace page. By releasing the entire Saul family under a permissive MIT license, Equall is fostering innovation and collaboration in the legal and AI communities. This move is poised to accelerate the integration of AI into legal practice, offering lawyers new tools to enhance their work and deliver more accessible legal services.
For those of us who are passionate about the pragmatic use of AI in law, Equall’s work is a significant milestone. Their models promise to reshape how legal services are delivered, making the practice of law more efficient and, ultimately, more equitable. Keep an eye on Equall—they’re leading the charge into the future of legal AI.
AI Model Notables
► 🚨🚨 HUGE SECURITY VULNERABILITY IN MICROSOFT COPILOT! For an attacker to manipulate your Copilot into doing whatever they want, all they need to do is share a simple document with you. That’s it.
Researchers find relatively easy ways to manipulate Copilot showing how easy it is to manipulate its outputs and thereby manipulate YOU.
► This post by OpenAI’s CEO is driving people crazy - many think it is a hint of a new AI model coming soon . . . It has been rumored that an internal project at OpenAI has been using the code name of “strawberry.”
► OpenAI says its latest GPT-4o model is ‘medium’ risk.
► Elon Musk to pause X’s AI training on some EU data, Ireland says.
► Palantir partners with Microsoft to sell AI to the government.
► Apple could charge up to $20 for some Apple Intelligence AI features, analysts say.
► Too late for the Olympics, Google creates AI-driven table tennis robot.
► Twitter/X tests a new “Ask Grok” function directly in your stream:
News You Can Use:
➭ 86% of enterprises see 6% revenue growth with gen AI use and 84% of organizations successfully transform a gen AI use case idea into production within six months, according to Google Cloud survey.
➭ But, Gartner predicts 30% of Generative AI projects will be abandoned after proof of concept by end of 2025.
➭ ChatGPT’s mobile app just had its biggest month yet.
➭ Zoom challenges Microsoft and Google with AI-powered Docs.
➭ LG unleashes South Korea’s first open-source AI, challenging global tech giants.
➭ AND, LG, Samsung eye South Korea's AI textbooks as edtech springboard.
➭ Colorado schools have AI roadmap to guide students and teachers into brave new world.
Was this newsletter useful? Help me to improve!With your feedback, I can improve the letter. Click on a link to vote: |
Who is the author, Josh Kubicki?
Some of you know me. Others do not. Here is a short intro. I am a lawyer, entrepreneur, and teacher. I have transformed legal practices and built multi-million dollar businesses. Not a theorist, I am an applied researcher and former Chief Strategy Officer, recognized by Fast Company and Bloomberg Law for my unique work. Through this newsletter, I offer you pragmatic insights into leveraging AI to inform and improve your daily life in legal services.
DISCLAIMER: None of this is legal advice. This newsletter is strictly educational and is not legal advice or a solicitation to buy or sell any assets or to make any legal decisions. Please /be careful and do your own research.8