• AI Academy
  • Posts
  • 🤖 Anthropic Redefines AI Safety Standards

🤖 Anthropic Redefines AI Safety Standards

Plus, OpenAI’s CriticGPT and Anthropic’s Projects Feature

July 03, 2024

Hello AI Enthusiast,

Ever wondered how AI models are measured today and if the standards are really up to the mark? Anthropic is exploring a more effective way to tackle this question. And that's not all from Anthropic this week. Plus, OpenAI always gives us something to discuss.

Speaking of integrating AI effectively, do you work with a computer or have a software-related hobby? Chances are you have a process in place, but most people don’t know how to speed it up with AI.

That’s where our GenAI Project Bootcamp comes in. Learn to integrate AI into your daily, weekly, or monthly routines.

No idea where to start? No problemo! We’ll help you frame your process and find the sweet spot.

You still have a couple of days before subscriptions to our seventh edition, starting on Monday, end.

Some of our past students have started working on AI projects within their companies after the course.

The Big Picture 🔊

Anthropic Addressing AI Safety in Third-Party Evaluations

Anthropic has launched a new initiative to fund third-party evaluations of AI capabilities and risks, focusing on areas like cybersecurity, CBRN threats, and model autonomy. This aims to improve safety assessments and address the gaps in current methods that can't keep up with rapid AI advancements.

Helin Yontar
HelinCPO and Podcast Host

There is a growing need to assess AI capabilities and risks, but the evaluation landscape for areas such as cybersecurity, CBRN risks, national security, and societal impacts varies significantly.

With this initiative, Anthropic plays an intermediary role by collecting and evaluating projects on all safety-related topics.

OpenAI Uses CriticGPT to Enhance AI

CriticGPT, a GPT-4-based model, critiques ChatGPT's responses to help human trainers identify mistakes during Reinforcement Learning from Human Feedback (RLHF). Using CriticGPT helps trainers find more errors and make better critiques. This tool will be added to the training process to make AI more accurate and reliable.

Gianluca Mauro
Gianluca MauroCEO and AI Rockstar

RLHF is a way to make AI better by using help from people. Human trainers look at the AI's answers, correct them, and rate different responses. This helps the AI learn and improve over time.

Using a special AI like CriticGPT speeds up this process and saves money because it reduces how much human help is needed. This idea of using technology to improve itself has been common in the tech world for a long time.

Anthropic Launches Projects for Better Team Collaboration

Claude Pro and Team users can now organize their chats into Projects, allowing for better workflow integration and team collaboration. The new Projects feature, powered by Claude 3.5 Sonnet, includes a 200K context window, supports files like docx and pdf, and enables custom instructions, enhancing Claude’s assistance with tasks. Users can also share their best Claude-generated work with teammates.

Gioele Mottarlini
GioeleCOO and Image Addict

Last week, we discussed their ‘Artifacts’ feature, and today, we’re talking about ‘Projects.’ Anthropic is definitely creating a more user-friendly product. This new feature is crucial for adoption, as it allows people to share their work, encouraging even those who don’t currently use the tool to engage with it.

Anthropic also offers an alternative to OpenAI’s task-specific GPTs and their Team plan, which we use and believe still has room for improvement.

Have you ever used ChatGPT Team, and if so, how was your experience?

Login or Subscribe to participate in polls.

Bits and Bobs 🗞️

Tribal News 🫂

Tomorrow’s the big day for our GenAI Project Bootcamp students! They’ll pitch their projects to the instructor, other students, and our AI coaches. With just a short time to explain their project’s problem and solution, it’s going to be intense. We can’t wait to see their 2-month work in action!

From Our Channels 🤳

Gianluca shared a quick tutorial on TikTok: “How to integrate an AI LinkedIn copywriter within Slack in 2 minutes.”

Building simple projects like this helps you understand how to turn problems into clear AI projects, combine tools to create solutions, prompt effectively for product development, and develop critical thinking about what makes a good AI product or automation.

@gianluca.mauro

Let’s build an AI automation within Slack in two minutes 🤖 #Ai #learnontiktok #artificialintelligence #business #machinelearning #product ... See more

LOLgorithms 😂

This is the mother of all memes!

That's a wrap on our newsletter! Here’s a quick recap before you go:

Catch you next week! 👋