Anthropic Calls All Model Evaluators

In partnership with

Hello readers,

Welcome to another edition of This Week in the Future! Anthropic is calling all model evaluators in an initiative to improve safety assessments of advanced AI models. Plus, AI-driven innovations in payment processing and prioritizing the human element in AI.

Better Safe Than Sorry

Anthropic, the OpenAI rival behind Claude, is taking the initiative to encourage the development of third-party model evaluations. The reason? Safety. Anthropic says a robust evaluation ecosystem is needed for assessing AI capabilities and risks.

Anthropic is asking for input from experts on a variety of AI Safety Level assessments in areas such as cybersecurity, chemical, biological, radiological, and nuclear risks, model autonomy, national security, social manipulation, and misalignment. If you think you have something to contribute, you can submit a proposal here.

Beyond safety assessments, Anthropic wants to develop evaluations for advanced model capabilities and infrastructure, tools, and methods for developing evaluations.

Our Take

Given the nature of the evaluations being asked for, Anthropic must fully expect these models to scale to dangerous levels of capability. Both the utopian and dystopian predictions for AI depend on the scaling hypothesis being true and/or a breakthrough. As of now, we have chatbots based on technology that hasn’t changed in seven years. This is not to say Anthropic shouldn’t research safety. After all, better safe than sorry.

🔥 Rapid Fire

• Value of gen AI scrutinized in new report from Goldman Sachs

• Retool takes pragmatic view on AI in State of AI 2024 report

• Cloudflare releases new feature to block AI web scrapers

• GXO Logistics signs multi-year deal with Agility Robotics

• HCLTech and IBM announce Generative AI Center of Excellence

• AI startup Adept gets absorbed by Amazon’s AGI organization

• French startup Kyutai releases GPT-4o competitor with voice mode

• Salesforce releases small model for API function-calling

• Google’s electricity needs for AI set back climate goals

• Apple launches public demo of its 4M AI model

• Meta introduces 3D Gen for text-to-3D asset generation

• Perplexity adds Pro Search updates for advanced problem-solving

• OpenAI demonstrates using GPT-4 to find GPT-4’s mistakes

• Apple set to get an observer role on OpenAI board

• ChatGPT Mac app was storing conversations in plain text

• Mind-reading AI recreates what you’re looking at

Join the live session: automate compliance & streamline security reviews

Whether you’re starting or scaling your company’s security program, demonstrating top-notch security practices and establishing trust is more important than ever.

Vanta automates compliance for SOC 2, ISO 27001, and more, saving you time and money — while helping you build customer trust.

And, you can streamline security reviews by automating questionnaires and demonstrating your security posture with a customer-facing Trust Center, all powered by Vanta AI.

Join Vanta’s 45-minute live session on July 9th at 12 pm PST to see the platform in action and ask your questions.

📖 What We’re Reading

AI-Driven Innovations in Payment Processing

Prioritizing the Human Element in AI

💻️ AI Tools and Platforms

• Eraser → AI copilot for technical design

• BuildBetter → AI for product teams

• Letter AI → AI for revenue teams

• Touch → Automate sales prospecting

• Daloopa → AI data for financial modeling