Anthropic is now building AGI?

In today's newsletter, we'll explore Mistral's powerful new coding model, discuss WhatsApp’s latest feature, discuss Jan Leike's move to Anthropic, and introduce the AI safety program.

A leading researcher at OpenAI recently left the company to join Anthropic. He was working on “superalignment”, OpenAI’s codename for AGI Project, he recently joined AGI.

In fact, Ilya left a few weeks back as well. Musk raised $6bn, so there’s probably some “realignment” on the company who will end up building AGI.

Let’s dive in!

Stuff you should know

Mistral launches an Open Source Coding Model

Codestral, a new AI code assistant from Mistral, is trained on over 80 programming languages, including popular ones like Python, Java, C++, and JavaScript, as well as more specific ones like Swift and Fortran.

As a 22B model, Codestral sets a new standard for code generation performance, particularly in the performance/latency space compared to previous models used for coding. With its larger context window of 32k (compared to 4k, 8k, or 16k for competitors), Codestral outperforms all other models in RepoBench, a long-range evaluation for code generation. This broad language base ensures Codestral can assist developers in various coding environments and projects.

Codestral saves developers time and effort by completing coding functions, writing tests, and filling in partial code using a fill-in-the-middle mechanism, helping to level up the developer's coding game and reduce the risk of errors and bugs.

Generate Images directly inside WhatsApp

WhatsApp is testing a new feature that allows users to quickly generate AI-powered images within chats. This feature utilizes Meta AI, which is currently available in select countries. The feature is currently in testing and is expected to simplify the image creation process for users by adding a shortcut to the chat attachment sheet. This shortcut will enable users to generate images efficiently without having to switch to their Meta AI chat.

The new feature is part of WhatsApp's efforts to integrate AI more deeply into its app, making it easier for users to create and share AI-generated content. The feature is currently available only to select users in certain countries, with plans for broader availability in future updates. The shortcut will save time and effort, especially for those who frequently use images in their conversations, and offers a more personalized way to express oneself, particularly for those who struggle to choose a suitable profile picture.

Anthropic Poaching OpenAI?

Jan Leike, a prominent AI researcher, has resigned from OpenAI to join Anthropic, where he will lead a new "superalignment" team focused on AI safety and security. Leike's decision to leave OpenAI came after he publicly criticized the company for prioritizing commercial products over crucial safety research.

At Anthropic, Leike will report directly to Chief Science Officer Jared Kaplan, and some Anthropic researchers currently working on scalable oversight will join his team. The Superalignment team aims to address key technical challenges in controlling superintelligent AI within the next four years, a goal that faced obstacles due to OpenAI's leadership priorities.

Anthropic has positioned itself as more safety-focused than OpenAI. The company was founded by former OpenAI VP of research Dario Amodei, who reportedly left OpenAI due to disagreements over the company's increasing commercial focus. Amodei brought several former OpenAI employees with him to launch Anthropic, including former policy lead Jack Clark.

What is ARIA?

NIST (National Institute of Standards and Technology) has launched a new program called ARIA (Assessing Risks and Impacts of AI) to advance sociotechnical testing and evaluation for AI. This program aims to help organizations and individuals determine whether a given AI technology will be valid, reliable, safe, secure, private, and fair once deployed

Key Stats: NIST's ARIA Program

• Program Name: Assessing Risks and Impacts of AI (ARIA)

• Launched by: National Institute of Standards and Technology (NIST)

• Goal: Advance sociotechnical testing and evaluation for AI systems

• Aims to: Help organizations and individuals determine the validity, reliability, safety, security, privacy, and fairness of AI technologies once deployed

Objectives

1. Assess risks and impacts of AI systems within societal contexts

2. Develop methodologies and metrics to quantify AI risks

3. Support AI safety and ensure alignment with societal values

4. Enhance transparency and accountability in AI development and deployment

5. Support AI regulation by providing a framework for evaluating AI risks and impacts

Expected Outcomes

• Significantly advance AI safety by providing a framework for evaluating AI risks and impacts

• Ensure AI systems are designed and deployed to align with societal values and norms

• Enhance transparency and accountability in AI development and deployment

• Support policymakers and regulators in making informed decisions about AI

Hand Picked Video

In this video, we'll look at the EU's groundbreaking AI Act. The EU AI Act is a comprehensive legal framework designed to regulate artificial intelligence within the EU.

Top AI Products from this week 

• Frontly - Generate apps with AI and edit them in a no-code drag-and-drop interface. Quickly launch your own AI apps with modern responsive layouts, trigger advanced actions, sync with automation platforms via webhooks and create your own library of reusable blocks.

• Lawformer AI - Lawformer's AI powered contract deconstruction tool allowing legal teams to create personalized clause libraries and manage their internal knowledge in one place.

• VIVA - Generate AI videos that captivate with stunning results, coupled with 4K Video Enhancement and beginner-friendly automatic prompt optimization, delivers an unparalleled video creation experience for you.

• timeOS 2.0 - timeOS is an AI-powered New Tab page designed to help you prepare for your next meeting. Hover over events to access relevant context from Gmail, past meeting notes, and LinkedIn.

• Syllaby V2.0 - Syllaby, a 5-in-1 digital marketing powerhouse tool, harnesses AI to simplify viral video creation. It aids in ideation, scheduling, scripting, and even avatar-based video production.

• Coverse.one - Coverse is a new way to work with documents. Gather Google Docs, Sheets, Word; Excel, Figma; Miro, Airtable, PDFs; screenshots, and others in one place.

This week in AI

• 3D Product Video Generator - Doly, a revolutionary app by AniML, simplifies 3D video creation for e-commerce. Using Gaussian splatting technology, it transforms still images into lifelike 3D models, offering diverse templates for dynamic product scenes. With a user-friendly interface, Doly enables users to generate professional product videos directly from their iPhones, potentially reshaping online retail marketing

• OpenAI Board create new Safety and Security Committee - OpenAI forms Safety and Security Committee to assess and strengthen processes and safeguards amid growing concerns over AI technology. The committee, led by Bret Taylor, Adam D'Angelo, Nicole Seligman, and Sam Altman, will evaluate and develop OpenAI's processes over 90 days, sharing recommendations publicly

• Character.ai Partnerships with Meta and xAI - Character.ai in talks with Meta and xAI about potential AI research partnerships, as tech giants seek to collaborate with startups while navigating regulatory concerns around acquisitions.

• Now YouTube also has similer feature like Shazam - YouTube Music for Android now allows users to search for songs by humming or singing into their phones. The feature uses AI to recognize and display song details within seconds, similar to Shazam.