Welcome to another edition of AI Cohort. As artificial intelligence continues to evolve, it’s crucial to stay informed about the innovations, challenges, and ethical considerations that come with this transformative technology. Here's a roundup of some of the most notable AI releases, launches, innovations and breakthroughs.

Google's secretly working on "Project Jarvis"

Google is reportedly developing an advanced AI agent codenamed "Project Jarvis" that can autonomously operate within the Google Chrome browser to perform various web-based tasks. Google may demonstrate this innovation as early as December 2024. This aligns with the expected release timeframe for the next major version of Google's Gemini large language model.

It's something Google CEO Sundar Pichai spoke about in the recent past - "I think about AI agents as intelligent systems that show reasoning, planning, and memory. Are able to think multiple steps ahead, work across software and systems, all to get something done on your behalf, and, most importantly, under your supervision."

According to 'The Information' Project Jarvis is designed to take over a user's web browser to complete various tasks automatically. This is a major step beyond traditional chatbots or language models, as it can directly interact with web interfaces.

The AI agent is expected to:

• Conduct online research

• Make purchases

• Book flights

• Fill out forms

• Navigate web pages

• Click buttons and input text

While the report provides an intriguing glimpse into Google's plans, details about Project Jarvis's specific capabilities or how it might integrate with Gemini remain unclear. The codename suggests it could be an AI assistant or interface, but this is speculative without further information.

Anthropic Unveils Claude 3.5 Models with Groundbreaking Capabilities

Anthropic made waves with the release of upgraded AI models and a revolutionary new feature:

• Claude 3.5 Sonnet: An improved version of their mid-tier model, now available to all users. It shows significant performance gains, especially in coding tasks.

• Computer Use Feature: A groundbreaking capability allowing Claude to interact with computer interfaces by moving the cursor, clicking, and typing. This opens up new possibilities for AI-powered task automation.

• Claude 3.5 Haiku: A new model combining rapid response times with improved reasoning, set to be released later this month

The revolutionary "computer use" capability allows Claude to interact with computers just like we do! Imagine an AI that not only understands your commands but can also move the cursor, click buttons, analyse screenshots, and even type filling out forms - it’s like having a digital assistant that truly gets you.

Apple Study Uncovers Fundamental Flaws in AI Reasoning Capabilities

Apple's AI research team has conducted a study that reveals significant limitations in the reasoning abilities of large language models (LLMs). Here are the key findings:

1. Lack of True Reasoning: The study suggests that LLMs rely on sophisticated pattern matching rather than genuine logical reasoning.

2. Performance Degradation with Minor Changes: Even minor changes in question phrasing or the addition of irrelevant information can cause major discrepancies in model performance.

3. Mathematical Reasoning Fragility: LLMs struggle with mathematical problems when presented with unfamiliar representations or additional contextual information.

4. Introduction of New Benchmark: Researchers introduced the GSM-Symbolic benchmark to better assess LLMs' mathematical reasoning capabilities.

5. Industry-wide implications: The study tested multiple LLMs from major companies, finding similar limitations across the board.

6. Challenge to AGI claims: The findings question the notion that current LLMs are close to achieving artificial general intelligence.

7. Future Research Directions: The study emphasizes the need for fundamental advancements in model architecture to develop genuine reasoning capabilities.

This research has significant implications for the AI industry, highlighting the current limitations of LLMs and the challenges ahead in developing truly intelligent systems capable of human-like reasoning.

Academic Research and Conferences

Continuing on the topic of AI research, Imperial College London's I-X centre recently held the second edition of its Breaking Topics in AI Conference, bringing together leading experts in artificial intelligence and machine learning. The conference featured keynote speakers presenting the latest trends and emerging challenges in AI research, covering topics such as machine learning in molecular studies, interactive explanations for contestable AI, AI safety, and the future of healthcare in light of recent AI developments.

And that's a wrap of this edition, thanks for reading the newsletter! Stay tuned for more and keep your subscription active with AI Cohort.