A new AI model can explain its thought process

Hola! Here’s what you need to know about AI today:

👉 UAE’s new AI model can explain how it arrived at an answer

👉 US lays out new set of AI export rules

👉 AI might replace midlevel engineers in 2025

and many more!

📧 Did someone forward you this email? Subscribe here for free to get the latest AI news everyday!

Read time: 4.8 minutes

What’s going on: Researchers at the MBZUAI, a top AI university in UAE, released LlamaV-o1, a state-of-the-art AI model capable of tackling some of the most complex reasoning tasks across text and images. LlamaV-o1 establishes a new standard for step-by-step reasoning in multimodal AI by integrating advanced curriculum learning with optimization methods such as Beam Search.

What does it mean: Traditional AI models typically just give you the end result without showing how they arrived at that answer. In contrast, LlamaV-o1 focuses on step-by-step reasoning, much like how humans solve problems. This feature lets users follow the logical progression of the model's thought process, which is especially useful in scenarios where understanding how decisions are made is crucial.

More details:

• LlamaV-o1 was trained on LLaVA-CoT-100k, a dataset optimized for reasoning tasks, and evaluated its performance using VRC-Bench benchmark. VRC-Bench is a new benchmark with over 1,000 samples and 4,000 reasoning steps.

• LlamaV-o1 achieved a reasoning step score of 68.93, outperforming some closed-source models like Claude 3.5 Sonnet.

• LlamaV-o1 shines in areas like medical image processing. When it arrives at a certain diagnosis, it can provide transparent, step-by-step reasoning that professionals can review and validate.

• This open-source model understands charts and diagrams very well, which can be very helpful in financial analysis and decision-making. It also can be used in content generation and to create conversational agents.

What’s going on: The Biden administration has introduced a new policy designed to simplify the sale of AI technology and computing resources to US allies, while simultaneously preventing these technologies from reaching nations that are considered a concern, mainly China and Russia.

What does it mean: It's important for US safety and economy that this key tech stays in America and that the world uses American systems for AI and put in place critical security and trust standards while building their AI ecosystems.

More details: 

• The rule does not restrict chip sales to 18 US allies and partners.

• Chip orders with collective computation power up to roughly 1,700 advanced GPUs do not require a license and do not count against national chip caps.

• Non-verified end user entities located outside of close US allies can still purchase large amounts of computational power, up to the equivalent of 50,000 advanced GPUs per country.

• This new rule expands on earlier measures from October 2022 and October 2023, which were designed to safeguard US national security, specifically through controls on chip exports.

🤖 Mark Zuckerberg stated on a recent podcast, The Joe Rogan Experience, that AI could automate the work of Meta's midlevel engineers in 2025, eventually handling all coding for the company's apps. (Full podcast episode on YouTube)

🖥 Microsoft has created a new engineering organization within the company aimed at accelerating AI infrastructure, led by a former Meta’s head of engineering, Jay Parikh.

📄 AI chip startup, Blaize, is set to go public on the Nasdaq on Tuesday. Blaize is founded by former Intel engineers and has raised $335 million from investors like Samsung and Mercedes-Benz since its 2011 launch.

🧠 Anthropic, the company behind Claude AI chatbot, achieves ISO 42001 certification for creating responsible AI.

📷 Adobe’s new AI tool, Firefly Bulk Create, allows users to replace, remove, or extend image backgrounds in large batches, up to 10,000 images.

🚗 Google Cloud and Mercedes-Benz are collaborating to introduce the Automotive AI Agent, enhancing the MBUX Virtual Assistant with advanced conversational AI capabilities in the new Mercedes-Benz CLA.

🚌 Montgomery County Schools are implementing cameras equipped with AI to identify and penalize drivers who illegally pass school buses.

📽 In the new film "Here", Tom Hanks and Robin Wright are digitally de-aged using AI technology, allowing them to portray characters across different stages of life.

AI and natural disasters

This prompt is highly applicable for those in environmental science, emergency management, or AI for good initiatives. It encourages thinking about the integration of AI with real-world problems, like disaster prediction, which involves data analysis, machine learning models, and real-time response systems.

Blue Origin - Senior Data Scientist - Operations Technology

Typeface - Staff Product Designer

Synechron - Gen AI Developer

Warner Bros. Discovery - Sr Director, AI Lab

Thank you for staying with us like always! If you are not subscribed, subscribe here for free to get more of these emails in your inbox! Cheers!