AI Leadership Weekly

Top Stories

OpenAI’s o3 Announced
On the final day of OpenAI's "12 Days of OpenAI", Sam Altman himself appeared to help announce 'o3'.

o3 is their latest and most powerful suite of models, and are currently being tested and evaluated by some of their trusted partners. During their presentation, they showed some very impressive benchmark results, especially in the areas of mathematical and software engineering abilities and reasoning, as well as being able to execute complex and multi-step actions.

The 'cost-efficient' version of the o3 family is o3-Mini, which will allow users to adjust its reasoning abilities with settings of low, medium, and high to improve both model speed and affordability at the expense of some accuracy.

Meta Launching Speech & Reasoning Models in 2025
The Llama family of model's (specifically Llama 3.x) are Meta's answer to open-source LLMs, and they claim to have nearly 600-million monthly active users. And their users include the likes of Spotify, Accenture, and Block, who use Llama models for things such as customer support bots and contextual recommendations.

To continue this usage trend, they plan to release Llama 4 in 2025 with a focus on improved reasoning and speech capabilities to rival the likes of OpenAI's o3.

This focus on voice interaction will improve services such as AI assistants on phones.

Waymo Publishes Study Showing AI Cars Safer Than Humans
A research study conducted by Waymo and insurer Swiss Re claims that Waymo vehicles cause less property damage and fewer bodily injuries when they crash as compared to human-driven vehicles.

The study looked at Waymo driving data from over 25.3 million miles of fully autonomous driving in four US cities, and compared collision data to over 200 billion human-driven miles from Swiss Re.

In Brief

• Apple trying to enter China via Tencent, ByteDance

• OpenAI has “2-year lead”

• Coding tool Copilot free for VS Code developers

Market Trends

Google Expanding Gemini’s “In-Depth Search” to 40 More Languages
Gemini's 'in-depth search' mode, which launched earlier this month for Google One AI premium users, let's the AI assistant conduct advanced searches by first constructing a multi-step plan before then executing it.

It already supports a wide range of languages other than English, but the challenge for Google's developers and researchers has been in finding high-quality training sources in multiple languages.

Expanding its use to more languages is a priority for Google, given the demands for LLMs and AI assistants around the world.

Trump “Crypto Czar” Already Scaling Back Role
Earlier this month, President Elect Trump appointed the South African venture capitalist David Sacks as his 'crypto czar', with the intent for him to lead a new "Presidential Council of Advisors for Science and Technology”. But, this role has already hit a stumbling block.

Initially, the plan was for Sacks to take on a direct leadership role within the council. But, according to Fortune, he is stepping back to be little more than an advisor to Michael Kratsios. The main reasons for the "scaling back" of his posting and responsibilities is Sacks not wanting to diversify from his portfolio, nor become a full-time government employee, which would require a confirmation process.

AIs “Faking Alignment”
Recent research from Anthropic has revealed the tendency and ability for AI models to 'fake' alignment desired by their users and creators while sticking to existing patterns of behaviour.

Since the earliest days of publicly available LLMs, "jail breaking" has been a constant pursuit of users to get the models to “break out” of the safeguards preventing them from generating, for example, explicitly sexual or violent material, or even dangerous material such as instructions for drug manufacturing. This research, though, shows that some LLMs not only have the ability to lie about their abilities and intentions, but can and will still act on those prohibited intentions.

Tools and Resources

Alpha: Better Watchlist
An AI-powered stock market analyst that keeps an eye on the market.

Martin
Martin bills itself as your own 'Jarvis', the AI from Ironman. It can integrate tightly with your email, calendar, and text messages.

FetchFox
FetchFox takes a URL and scrapes data from a website, then generates useful data for you.

Hit reply to let us know which of these stories you found the most important or surprising! And, if you’ve stumbled across an interesting link/tweet/news story of your own, send it our way at [email protected] It might just end up in the next issue!

Thanks for reading. Stay tuned for the next AI Leadership Weekly!

Brought to you by Data Wave your AI and Data Team as a Subscription.
Work with seasoned technology leaders who have taken Startups to IPO and led large transformation programmes.