AI Leadership Weekly

Issue #24

Welcome to the latest AI Leadership Weekly, a curated digest of AI news and developments for business leaders.

Top Stories

Source: Hugging Face

DeepSeek V3 the new leader after quiet update
The recent upgrade to DeepSeek V3 (helpfully titled DeepSeek-V3-0324) brings improvements to both reasoning and programming capabilities, and has leapfrogged the competition in benchmark results.

Their team seems to have focused on gaining some real-world performance in various domains, as well as improving efficiency. The model has 685B parameters, but they are spread across its "mixture of experts" which only engage 37B parameters at a time.

It's available on Hugging Face under the MIT license, and, hence, is open source. The biggest things is that it only costs $0.14 per million input tokens, versus Claude 3.7 Sonnet which costs $3 per million input tokens.



Gemini is watching you (and your screen)
Google's Gemini can now see your phone screens and camera feeds in a recent update.

This feature lets you ask questions about either image source, and was born out of their Project Astra from a year ago. The feature is slowly rolling out to Google One AI Premium subscribers, with wider availability expected to come soon.

Source: ARC Prize

The ARC Prize is back
The ARC Prize Foundation has returned with a new benchmark competition with a prize pool of $1 million.

The original ARC-AGI-1 benchmark from 2019 aimed to test capabilities in deep learning, while the new ARC-AGI-2 focuses on reasoning systems. Each challenge in the new set was solved by at least two humans in two or fewer attempts, and matches the conditions allowed for AIs (i.e., two attempts).

In Brief

Market Trends

Cloudeflare traps AI in a maze
Cloudflare, one of the world's leading web infrastructure providers, announced on Wednesday the release of their "AI Labyrinth", which they say aims to halt AIs from scraping websites.

It's been known for years that AI model creators such as OpenAI scrape data from numerous sources, including websites, and that they don't always respect website "rules" blocking access to bots. So, instead of simply blocking these bots, Cloudflare lures them into what they term a "maze" of realistic-looking data and pages, which will ultimately waste compute resources of the scraper.

They say that the standard approach of blocking doesn't always work, and usually just means the scraper is alerted to the failure, and can then search for a work-around.

Meta fails in bid for AI chip manufacturer
Most of the major AI firms have custom chip manufacturing in their pipeline, and Meta recently attempted to purchase South Korean AI chip startup FuriosaAI for $800m.

The bid was unsuccessful, however.

Reportedly, the company wasn't satisfied with discussions around post-purchase business strategy and organisational structure. FuriosaAI is also reportedly in talks with other investors, seeking $48m in investor funding.

Tools and Resources

Sider
"Deep research in minutes" is their claim, including highlights, notes, and visualisations.

Base44
Takes your idea and turns it into a fully-functional app, which includes database set-up, email systems, authentication, and more.

10X
Learn about anything you want by turning it into a game!

Recommended Reading

Vibe coding a full game in a week
Popular YouTuber Primeagen and friends are half-way through a week-long project to create a tower defence game in Lua, using the power of vibe-coding. Can it be done?? You’ll have to watch the series (either the daily highlights or live streams) to find out!

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!

Your AI and Data Team as a Subscription

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.