From scrutiny by Elon Musk's lawyers to lawsuits by news publications, OpenAI can't seem to catch a break, and a protest group leaking its unreleased GenAI video model to the public is probably the last thing it needed. Other key highlights include:

• Elon Musk looks to create an AI games studio with xAI to "make games great again"
• Alibaba joins the fray as it launches an "open-source" reasoning model to combat OpenAI's o1
• Amazon may look to lessen its reliance on Anthropic by beginning work on a new generative AI model

Join us at AI Tangle as we untangle this week's happenings in AI!

THE BIG AI STORY

After months of showcasing and teasing the public with its unreleased AI video generator Sora, a protest group may have come back to bite OpenAI when it was least expecting it. For a short while, an upgraded variant of OpenAI's multimodal AI model was leaked to the public for users to take it out for a spin with nothing but a queue to hold them back - but how and why did this happen in the first place?

​

Let's elaborate on the situation a little.

Earlier this week, a group called "Sora PR Puppets" leaked access to OpenAI's Sora to protest what it describes as unfair practices and "art washing" by the company. The group is accusing OpenAI of pressuring early testers, including red teamers and creative partners, into promoting Sora while offering little to no compensation for their efforts. Using early access tokens, the group enabled 1080p, up to 10-second-long video generation of a presumably "turbo" variant of the model via a public frontend before OpenAI shut it down three hours later, according to a claim by the group.

5 QUICK HITS

Elon Musk recently announced plans via a post on X/Twitter to enter the video game industry through his AI company, xAI, by creating an AI game studio to "make games great again." Musk stated that his reason for wanting to create a game studio was due to the unnecessity of left-wing politics (or "wokeness" in gaming) and his discontent with corporate-owned game studios. However, some were quick to point out the irony in his latter statement, as Elon Musk is the CEO of multiple massive corporations himself.

​

Not long after the launch of the Chinese-developed DeepSeek-R1 reasoning model, another one joined the fray, but this time by Chinese giant Alibaba, codenamed QwQ-32B-Preview on HuggingFace. Much like the company's other models, this one was developed by Alibaba's Qwen team and, at least per Alibaba's own testing, outperforms the likes of OpenAI's o1 and o1-mini on certain benchmarks. However, though Alibaba's QwQ-32B-Preview is "open-source" under the Apache 2.0 license, meaning it can be used commercially, this only includes specific components, making it impossible to reproduce the entirety of the model.

​

​According to an article by The Information, e-commerce bigwig Amazon is reportedly developing a new generative AI video model, code-named Olympus, for processing images and videos, obviously. The goal is cited as likely to reduce Amazon's reliance on Anthropic's Claude chatbot, widely used on AWS. Olympus will allow users to use text prompts to analyze, dissect, and search whole scenes, such as specific basketball moments. Still, despite this, Amazon was still keen to pour another $4 billion into Anthropic, doubling its total.

​

British Prime Minister Keir Starmer recently inaugurated London's first Google-funded AI Campus in Camden with the goal of teaching young people AI and machine learning skills. The campus, launched in partnership with Google DeepMind and local authorities, is piloting a program with 32 students aged 16-18 who will work on real-world AI projects in fields like health, social sciences, and the arts. Google UK's Debbie Weinstein also announced £865,000 in funding for an AI literacy program to train teachers, aiming to reach 250,000 students by 2026.

​

​Pathway, a startup focused on "Live AI" systems that update and learn in real-time, recently announced that it had raised $10 million in a seed round led by TQ Ventures. The company offers infrastructure that allows enterprise AI platforms to process live, structured, and unstructured data, with companies like NATO, Intel, and F1 leveraging it handily. Company CEO Zuzanna Stamirowska emphasizes the challenge of integrating memory and dynamic knowledge into AI systems, calling current models "very smart interns" that "can't really memorize."

4 AI TOOLS

​TwinMind - Designed to be a creative AI sidekick that understands what you see and hear on your screen, TwinMind aims to boost productivity, be it writing, understanding tabs, or answering proactively.

​Dicte - Dicte is an AI-powered meeting assistant that turns voice notes and recorded meetings into reports and minutes automatically with seamless integration.

​Roundtable - Clean your survey responses with Roundtable, an easy-to-integrate API with behavioral tracking, dynamic clustering, and more to cut down on time spent analyzing survey data.

​Speax - Dub your videos with seamless voice synchronization and turn them multilingual across 29+ languages with Speax's advanced AI-based platform.

AI EXTRA READ

Discovering Hidden Weather Patterns With AI (3-min read)

With the amount of data to process growing ever-larger, researchers have begun employing AI to analyze historical climate data in Europe, finding new historical temperature extremes in the process and improving accuracy over time-tested measures in regions where data is still lackluster.