Worms In AI 🐛 ‘Extinction-Level’ Threat & iRobot Is Here 🤖
Your Weekly AI Recap
Welcome to the biggest headlines in AI this week. Let me know what you think.
iRobot Is Basically Here & It's Mind-Blowing
Figure just showcased their Figure 01 humanoid robot.
It uses OpenAI’s multi-modal models to understand the world around it. This feels like a watershed moment: Advanced robotics coupled with bleeding-edge AI.
The demonstration follows Figure’s successful February 29 raise of US$675M from the likes of Microsoft, OpenAI, Amazon, Nvidia, Ark Invest, Intel, and others.
The race is on
Tesla’s Optimus is not far behind, Tesla cars send back “billions of miles” of data to be processed by Dojo, their supercomputer, and this data is also being used to train Optimus.
Cyber Worms Threaten GenAI
An academic research paper on March 5 introduced a new threat known as GenAI worms which explored how malware could be used to exploit the GenAI component of an agent and “launch cyber-attacks on the entire GenAI ecosystem.”
Interconnected models like ChatGPT and Gemini are vulnerable to this novel hack, the study found.
When prompted, the AI will use a method called RAG to reference knowledge outside of its training data before generating a response.
The trouble is the information the AI is prompted to retrieve is from a malicious source which then “poisons” it.
The worm then alters the GenAI model's output to control an application's behaviour, and enable it to spread even further.
Why it Matters
The worm's self-replication and zero-click spread make it a formidable threat and integration with critical applications could lead to devastating effects.
This is the first GenAI virus, and we’ll likely start to see an explosion of new novel attacks as this technology evolves.
🍿 This video explains it well (don’t mind the voice)
Devin, the World's First Fully Autonomous AI Software Engineer
This autonomous AI can build apps, fix bugs, and seamlessly collaborate with human developers.
With remarkable reasoning and problem-solving abilities, Devin can plan complex tasks, learn over time, and even train its own AI models.
🍿 Another 2 minute paper explanation video
Elon Open Sources 𝕏’s AI Grok in Response to the OpenAI Lawsuit
Elon Musk is escalating his criticism of OpenAI by open-sourcing xAI's AI chatbot, Grok.
This move is a direct response to his lawsuit against OpenAI, where he alleges they have abandoned their original open-source mission in favour of a closed-source, profit-driven model backed by Microsoft.
Musk positions this as a return to his commitment to transparent AI development and aligns with his history of supporting open-source principles at Tesla and X.
📄 More information in this article
'Extinction-Level' Threat From AI, US Government-Commissioned Report
A new U.S. State Department report warns of catastrophic security risks from advanced AI, citing interviews with over 200 industry experts. The report argues that AI could be weaponised or become uncontrollable, presenting an "extinction-level threat".
“The rise of advanced AI and AGI [artificial general intelligence] has the potential to destabilize global security in ways reminiscent of the introduction of nuclear weapons.”
The report proposes drastic restrictions on AI development, including a ban on training models above a certain power level, government oversight of smaller advanced models, potential bans on publishing AI code, and tighter control over AI chip exports.
These measures seek to limit the uncontrolled growth and potential weaponisation of AI technology.
Such restrictions would hinder U.S. progress in a field where rivals like China are surging ahead.
Other countries are looking at similar regulation
The EU's also just introduced the AI Act will ban harmful AI systems like social scoring and restrict the use of high-risk technologies like facial recognition.
India is also preparing to introduce the Digital India Bill, aiming to regulate AI, online content, and semiconductor R&D. The bill, championed by Minister Rajeev Chandrasekhar, seeks to address
📄 Excellent article on the report from The Times
Indie Filmmakers Leverage GenAI
GenAI is allowing Indie filmmakers to use stunning VFX previously reserved for big-budget productions.
The project combined live-action footage with AI-generated VFX supervised by Robert Nederhorst (John Wick 3, X-Men).
AI-powered tools put Hollywood-level effects within reach, empowering indie filmmakers to tell visually ambitious stories.
This levels the playing field, giving independent voices a powerful new tool for cinematic expression, though there is sure to be pushback from the Screen Actors Guild-American Federation who’ll likely perceive it as another threat to the U.S. filmmaking industry.
🍿 Here's a library of AI shorts (worth a watch!)
Data Is the New Oil
Companies like Reddit and Substack are capitalising on their user-generated content. With over 100,000 niche communities (subreddits), Reddit offers a rich dataset of user interactions and preferences.
Despite its long history and potential for an IPO, the platform has yet to turn a profit in its nearly two-decade existence.
As the landscape remains dry for social media IPOs since Pinterest in 2019, Reddit found a lucrative revenue stream by selling this valuable data.
Companies like Google pay Reddit a reported $60 million annually, in a non-exclusive agreement, to train their advanced AI models.
This was a great read! I really enjoyed this weeks newsletter. Thank you for the great and engaging content, I feel like I am starting to wrap my head around some of the AI information out there!! Much appreciated and keep it coming!