AI Rats + Politicians 🐀 Nvidia Becomes World's Most Valuable Company 💰 VideoGen Ramps Up 🎥
Your weekly dose of AI insights
Welcome to your weekly dose of AI innovation.
In this edition, we explore advancements in deep tech, consumer-focused video generation, virtual rodents, and the growing influence of AI in politics.
Let’s get into it!
👋 If you’re new here, welcome!
Subscribe to get your AI insights every Thursday.
Dream With Luma AI
The field of text-to-video AI continues to make inroads.
With models like Luma AI's Dream Machine, OpenAI's Sora, Google's VEO and Runway's Gen-3 Alpha, the tech is showcasing impressive capabilities in generating realistic videos from simple text prompts.
While the promise of text-to-video has generated hype for some time, many companies have yet to deliver on their promises. This makes Luma AI's Dream Machine and Runway's Gen-3 Alpha leaders in this field.
Why it Matters
This technological breakthrough has far-reaching implications across industries.
Businesses stand to gain immense value from streamlined video production, reduced costs and the opening of new creative avenues. For individuals, it democratises video creation, allowing anyone to easily bring their ideas to life.
The race to dominate this space is intensifying, with various companies fiercely competing for a leading position in this rapidly evolving landscape.
Here’s some of the best generation we’ve found:
A Tiny Model Beats GPT-4 in Math Benchmarks
Researchers have developed a method called MCTSr combining the language processing power of Large Language Models (LLMs) with a decision-making algorithm known as Monte Carlo Tree Search (MCTS).
This combination improves the ability of LLMs to solve complex math problems. The MCTSr algorithm works by continuously refining and evaluating potential solutions, similar to how a human might approach a challenging problem.
It also uses a smart strategy to explore different solution paths efficiently.
Why It Matters
The research addresses the limitations of LLMs in areas requiring strategic and logical reasoning, especially in mathematics. By combining the strengths of LLMs and MCTS, MCTSr offers a more robust framework for solving intricate reasoning tasks.
The algorithm's effectiveness has been demonstrated through experiments, showing vast improvements in success rates across various mathematical benchmarks, including Olympiad-level problems.
Notably, MCTSr, using the smaller LLaMa-3 8B model, achieved performance comparable to state-of-the-art closed-source models like GPT-4 on certain mathematical evaluations.
With the potential to revolutionise the application of LLMs in complex reasoning tasks, MCTSr could lead to more accurate and reliable AI-driven applications in education and research.
Make AI Great Again!
As the UK approaches its next election, AI Steve emerges as a candidate, representing the first instance of artificial intelligence running for Parliament.
AI Steve, an avatar of Sussex businessman Steven Endacott, will appear on the ballot in Brighton Pavilion. The approach allows voters to engage with AI Steve, ask questions, and suggest policies, with Endacott as the in-person representative in Parliament.
The high volume of interactions with AI Steve hints at greater public curiosity and a potential shift towards more innovative and technology-driven political campaigns.
Why It Matters
AI Steve's candidacy represents a novel experiment in integrating artificial intelligence into democratic processes, suggesting a future where politicians could rely on AI to maintain constant communication with constituents and shape policies based on real-time feedback.
It could also transform voter engagement and policy-making, making politics more responsive and inclusive. For voters, this means a more direct say in policy-making, with their concerns and suggestions promptly addressed and incorporated.
Scientists Create a Virtual Rat
Scientists have successfully created a virtual rat using AI and deep reinforcement learning. This digital rodent not only performs various tasks but also exhibits brain activity remarkably similar to real rats.
Why It Matters
This breakthrough allows researchers to study the neural mechanisms underlying animal behaviour in unprecedented detail.
By observing the virtual rat's brain activity as it performs different tasks, scientists can gain a deeper understanding of how the brain controls complex movements and behaviours.
Key Findings
Inverse Dynamics Models: The virtual rat's AI brain, designed to mimic how the brain controls movement, proved to be a more accurate predictor of real rat brain activity than simply tracking the rat's movements. This suggests the AI successfully captured the complex process of how the brain transforms intentions into actions.
Brain Regions: The virtual rat's neural activity closely mirrored that of the striatum and motor cortex in real rats, suggesting these brain regions play a crucial role in implementing inverse dynamics.
Virtual Neuroscience: This work demonstrates the power of "virtual neuroscience," where complex animal behaviours can be simulated and studied in silico. This approach allows researchers to investigate aspects of neuromotor control that are difficult to study experimentally.
📝 Read an X post by the paper's author
Nvidia Tooling to Boost Large Language Models
Nvidia's journey from an IPO in 1999 to becoming the world's most valuable company with a market cap of $3.34 trillion in 2024 is a testament to its technological foresight and adaptability.
This achievement was fueled by a staggering 591,078% return on its stock since its debut, driven largely by the company's strategic focus on graphics processing units and accelerated computing.
Not only is Nvidia supplying the hardware, but they’re also releasing tools to further lock developers into their ecosystem.
Nvidia just launched Nemotron-4 340B, a family of open models designed to revolutionise the development of large language models. The toolset lets developers generate high-quality synthetic data, a critical but often scarce resource for training LLMs.
Nemotron-4 340B's Instruct model generates diverse data, while the Reward model notably outperforms GPT-4o on reward model training by evaluating and refining the process.
Why it Matters
Nemotron-4 340B solidifies NVIDIA's position as a leading "pick and shovel" provider in the AI industry.
It's not just selling graphics processing units (GPUs), the hardware powering AI models; it's providing the essential tools and resources that make AI development easier and more accessible.
It also aligns with NVIDIA's broader strategy to build a comprehensive AI ecosystem.
By offering a suite of open-source tools like NeMo and TensorRT-LLM combined with the powerful Nemotron models, NVIDIA empowers developers and businesses to build their own custom AI solutions.
This approach reinforces NVIDIA's influence in the industry, as it becomes an integral part of the AI development process.
By lowering the barriers to entry and democratising AI development, NVIDIA hopes to expand its market reach and ensure its continued relevance in the rapidly evolving AI landscape.
📝 See the blog post on NVIDIA's website
📰 Article by Bloomberg on Nvidia's share price
AI Can Predict Your Anxiety
Researchers have developed a new method for predicting anxiety levels using a combination of a short picture-rating task and machine learning. This approach achieved up to 81% accuracy in classifying individuals into 'higher' and 'lower' anxiety groups.
The picture-rating task, which takes only 2-3 minutes to complete, assesses biases in reward/aversion judgment, providing insights into the underlying psychological processes related to anxiety.
Why It Matters
This research could revolutionise mental health assessments by offering a more efficient, scalable, and accessible method for identifying individuals who may be experiencing anxiety.
The use of a simple picture-rating task, easily administered on personal electronic devices, could significantly improve the early detection and intervention of anxiety disorders. Additionally, understanding the specific judgment biases associated with anxiety could inform the development of more targeted and effective treatments.
Smile Robot, Everybody’s Watching
Ex-Robots, a Chinese robotics company, is pushing the limits of humanoid robot development with a focus on enhancing facial expressions and emotions.
Their robots can mimic human facial movements, from smiling to sticking out their tongue, thanks to miniature motors embedded in their facial structures.
This advancement is made possible by the company's in-house software and algorithm teams developing proprietary technology to enable AI to recognise and express emotions.
They're also working on a multi-modal foundation model capable of perceiving the environment and producing appropriate facial feedback.
Why it Matters
Emotional expression and recognition represent a significant leap forward in humanoid robotics. It moves these machines beyond mere automation tools, enabling them to interact with humans on a more personal and engaging level.
This has far-reaching implications for industry, particularly healthcare and education, where emotional interaction is critical.
By developing robots that can understand and respond to human emotions, Ex-Robots is paving the way for a future in which robots are not just tools but companions and assistants capable of enriching our lives.
📰 Article by the South China Morning Post
McDonald's to End AI Drive-through Ordering Trial in Over 100 Locations
McDonald's has ended its AI drive-through ordering trial with IBM, pulling the plug on the voice ordering system in over 100 test locations.
While the official reason is unknown, the decision follows reports of order errors and hints at potential performance issues with the technology.
McDonald's remains committed to exploring AI solutions for its drive-throughs via its ongoing partnership with Google and developing an employee-assisting chatbot called "Ask Pickles."
Why it Matters
It’s a significant moment in the fast-food industry's ongoing experimentation with AI.
The move highlights the challenges and complexities of integrating AI into real-world operations, especially in customer-facing roles where accuracy and efficiency are paramount.
While McDonald's experience may be a setback for IBM, it doesn't signal the end of AI in fast food. Other chains like Wendy's and White Castle continue to invest in AI-powered solutions, indicating that the drive towards automation is far from over.
OpenAI Considers For-Profit Status
OpenAI is reportedly contemplating a significant change in its governance structure. CEO Sam Altman has revealed to shareholders the possibility of OpenAI transitioning into a for-profit entity.
In response to queries about the possibility, OpenAI said:
"We remain focused on building AI that benefits everyone. The nonprofit is core to our mission and will continue to exist."
Why it Matters
This potential shift could mean that OpenAI's non-profit board may no longer control the company's direction.
This move could have far-reaching implications for the future of OpenAI and the broader AI landscape, particularly in how it balances commercial interests with its stated mission of "building AI that benefits everyone."
That’s a Wrap!
If you want to chat about what I wrote, you can reach me through LinkedIn.
Or give my editor a bell through his LinkedIn here.
If you liked it, give it a share!