AI News Now - McKinsey Sees Generative AI Adding 4 Trillion to the World Economy, Next Gen LLaMA Goes Full Commercial, and the UK Invests Big in AI

Unleashing the Economic Superpower: The Rise of Generative AI Will Add 4.4 Trilion to the Economy

In a world where technology is rapidly evolving, Generative AI is taking center stage, promising to infuse a whopping $4.4 trillion annually into the global economy. In the latest report?from analysts at McKinsey they outline the potential economic impact of generative AI.?

Here’s some key highlights from the report:

• The report speculates that generative AI could add $2.6 trillion to $4.4 trillion annually across the 63 use cases.?To put that in context, the United Kingdom's entire GDP in 2021 was $3.1 trillion and the United States came in at 23 trillion in 2022.
• Much of the value of generative AI, as much as 75% comes in four areas: Customer operations, marketing and sales, software engineering, and R&D.?
• But it won’t just be those sectors that stand to benefit. Generative AI will have a significant impact across almost every major sector on Earth, including high tech, banking, and the life sciences.
• Generative AI also holds the potential to dramatically change the very nature of work, augmenting individual workers as much as small and big business. Current generative AI tech already shows the possibility to automate the boring tasks that take 60-70% of employees' time today, so they can focus on more productive and creative work. The acceleration largely comes from generative AI's ability to understand natural language and to reason about what’s being read or said, which makes up as much as 25% of total work time.
• The McKinsey folks see the pace of automation accelerating and they estimate half of today's work activities could be automated between 2030 and 2060, with a midpoint in 2045, or roughly a decade earlier than previous estimates.

The AI landscape is growing at a tremendous clip and its tools and resources that are making this technology more accessible and user-friendly every day. McKinsey expects AI tools to tremendously change how we work and that we’re on the verge of machines giving people "superpowers" that will turbocharge the global economy.

Meta’s New LLaMA Will Be Licensed for Commerical Use

Meta is stepping into the ring with OpenAI and Google DeepMind, ready to throw some punches with its commercial versions of the LLaMA language model. The last flavors of LLaMA (coming in 4 versions based on parameter count) made a big impact in academia as researchers raced to add capabilities and test ideas on the cheap.?But there was a big problem.?LLaMA wasn’t licensed for commercial use.?That’s about to change with Meta’s latest version which is set to release imminently.?If the company sticks to their promise and isn’t blocked by legislators worried about nonsense fantasies of AI economic apocalypse then expect LLaMA to have a massive impact on open source AI and commercial AI driven applications that might finally deliver competition to the OpenAI powerhouse.

UK Government’s ￡100M Leap into Next-Gen AI Development

The UK government has decided to up the ante in the AI game, with a cool ￡100 million investment to fund a joint government-industry taskforce. This isn’t just about accelerating the UK’s AI capabilities, it’s also about ensuring the safety and reliability of AI systems and building the UK’s ‘sovereign’ national capabilities. The investment also covers a new ‘exascale’ supercomputer dedicated to AI Research.?It’s clear that while many governments seem fearful or reticent about AI, the UK is embracing it with relish and positioning themselves to be one of the top places to build and do business with artificial intelligence.

Mastering Imitation: The New Era of Inverse Reinforcement Learning

In a new paper, researchers from CMU and Cornell outline Inverse Reinforcement Learning (IRL), where imitation is the name of the game. The researchers realized that traditional IRL methods suffer from a big computational weakness: they require repeatedly solving a hard reinforcement learning (RL) problem as a subroutine. In this work, they demonstrate that a more informed imitation learning reduction the uses the state distribution of the expert to alleviate the global exploration component of the RL subroutine, providing an exponential speedup. And in practice, they find that they’re able to significantly speed up continuous control tasks.

Exploring Liquid Neural Networks that Adapt on the Go

Dive into the world of “liquid” neural networks a new kind of network from MIT CSAIL researchers, featured in IEEE Spectrum’s latest piece. These networks are modeled on the brains of c-elegans and they’re not only compact in size but also offers a clear understanding of its decision-making process, a rarity in the realm of AI.?A far cry from their bulkier neural net counterparts, these streamlined systems not only adapt on the go to brand new environments.?The researchers show how their model allows a drone to rapidly adapt to an environment it’s never been trained on.?The researchers oversell the solution a little bit because its not actual continual learning on the fly, but it does demonstrate that a small NN can be much more flexible and adaptable.

RLP: A New Dawn in Conversational AI

In a new paper, quantum theoretical physicist, Kevin Fisher outlines his approach to bringing self-awareness and reflection to AI agents.?He calls it Reflective Linguistic Programming (RLP). RLP encourages models to introspect on their own predefined personality traits, emotional responses to incoming messages, and planned strategies, enabling contextually rich, coherent, and engaging interactions.?Fisher sees applications in everything from negotiations to mental health support systems.

Med-PaLM 2: Google’s AI Prodigy in the Medical Field

Google’s latest AI tool, Med-PaLM 2, is making quite a splash in the medical field. This new kid on the block, a variant of PaLM 2, has been undergoing tests since April in various hospitals. According to Google’s research, this chatbot has been flexing its muscles, performing on par with actual doctors in several metrics. But don’t worry about your secrets getting out - the hospitals testing Med-PaLM 2 have full control over their data, which is encrypted for good measure. Google is betting big on this, believing that Med-PaLM 2 could potentially increase the beneficial use of AI in healthcare by a whopping 10-fold.

Also this week:

• That Google memo about having ‘no moat’ in AI was real — and Google’s AI boss disagrees with it
• UK to host first global summit on Artificial Intelligence
• ‘Excited’ And ‘Scary’: After Sizzling First Half, Don’t Expect AI Investment To Cool Off
• Google unveils a better Bard and new NotebookLM service
• Chip wars: How ‘chiplets’ are emerging as a core part of China’s tech strategy
• Cyberpunk 2077 Gets ChatGPT-Fueled ‘CyberAI’ Mod
• LLM Powered Autonomous Agents (by Lilian Weng)
• Dr.?ChatGPT Will Interface With You Now
• Sarah Silverman is suing OpenAI and Meta for copyright infringement
• Announcing our $58M funding to Build Generative Models and Advance AI Research - Reka AI
• Let’s Talk about LangChain (by David Hershey)
• Spikformer: When Spiking Neural Network Meets Transformer
• GKD: Generalized Knowledge Distillation for Auto-regressive Sequence Models
• Midjourney V5.2 for Macro Photorealistic Images (With “Shortened” Prompts) (by Henry Zhang)
• WormGPT: New AI Tool Allows Cybercriminals to Launch Sophisticated Cyber Attacks
• Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks