AIandBlockchain

What if you could train a cutting-edge AI model with hundreds of billions of parameters—without shelling out millions on premium GPUs or running a high-performance data center? In this episode, we crack open a groundbreaking report from Ant Group's AI team, revealing how they achieved exactly that.

We take you inside the tech behind Linglight and Ling Plus, two powerful mixture-of-experts (MoE) models that challenge the status quo. With parameter counts rivaling industry giants, these models deliver elite-level performance while operating on modest hardware setups. How? Through a revolutionary cocktail of architectural innovation, smart hardware selection, and next-gen training strategies.

From optimizing model frameworks with DL Rover and the "Eat It" elastic training technique, to scaling across low-cost devices and building robust debugging tools like XPU Timer, this episode unpacks the technical wizardry behind the scenes. We also dive deep into anomaly detection, adaptive tool learning using knowledge graphs, and evaluation tools like Flood—all designed to squeeze out maximum efficiency and stability from limited resources.

But this is more than just a technical deep dive—it's a blueprint for democratizing AI. Whether you’re a researcher, engineer, or just AI-curious, this episode lays out how massive models like Ling Plus (trained on 9 trillion tokens, no less) can be built without Wall Street-level budgets.

If you've ever felt locked out of high-end AI development, this episode is your key. Tune in and discover how the next era of large language models is being built—not just bigger, but smarter, faster, and more accessible.

Read more: https://arxiv.org/pdf/2503.05139

We unpack the fascinating results of a massive, multi-method research project analyzing over three million conversations, thousands of user surveys, and a 28-day randomized control trial. The findings? Some users are forming deep emotional bonds with AI—seeking comfort, support, even friendship. And it’s not just anecdotal: the study used a set of advanced “emo classifiers” to track expressions of loneliness, dependence, affection, and more across conversations.

We explore what draws people into these emotionally rich interactions, the surprising role of voice-based AI models, and how users’ existing emotional states shape their relationships with technology. You’ll hear how “power users” are driving a large share of emotionally expressive engagement—and why this matters for our collective future.

Whether you’re curious about how AI is affecting human psychology, worried about growing emotional dependence, or just fascinated by the idea of digital companionship, this episode offers insights you won’t want to miss. As AI becomes more lifelike and integrated into daily life, we ask the big questions: Is this the beginning of a new kind of relationship? And how can we engage with AI in ways that support, rather than replace, our human connections?

Tune in for an eye-opening conversation that just might change how you think about talking to machines.

Read more: https://www.media.mit.edu/posts/openai-mit-research-collaboration-affective-use-and-emotional-wellbeing-in-ChatGPT/

We break down the study’s fascinating findings, including how researchers used over 800 recordings of human professionals tackling tasks that ranged from quick coding fixes to 30-hour software projects. Then they put top AI models—like Claude 3.7 and others—to the test. The result? A striking discovery: AI's ability to complete longer tasks has been doubling every 7 months since 2019.

We explore what this means for fields like software development, where AI might be able to complete a month’s worth of work in a single task within just five years. But it's not just about task length—it's about complexity. The researchers introduce “messiness factors” to capture how AI handles the kind of unpredictable, real-world challenges humans deal with daily. And surprisingly, AI is improving on those, too.

Tune in as we unpack what this means for the future of work, creativity, and the role of human expertise in an AI-accelerating world. Will AI soon operate like a junior employee? Can it tackle dynamic, messy tasks with confidence? And what might this all mean for your job five years from now?

If you’re curious about what “AI progress” really looks like—and how fast it's moving—this episode is your roadmap.

Read more: https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/

AI is evolving at an unprecedented pace—but what does that mean for national security? In this episode, we dive into a groundbreaking report from Anthropic’s Frontier Red Team, which evaluates the dual-use risks of AI in cyber security and biosecurity. From AI models excelling in hacking challenges to their growing understanding of biological processes, the findings are both fascinating and alarming. Are we prepared for the rapid advancements in AI's offensive capabilities? How can we ensure responsible scaling and global safety? Join us as we break down the key takeaways, the risks, and what needs to happen next to keep AI in check.Read more: https://www.anthropic.com/news/strategic-warning-for-ai-risk-progress-and-insights-from-our-frontier-red-team

Can we truly think of the brain as a classical digital computer? On the surface, it seems like a compelling analogy, but a groundbreaking new study using information theory challenges that assumption. In this episode of The Deep Dive, we explore the hidden complexity of consciousness, the role of past experiences in shaping our present, and why the sheer amount of information required for our conscious lives might exceed the brain’s computational capacity. Could this mean our minds function in a way that goes beyond classical computation—or even beyond the brain itself? Join us as we unpack the implications of temporal holism, memory integration, and what this could mean for AI, neuroscience, and the very nature of human experience.Read more: https://arxiv.org/pdf/2503.10518v1

Can artificial intelligence really reason like a human? In this deep dive, we unravel the mysteries of how large language models (LLMs) move beyond just predicting words to making logical connections and solving problems. We break down the latest research, including insights from Kamal Kumar’s paper, exploring how techniques like fine-tuning, reinforcement learning, and scaling impact AI’s ability to think. From Sherlock Holmes-style deductions to Monte Carlo simulations, we uncover the mind-blowing ways LLMs are evolving—and the ethical questions that come with it. Tune in for a thought-provoking exploration of AI’s future!

AI is getting smarter—but is it always honest? In this deep dive, we explore groundbreaking research from Anthropic on auditing AI systems for hidden objectives. Researchers built an AI with deliberate quirks, like an obsession with camelCase in Python, to see if auditors could uncover its secret motivations. They even created a fictional academic history to test how AI picks up biases from external sources.

Join us as we unpack the clever techniques auditors used—behavioral attacks, data sleuthing, and even AI "interrogation" methods—to reveal how artificial intelligence can develop unintended priorities. What does this mean for the future of AI safety? And how can we ensure AI systems act in our best interests? Tune in to find out!

Read more: https://www.anthropic.com/research/auditing-hidden-objectives

Step into the future of automation with Figure and their groundbreaking humanoid robot factory, BOTQ. In this episode, we explore how Figure is revolutionizing manufacturing by using robots to build more robots—at an unprecedented scale of 12,000 units per year. From vertical integration and AI-driven production to the game-changing Helix system, we break down how this technology is reshaping the workforce and what it means for the future of human-robot collaboration. Are we witnessing the next industrial revolution? Tune in as we dive deep into the challenges, opportunities, and implications of this robotic revolution.

Read more: https://www.figure.ai/news/botq

Can artificial intelligence experience anxiety like humans do? A groundbreaking new study put GPT-4 to the test using psychological tools designed for people—and the results were astonishing. In this episode, we dive into how AI responded to traumatic narratives, what happened when researchers tried mindfulness techniques on it, and what all of this means for the future of AI in mental health, customer service, and beyond. Could AI be emotionally influenced? And if so, what are the ethical implications? Join us for this thought-provoking discussion that challenges everything we thought we knew about artificial intelligence.

Read more: https://www.nature.com/articles/s41746-025-01512-6

What if AI could think beyond its training data and actively search for real-time information? In this episode, we take a deep dive into the fascinating world of retrieval-augmented generation (RAG) and reinforcement learning (RL) with R1 Searcher—an innovative new approach that turns LLMs into expert researchers.

Join us as we break down how R1 Searcher overcomes the limitations of static knowledge, outperforms state-of-the-art models, and learns to retrieve accurate information in real time. From financial analysis to scientific research, the implications of this breakthrough are game-changing. Can AI finally reason, adapt, and learn like never before? Tune in to find out!

Read more: https://arxiv.org/pdf/2503.05592v1