In this engaging episode of "AI with Shaily," host Shailendra Kumar, affectionately known as Shaily, dives into the fascinating realm of artificial intelligence and its impact on our daily lives. 🎙️✨ The focus of this discussion is on the advancements in fine-tuning large language models (LLMs), a topic that is gaining traction in the AI community, even if it hasn't gone fully mainstream yet.

Shaily kicks things off by painting a vivid picture of the possibilities that come with fine-tuning LLMs. He humorously imagines an AI that can converse in Shakespearean English while also delivering jokes in Klingon, showcasing the impressive capabilities that fine-tuning can unlock. Techniques like QLoRA and Spectrum are highlighted as game-changers in this field. QLoRA utilizes 4-bit quantization to enhance efficiency, while Spectrum smartly focuses on the most informative layers of the model, making the process of fine-tuning more precise and effective. Shaily cleverly poses the question, "Why use a sledgehammer when a scalpel will do?" to emphasize the importance of targeted approaches in AI development. 🔍⚙️

The conversation then shifts to the emergence of synthetic data, which Shaily likens to having a virtual assistant that can quickly generate high-quality datasets. This innovation is made possible through techniques like Generative Adversarial Networks (GANs) and model distillation, which streamline the often tedious process of data annotation. 🖥️📊

Shaily also discusses the trend of smaller, more efficient models, comparing them to smartwatches—compact yet powerful. He references DeepSeek's innovative use of chain-of-thought datasets to train these nimble models, suggesting that soon we might see LLMs that can handle everyday tasks without the need for extensive computational resources. ⏱️💡

The topic of open-source platforms comes up next, with Shaily mentioning InstructLab.ai as a prime example of how collaborative efforts and synthetic data are making advanced AI more accessible. He emphasizes how these developments are bringing AI innovations closer to everyday conversations and interactions. 🌐🤝

Reflecting on his personal journey in AI, Shaily shares a nostalgic anecdote about the challenges of fine-tuning in the past, likening it to searching for a needle in a haystack. However, with the advancements being made, he feels that it's now more like having a map to that needle, highlighting the progress in the field. 🧭✨ As a bonus tip, he encourages listeners to embrace synthetic data for their LLM projects, as it can significantly reduce costs and complexity while maintaining flexibility.

To wrap up the episode, Shaily shares a thought-provoking quote from AI expert Fred Brooks: "Good judgment comes from experience, and experience comes from bad judgment." This wisdom resonates particularly well in the context of adapting AI models. 

Shaily invites listeners to engage further by following him on various social media platforms like YouTube, Twitter, LinkedIn, and Medium for more insights. He encourages everyone to subscribe and share their thoughts in the comments, leaving the audience with a sense of curiosity and anticipation for the next episode. This is Shailendra Kumar signing off, reminding everyone to stay curious and tuned in! 🌟👋

AI Unleashed: Fine-Tuning the Future of Language Models

Discover NNetNav: Stanford's Groundbreaking Open-Source AI Agent

In this video, I explore NNetNav, Stanford University's innovative AI agent, inspired by how children learn through exploration. Unlike traditional Large Language Models (LLMs) relying on static data, NNetNav independently generates its own training data by interacting directly with websites—clicking buttons, filling forms, and navigating intuitively.


✅ Key Topics Covered:

1. What is NNetNav?

Developed by Stanford researchers including Shikhar Murty and Professor Chris Manning.

Learns unsupervised through exploratory, childlike interactions instead of static datasets.

2. Interactive Learning Paradigm

Explanation of synthetic training data generation via real-time web interaction.

Mirroring human cognitive development for more resilient AI.

3. Efficiency: Fewer Parameters, Greater Performance

Achieves comparable or superior outcomes relative to models like GPT-4 and Anthropic.

Operates using significantly fewer parameters (~one-third fewer), maximizing efficiency.

4. Privacy and Ethical Considerations

Advantages of open-source transparency over proprietary alternatives.

Addressing ethical challenges and advocating regulatory frameworks for responsible AI deployment.

5. Real-World Applications

Potential uses: Flight booking automation, data extraction, complex reporting, and web navigation.

Transformative implications across industries—education, software development, customer support, and more.

6. Impact on Education & Human-AI Collaboration

Influence on educational strategies emphasizing practical, interactive learning.

Promoting deeper human-AI partnerships through embodied learning, fostering intuitive and aligned interactions.

7. Future Outlook

Exploring reinforcement learning and "learning on-the-fly" to enhance generalization capabilities.

Predictions for future developments and innovations inspired by this approach.


🌐 Explore Further:

Original Stanford Article: https://hai.stanford.edu/news/an-open-source-ai-agent-for-doing-tasks-on-the-web

Join the conversation and follow my insights on:

X (Twitter): https://x.com/_Michael_Lorenz/status/1905876382258716685

Bluesky: https://bsky.app/profile/ml-the-ai-engineer.bsky.social/post/3llir4vvfkc26

Threads: https://www.threads.net/@michael_lorenz_/post/DHxhf_XM572?xmt=AQGzUqnqKS0eiPnEZBRe0ROitQMOv8fw1ecsvtfqJBBmag


🔗 More from me on other platforms:

TikTok: https://www.tiktok.com/@michael.lorenz.ai/video/7487229020501708054

Spotify: https://creators.spotify.com/pod/show/michaellorenz/episodes/Stanfords-NNetNav-How-Childlike-Learning-Could-Revolutionize-Open-Source-AI-e30qu4q

Medium: https://medium.com/@michael-lorenz/can-ai-learn-like-a-child-stanford-thinks-so-6314deb07ab7

Substack: https://michaellorenz.substack.com/publish/post/160134977

YouTube: https://www.youtube.com/watch?v=myRi6OmFk9U&t=2s

🔗 Connect with Me:

I'm Michael Lorenz, passionate about bridging technology, sustainability, and human-centric innovation. Subscribe and join me as we explore the fascinating intersection of AI, learning, and societal transformation.

Stanfords NNetNav How Childlike Learning Could Revolutionize Open-Source AI

Welcome to "AI with Shaily"! 🎉 I'm Shailendra Kumar, your enthusiastic host, guiding you through the fascinating world of artificial intelligence. Today, we're diving into some exciting developments, particularly focusing on Joelle Pineau, a key figure in Meta's AI advancements. After eight remarkable years, Joelle has announced her departure from Meta, which has stirred quite a buzz in the tech community. 🚀

As Joelle hangs up her coding hat at the end of May, many are left pondering the future of Meta's AI division. With a staggering $65 billion allocated for AI initiatives, Meta is not slowing down anytime soon. Even without naming a successor, the company is pushing forward, particularly with projects like their open-source Llama model, which is a significant step in democratizing AI technology. 🌍

Joelle's contributions to the AI landscape are monumental. She has played a pivotal role in developing influential models such as PyTorch and Llama, which have opened the doors for developers worldwide to access and innovate in AI. It's akin to handing over a treasure chest of AI possibilities to the global developer community, sparking excitement and creativity. 💡

This leads us to an intriguing question: Can these successful open-source models lay the groundwork for future AI innovations? The potential is enormous, especially as AI systems evolve to achieve human-like reasoning and flawless voice translation. 🤔

On a personal note, I recognize the significant impact of Joelle's departure. As someone deeply invested in the AI field, I see the vital connection between groundbreaking research and its practical applications. Meta’s strategy to seamlessly integrate AI into its products is crucial for fostering innovation and keeping pace with technological advancements. It's a delicate balance that can lead to exciting breakthroughs. ⚖️

Before we conclude, here's a cheeky Bonus Tip for aspiring AI enthusiasts: consider getting involved in open-source projects! It's like a free training camp with limitless opportunities for creativity and learning. 🌟

Remember this nugget of wisdom as we journey through the AI landscape: "The future of AI isn't about replacing humans; it's about enhancing capabilities and creating new experiences." Keep this in mind as we explore the ever-evolving saga of AI! 

Thank you for joining me on "AI with Shaily"! For more insights and updates, connect with me on YouTube, Twitter, LinkedIn, and Medium. Don’t forget to subscribe for the latest AI news, and I encourage you to share your thoughts and comments. Until next time, keep questioning, keep exploring, and stay curious! 🌈

AI Unplugged: Joelle Pineau's Departure and the Future of Meta's AI Revolution

Welcome to "AI with Shaily," where your host, Shailendra Kumar, takes you on an exciting journey into the world of artificial intelligence, robotics, and materials science! 🚀 In this episode, we focus on an innovative 3D-printed robot named FLUID, which is set to change the landscape of material synthesis around the globe. 🌍

Imagine being a young researcher bursting with ideas but facing financial hurdles. That's where FLUID comes to the rescue! Developed by a dedicated team led by Professor Keisuke Takahashi at Hokkaido University, FLUID is designed to make complex material creation accessible and affordable for everyone. Think of it as the Robin Hood of research tools, leveraging open-source technology and 3D printing to bring high-tech solutions to labs of all budgets. 💡

FLUID is constructed from off-the-shelf electronics, making it a cost-effective alternative to expensive machinery. Its modular design is reminiscent of assembling your favorite IKEA furniture, offering both flexibility and precision. Whether you're synthesizing materials or following a recipe, FLUID simplifies intricate processes, allowing researchers to focus on innovation rather than equipment limitations. 🛠️

One of the standout features of FLUID is its real-time software integration. Imagine having a virtual assistant like Siri guiding you through your research, making the process intuitive and user-friendly. This is a prime example of how technology can democratize scientific research, making it more inclusive and accessible worldwide. 🌐

Shaily also shares a valuable tip for researchers looking to enhance their lab capabilities: engage with open-source communities! These platforms are treasure troves of resources, support, and innovations like FLUID that can significantly transform your research experience. 🔍

As the segment wraps up, Shaily leaves us with a thought-provoking quote: "Innovation is the ability to see change as an opportunity – not a threat." This embodies the spirit of FLUID and encourages researchers to embrace opportunities for growth and advancement. 

Don’t forget to connect with Shaily on social media platforms like YouTube, Twitter, LinkedIn, and Medium to keep the conversation alive! 📱 He invites you to subscribe for more enlightening insights into AI and share your thoughts or innovative experiences in the comments. Together, let's break barriers and spread knowledge! Until next time, Shaily wishes you a curious mind and an inspired heart. 💖

Revolutionizing Research: How FLUID is Democratizing Material Synthesis

https://www.youtube.com/watch?v=aNmEDNN4Vdc&list=PLNP4Nm89WfbWQbQFEAMZsiSVyH9AylCN9&index=26
too much images just like information/news

Meta-Mashups & Tools of Recognition - Single

Meta-Mashups & Tools of Recognition

🎵 Cover of "Oxygene, Pt. 10" by Jean Michel Jarre  
🎛️ Created using LMMS 1.2.2 on Ubuntu 24.10  

This is a cover of the iconic "Oxygene, Pt. 10" by Jean Michel Jarre. I tried to stay as close to the original as possible, recreating its atmospheric synth textures using only free tools on Linux.

🧰 Plugins & Effects Used

🔊 Instruments:
- ZynAddSubFX (multiple instances)
- tripleoscillator
- bitinvader
- organic

🎚️ LADSPA Effects:
- autowah
- Eq, Eq2x2, Eq10X2
- invada_hp_stereo_filter_module_0_1
- PingPongPan
- Plate2x2
- VintageDelay

All created and mixed natively on Linux with open-source tools.


Oxygene, Pt. 10 (Jean Michel Jarre Cover)

Hey there! 👋 You’re tuned into AI with Shaily, hosted by Shailendra Kumar, your friendly guide to the latest and coolest AI news from around the world. Today’s spotlight is on an exciting breakthrough that’s perfect for both LEGO fans and AI lovers—introducing LegoGPT! 🧱🤖

LegoGPT is a groundbreaking AI developed by Carnegie Mellon University that can turn simple text prompts into stable LEGO building plans. Imagine telling the AI, “Build me a castle with a turret and moat,” and it responds with a detailed blueprint that’s been tested through physics simulations to make sure it won’t collapse. No more wobbly towers or failed builds! 🏰💪

The magic behind LegoGPT comes from combining a powerful language model from Meta called LLaMA-3.2-1B-Instruct with a massive dataset named StableText2Lego. This dataset includes over 47,000 LEGO models representing nearly 28,000 unique 3D objects, giving the AI a rich understanding of how LEGO bricks fit and work together. 📚🧩

One of the coolest features is its rollback mechanism—a kind of safety net that lets the AI “rewind” and try again if it detects instability in the design. Thanks to this clever trick, the success rate of creating stable LEGO structures skyrocketed from just 24% to an astonishing 99%! It’s like having a perfectionist builder who won’t stop until everything is rock solid. 🔄✅

As a LEGO enthusiast myself, I know the joy and frustration of building and rebuilding sets. So, an AI that nails it nearly 99% of the time? That’s seriously impressive! It opens up exciting possibilities not just for play, but also for education, prototyping, and creative design in the future. 🎓🚀

Even better, LegoGPT is open-source and available on GitHub, inviting developers and hobbyists to explore, experiment, and push the limits of what’s possible. There’s also buzz about expanding its capabilities to create LEGO plans from images—imagine turning your favorite photos into custom LEGO masterpieces! 📸➡️🧱 How awesome is that for personalizing your space?

Here’s a quick tip if you want to try LegoGPT or similar AI tools: start with simple prompts and gradually add complexity. This approach helps the AI produce cleaner, more stable designs and saves you from a lot of trial-and-error. 📝✨

To close, I’ll share a wise quote from Alan Kay: “The best way to predict the future is to invent it.” LegoGPT perfectly embodies this spirit, using AI to invent the future of creative building. 🛠️🌟

Don’t forget to follow me on YouTube, Twitter, LinkedIn, and Medium for more awesome AI updates. Subscribe to AI with Shaily, and if you have ideas or experiences with AI and creative tools like LegoGPT, drop a comment—I love hearing your thoughts! 💬❤️

Until next time, keep building your AI knowledge one brick at a time! 🧱🤓

LegoGPT: The AI Revolution Building Stable LEGO Creations from Text Prompts

Tamraght is a small town for surfers on the Moroccan coast. I hope this piece of music reflects the relaxed and laid-back atmosphere I experienced there. It is my first attempt at producing electronic music.

I created this track exclusively with free and open source tools such as Ardour, Geonkick, Odin2, LSP or the CALF and x42 plugins.

This track is made available under the CC0 (Creative Commons Zero) licence.

Free download at https://matisse.unwo.space .

Tamraght

Welcome to AI with Shaily! 👋 I’m Shailendra Kumar, your friendly guide through the ever-evolving world of artificial intelligence. Today, we’re exploring a hot and timely topic that’s buzzing across developer forums, social media, and AI communities worldwide — Microsoft’s decision to retire the Bing Search API and what this means for the future of generative AI search innovation. 🔍🤖

Imagine you’re building a smart AI assistant, your personal helper that fetches information from the web in real time. Until now, many developers relied on Bing’s Search API to get raw search results — links, snippets, and detailed data — giving them the power to create highly customized, data-accurate responses. But Microsoft is changing the game by retiring this API and encouraging users to switch to Azure AI Agents, which use “Grounding with Bing Search.” Instead of raw data, these agents provide AI-generated summaries that are integrated with real-time web information. This means cleaner, more concise answers, but less access to the granular details behind the scenes. ⚙️✨

This change has sparked lively debates online. Is this a bold step forward, allowing AI to present smarter and more user-friendly summaries? Or is it a setback, limiting access to detailed data and locking innovation inside Microsoft’s ecosystem? I personally recall working on projects that needed precise search data for analytics — losing raw access felt like losing the sharpest tool in my toolbox. Yet, on the flip side, AI-curated insights can save time and reduce information overload for many users. ⏳🛠️

Developers are getting creative in response. There’s a growing buzz about building DIY search engines and leveraging open-source indexes to avoid vendor lock-in. Tutorials and code snippets are flying across Twitter and GitHub as innovators strive to keep search transparent and customizable. Meanwhile, entrepreneurs on LinkedIn are debating how this shift impacts startups that rely on Bing’s data feed. 💡🚀

Quality and trust are front and center in this discussion. When AI summarizes the web, how do we prevent bias from creeping in unnoticed? How can we verify the sources behind those neat, AI-generated paragraphs? These critical questions dominate conversations on X (formerly Twitter) and LinkedIn, fueling vibrant and thoughtful debates. ⚖️🔍

Here’s a bonus tip for AI developers and enthusiasts: if you’re building generative AI applications that depend on search data, consider blending multiple sources. Combining AI summaries with selective raw data or trusted APIs can strike a balance between convenience, transparency, and control. This hybrid approach might just be the best of both worlds! 🧩🤝

So, what do you think? Is Microsoft’s new approach pushing AI search innovation into exciting, uncharted territory — or is it clipping the wings of open, granular data exploration? 🤔🌐

As the legendary AI pioneer Alan Turing once said, “We can only see a short distance ahead, but we can see plenty there that needs to be done.” That vision inspires us to keep pushing forward, questioning, and innovating. 💭🚀

Don’t forget to follow me, Shailendra Kumar, on YouTube, Twitter, LinkedIn, and Medium for more insights into the AI universe. If you enjoyed this episode, hit subscribe and share your thoughts below — let’s keep this conversation alive. After all, AI isn’t just technology; it’s our shared journey into the future. 🌟🤖

Thanks for tuning in to AI with Shaily — until next time, keep experimenting and stay curious! 🔬✨

Microsoft Retires Bing Search API: Revolution or Restriction for AI Search Innovation?

https://www.youtube.com/watch?v=toBDlHUd_wc&list=PLNP4Nm89WfbWQbQFEAMZsiSVyH9AylCN9&index=16
s/o 2_my A$sociaTes

Trial & Error: Death of Playlist

Unlock the secrets of data mastery with Shailendra, a 27-year veteran in AI, ML, and Data Science, driving AU$ 6 Billion in transformative value. As the author of 'Making Money out of Data,' their strategies redefine digital success. With expertise honed at IBM and Accenture, they drive innovation and data monetization. Explore how loyalty programs like Flybuys and Everyday Rewards revolutionize business—connect with Shaily today.

𝐰𝐎𝐫𝐋𝐝 𝐂𝐥𝐀𝐬$ 𝙥𝙃𝙞𝙇𝙖𝙉𝙩𝙍𝙤𝙋𝙞$𝙩

Download my stuff at https://matisse.unwo.space .

🎙️ From Sound to Systems: The Journey from Music to Artificial Intelligence

Music has always played a central role in my life. I started playing guitar at 19—not to become a musician, but to connect with creativity in a hands-on way. That same curiosity now drives my work in artificial intelligence, engineering, and technological innovation.

What began as a private SoundCloud account for listening and inspiration is now transforming into a platform for something new:
The launch of my AI & Technology Podcast.

🎧 New Podcast on Artificial Intelligence, Engineering & the Future of Work

As an AI engineer and systems thinker, I explore the deep transformation underway at the intersection of:

Artificial Intelligence & Machine Learning

Digital Engineering & Cloud Technologies

Sustainable Innovation & Future Skills

Neuroscience, Creativity & Human Potential

The podcast—available here on SoundCloud and on Spotify—features thought leadership, expert interviews, and practical insights for engineers, technologists, and future-minded thinkers.

Whether you're interested in:

Real-world AI use cases,

How to thrive in a world shaped by automation,

Interdisciplinary insights from engineering, psychology, and design.

Or building a future where tech serves people, not the other way around –
this podcast is designed for you.

🌐 Who am I?

AI Engineer | Cloud Architect | Battery Tech Innovator

Deep interest in cognitive science, systems thinking, and transformational technologies.

Passionate about bridging human values and technological progress.

New tracks tagged #open-source