I’ll share which AI tools I’m exploring, some experiments I’m conducting, and insightful information about what I’m observing in the world every week.

🔧 Three Tools I’m Testing

📈 Plus AI - I’m using this to help ideate on Powerpoint slides based on longer form writing and information that I have. It created some excellent slides around a strategy document I created, and then I took those and refined them. Saved hours and hours on one presentation alone. I haven’t tried to teach it our company template yet, fwiw.

🤖 Martin - Self proclaimed J.A.R.V.I.S for yourself. I wouldn’t go that far, but I’m using it for multiple standard calendar and daily workflows. The most interesting thing right now is having it create a research profile on people in business meetings ahead of time and then email me the information (might move this to Slack).

💻 Vercel V0 - My software AI tool of the week. I’m testing it compared to Replit because I like the idea of being one level lower than Replit (it’s very close). I think if you’re used to Vercel, it is the right tool. However, for me, I’m going to stick with Replit.

🧪 AI Experiment of The Week

I’ve been slightly obsessed with the idea of conversing my children’s voices through a custom model trained on their speech. Mainly for nostalgic purposes for my wife and me down the road. But, also to be able to make my kids say they actually like Dad more than Mom and love eating their vegetables.

My goal is to build these models using open source models to own and retain them personally. However, I used Resemble.AI for this experiment, which allows you to create clones of voices through Voice Design, Rapid, and Professional training. I tried Voice Design and Rapid with my son. The Voice Design just takes a little bit of reading of content (which is a bit of a chore with my six-year-old) and modifies one of their models - not great results. The Rapid model took multiple reading sessions and accepts up to twenty recordings. The Professional requires a minimum of fifteen minutes of training data (readings / audio files).

Once trained, you can use it for text-to-speech or speech-to-speech generation. I just played around with text-to-speech. I give the current model for my son about an 85% accuracy, and if I get him to sit down for more training, I bet it gets really close. I am paying for this at $29 a month. However, I’m going to play around building local models for them using Mask GCT (more to come on that).

📰 Article of The Week

This is a comprehensive exploration of the fundamental mechanics behind Large Language Models.

The article breaks down complex AI concepts into understandable components, explaining how LLMs process and generate text through computational patterns rather than traditional understanding. It demystifies the technology while acknowledging both its capabilities and limitations.

As AI tools become increasingly integrated into our daily work and lives, understanding their core mechanics isn't just academic curiosity—it's practical necessity. Whether you're a developer, business leader, or curious user, grasping these fundamentals helps you make better decisions about when and how to use AI tools effectively. Wolfram's explanation provides that crucial foundation without requiring advanced technical knowledge.

🌎 Where the World is Going

Even as open-source challengers like DeepSeek demonstrate impressive capabilities, tech giants aren't slowing their AI infrastructure arms race – they're doubling down with a staggering $300 billion investment in new data centers. But while the headlines focus on the eye-popping numbers, I'm seeing a more nuanced story unfold. It's like the early days of cloud computing – everyone focused on who was building the biggest data centers, while the real revolution was happening in how we used them.

What fascinates me is how these massive investments will reshape the AI landscape. The common narrative suggests these data centers are primarily for training the next GPT or Gemini. However, my experience working with these models suggests we're underestimating the computational demands of inference and reasoning – the everyday work of AI systems thinking through problems and generating responses. It's like everyone's focused on the $30 million per episode cost of making Stranger Things while overlooking the massive infrastructure needed to stream it to millions of households every day.

The silver lining for innovators and builders is clear: as these tech giants compete for dominance, AI capabilities are becoming commoditized at an accelerating pace. We're already seeing this with Gemini's pricing structure undercutting previous benchmarks. This commoditization isn't just about cheaper access – it's about democratizing AI innovation. The real opportunity isn't in building bigger models or more data centers but in leveraging these increasingly affordable AI capabilities to solve real-world problems in novel ways. The future belongs not to those who own the most computing but to those who can turn these commoditized AI resources into meaningful solutions.

👨‍💻 About Me

🔗 Extra Links to Check Out

• Jensen Huang’s Conversion with Cleo Abram 📺

• Tally Form Builder 📄

• Unitree Robots 🚷

• Apple Event Invite App 🍏