- AI with Kyle
- Posts
- AI with Kyle Daily Update 040
AI with Kyle Daily Update 040
Today in AI: Microsoft enters the AI race + Engineer steals from Musk?
Kyle Balmer
September 02, 2025
The skinny on what's happening in AI - straight from the previous live session:
Highlights
🛠️ Microsoft Finally Builds Its Own AI Models!
Microsoft launched two new models - MAI Voice One for speech generation and MAI-1-Preview for general tasks, marking their first serious attempt to move away from OpenAI dependency. These are their first foundation models - at last!
The voice model can generate a minute of audio in under a second on single GPU, whilst MAI-1-Preview sits at 15th place on LM Arena rankings.
Kyle's take: This is massive news even though the models aren't blowing doors off yet. Microsoft's been stuck in the bizarre situation of using ChatGPT underneath Copilot but somehow making it worse with additional layers of corporate guardrails.
Having their own models means real competition and investment in the space. The fact they can even compete in the top 20 with their first attempt is impressive - same way Grok managed to brute-force into 8th place in just two years.
More players means better models for everyone, and hopefully they can finally make Copilot actually useful instead of being a lobotomised version of ChatGPT. Maybe!
Source: Mashable coverage and LMArena
⚔️ Musk Sues Ex-Engineer for Stealing Grok Secrets
xAI filed a lawsuit against former engineer Xuechen Li for allegedly stealing trade secrets about Grok and taking them to OpenAI. Lee sold $7 million in XAI stock in July and started work at OpenAI this month. The complaint claims he stole "confidential information related to cutting-edge AI technologies with features superior to those offered by ChatGPT."
Kyle's take: This is messy. Of all the places to (maybe!) take stolen secrets, OpenAI is the worst choice given Musk and Altman already hate each other. There are now multiple lawsuits flying between these companies - Musk suing OpenAI for deviating from original mission, OpenAI countersuing for harassment, plus Musk's new lawsuit against OpenAI and Apple for allegedly monopolising AI chatbots on Apple devices.
Two billionaires having a legal street fight whilst the rest of us just want better AI tools. The talent war is getting increasingly bitter, and this won't be the last case we see, especially when the potential payout of espionage like this is so high.
Source: Reuters
🏗️ Kyle's Building Challenge: 30 Days, 1 Hour Per Day
Fresh back from visiting Lovable in Stockholm, I’m launching a public building challenge - creating a complete application in 30 days using just one hour per day.
First project: an AI tool that generates custom lead magnets for any piece of content, inspired by an entrepreneur in one of my Whatsapp groups who doubled his business by creating tailored lead magnets instead of generic ones.
I met a bunch of amazing builders when I visited Lovable last week and its inspired me to build myself and help others.
Here’s where I’ve got to after 45 minutes yesterday:
I want to show that vibe coding isn't just for people with unlimited time. Most of you have jobs and families - spending 8 hours a day building isn't replicable.
So I'm limiting myself to one hour daily, taking something from idea to launch in 30 hours total.
Here’s day 1!
Member Question from Sheikh: "What do you say about GPT-5, especially for brainstorming SaaS ideas?"
Kyle's response: For brainstorming specifically, I tend to prefer Claude - that's subjective though, so try different models yourself. I find Claude goes into more detail and communicates better for how I like to brainstorm. That said, ChatGPT has better voice modes, so if you prefer talking to get ideas out of your head, GPT-5 might work better.
The key thing with any model is being careful when brainstorming because they tend to gas you up and glaze you. If you say "here's my business idea, what do you think?" they'll just say it's brilliant.
Better approach: give it 4-5 different ideas and ask it to help you create metrics for deciding which to proceed with - speed to deploy, funding needed, revenue potential, etc. Then work collaboratively on comparative analysis. Forces the AI to actually help you think rather than just telling you what you want to hear.
Want the full unfiltered discussion? Join me tomorrow for the daily AI news live stream where we dig into the stories and you can ask questions directly.