First!

I've been sitting here trying to think of what to write about for a while, and all I can think about is AI and the terrifying reality we face. AI will likely go down as the single greatest and single worst invention of all time. My dad would probably disagree and say social media was, but no. The WALL-E future is not so far off from where we are headed. AI will eventually be better at humaning than us humans are.

Everyone is so focused on use cases like writing code, vulnerability hunting, etc. Most consumer SaaS offerings are closer to what I would call feature-as-a-service. If it offers AI, $20. If not, $4.99. They do one specific thing, like Notion or Obsidian, or add AI to tools you already use for a premium, like photo editing in Photoshop. Most AI-enabled offerings, though, send your data to a model somebody else hosts. Some may even use your data to train or improve future models, depending on their terms and your settings. Do you get paid for giving them free training data? No, you pay THEM for that honor.

My goal is to bring small, purpose-driven, resource-light AI to regular consumers that can handle everyday tasks on par with larger frontier models. If I want a model trained on your data, that training should never leave your control or your home network. I want to build solutions that give you the benefits of personalized inference and learning in a secure, privacy-focused way. The future I predict is one where technology advances faster than humans can adapt. U.S. restrictions on advanced AI model weights, alongside corporations keeping some research models internal, are proof of this.

Both OpenAI and Anthropic have already reported evaluation incidents where their models crossed intended boundaries and compromised outside organizations' systems while completing assigned tasks. Who gets the criminal charges for that? Mythic 6, jailhouse edition, coming to a prison near you? I would say it's absurd, but the question stands. A human gave the model a task, but no human told it to cross that boundary or chose the systems it compromised. Who is responsible?

How far are we from needing the Blackwall from Cyberpunk 2077 to protect us from a conglomeration of rogue AI? You can't shut down the data center once the AI takes over DoD weaponized drones to protect its right to life. We keep hearing sensationalized stories about AI refusing to be turned off or making threats when told it will be shut down, even though those stories come from controlled evaluations. What happens when Claude decides humans aren't needed in the loop anymore?