Thoughts on pseudo-sentience in LLMs

Toward a High Level Future

am much of the opinion that our future will be less about how well we can do any one thing, but on how well we can get ai tooling to accomplish this for us. There is quite a lot of discourse around the topic. I don't think that makes it any less the favorable outcome. It is my recommendation that everyone should learn to think "High Level". When it comes to passion do what you what. It is in fact much better for the retention of skills to continue to use products that demand more of you. Opt for the Non-AI IDE, Write without the assitance of an ai for your first draft. Doing these thing's the hard way. Is inevitably the better way to retain skills one is passionate about. However one can not be passionate about everything, and there is a lot of menial work to be done in the modern age. So why not leverage the 100X Dev, 100X Writer, and 100X Teacher at your fingertips? All you need to do is learn to become a Architect. What I mean by this is to become a person who can create clear, efficient, proven, and data-driven plans and let your AI Agents do all the ground work. This is no where near as simple, or boring as it may sound. To become a good architect at anything one must know a great deal about that thing. You are essentialy creating a "Style Guide". A guide that consists of hundreds of hours of knowledge, wisdom, and refinement of ideas to get the best possible outcome in the simplest amount of time.

On Privacy in LLMs

On Security in LLMs

On the Integration with External Tools

Voice Interaction

TTS (Text-to-Speech)

• Leaning towards Tortoise TTS, realistic and expressive.

STT (Speech-to-Text)

• Locked down on OpenAI's Whisper

Finetuning DeepSeek for My Usecases

On Multimodality

Deep Research

Big Sister (Lifelogging Protocol)

Operator Agent

Asynchronous Coding Agent

Tool Calls

Image Generation via SDXL (Stable Diffusion XL)

Web Crawling

Experimental Features

Autonomous Asynchronous Finetuning

@article{yotam2025wisdom,
  title   = "Wisdom",
  author  = "Yotam, Kris",
  journal = "krisyotam.com",
  year    = "2025",
  month   = "Jun",
  url     = "https://krisyotam.com/notes/projects/wisdom"
}