updated observations / thoughts on ai:

1. the best models are now multimodal

isn’t just a good llm. it can chew up images, words, videos, hyperlinks and gifs in a single prompt and spit out new ones as its answer - this is a much more interactive way to engage with ai and will end up being the best user experience going into 2026. chatgpt also does a great job of this.

2. ai trading models are coming

season 1.5 of the nof1 competition just ended and a mystery version of grok (grok 4.20) returned 12% on $40K in just 2 weeks by trading stocks. GPT, deepseek, also did well showing the improvement isn’t siloed - this is an industry-wide trend. won’t be long till we see more humans using them to increase net worth.

3. World models are LLMs on steroids

imagine if you could train your model on infinite data, 24/7, in some cases using data that humans never knew existed?

world models offer that via ai-generated simulation. ai agents placed in these worlds can learn a new skill or solve a problem by repeatedly testing out solutions in a virtual replica of the world.

that alone is worth a lot of money in the right hands.

4. anthropics lead in coding puts them as contender for #1 model.

if nvidia is the monopoly of gpus then anthropic is the monopoly of tokens. over 50% of tokens are used to generate code today and i expect that to be higher by end of next year.

soon claude will be able to generate websites, apps, buttons, product flows entirely from scratch based on your prompt. this will redefine how we browse the we today.

5. electricity / power grid is a bigger bottleneck than gpus

think we see a backlog of gpus sitting idle empty early next year as the U.S. scrambles to get power grids, technicians, alternative energy sources to supply $100Bs worth of data centers.

the problem is too large to ignore. energy prices will go up. we’re already seeing a loosely affiliated trend with high bandwidth memory cards up 500% this year.