It's all derivative

What it is tho? but can I ask why tho?

I wanted to roll my own blog, so I did. I’m using mdsvex which is a pretty nice way to have svelte in markdown… or vice versa?

That made me realize… OG HTML tags inbound.

i unapologetically use the HTML marquee element ¯\_(ツ)_/¯

Bring <marquee> back… /j. If you can’t see it, it’s just the scrolling marquee element introduced way back in the days of early Internet Explorer.

If you click that inline link you’ll see Wikipedia calls it “distracting”.

vite-svelte-plugin is mad at my usage of <marquee> but i don’t care, my HS computer lab teacher would be proud

1:42:51 PM [vite-plugin-svelte] src/routes/dispatches/the-first-dispatch/+page.md:15:0
Avoid `<marquee>` elements
https://svelte.dev/e/a11y_distracting_elements

Look I get the reasoning behind it. It’s distracting, but also not very accessible.

I promise I’ll limit my usage.

It’s all derivative

• Everything we’re building with LLMs is almost guaranteed to be derivative.
• LLMs only exist in the presence of pre-existing data. Therefore what is generated already exists in some distilled format.
• The problem is everyone using it to create what they think are ‘novel solutions’ or creations. The problem is that fundamentally they cannot be novel.
• Only the human mind can create something novel and in the tech-forward era, we’re saturated with information and data that is conditioning our mind.

It’s cognitive surrender I worry about

• Really smart people giving up their own individual thinking prowess for the low-effort wins
• The motivator is almost always economic, unless it’s ‘clout’, which is also a very real (and lame) one.
• “How can I quickly get the reward I so desire with the least amount of effort?”
• We’re being trained by the thing that we trained to optmize for the reward.
• Except our optimization is more like a reward-hack = ‘Use AI’

When will people figure it out?

• Probably around the time they grasp how LLMs actually work, or at least what a good application for an LLM is. Which is starting to seem like it won’t be any time soon.
• Hard to figure it out when the heads of industry, motivated by only one thing ($), keep selling silly concepts like consciousness or sentience, and presenting them like “well, maybe it is, maybe it isn’t”.

And now, a diagram of LLM from training to inference

A reasonable generalization.

In future dispatches, I’ll walk through each of these rounded rectangles and the transitions between.