Great article. I’ve been in the UI/UX space for years. Prior to that I studied gamedesign at uni and graduated engineer in art and tech. Principles of gamedesign always allowed me to solve ux problems in sometimes unusually delightful ways. One tiiiny example was a search function for a piece of software that was incredibly slow. The engineers couldnt get it to go any faster. I helped solve it with an old gamedesign technique, which is to add story to the loading screen. Just adding a fun animation and text on what’s happening helped decrease dropoff and increase delight. Now transitioned to more design and engineering allows me to own the full ux even more. [1]
Speculating here, but I assume researchers can make a reasonable estimate of the size of closed models based on factors like training time, training speed, and the number of tokens processed.
Also, Anthropic and OpenAI probably want to keep each other on their toes so they don’t end up on the wrong side of another Opus 4.6 / GPT-5.3-Codex situation, where one lab releases a model only for the other to drop a better one hours later.
I think first of all it’s not an obvious idea, also the marketing surplus for other providers is not as big for openai/anthropic as for xiaomi and last but not least I’m pretty sure you can withdraw methodology from here.
I’m saying who has a million dollars for me, so I can make my own model?
Cattail tubers and bark from smallish hardwood trees that they embed underwater for winter food. Hidden affect of that is the cattail is not particularly good habitat or food source for waterfowl so then you have one more item pushing down on waterfowl populations.
That was my thought too. Nothing in an ecosystem exists in isolation. Change a single variable, especially a dramatic change like going from 8% to 60% on a single variable, and there will be ripple effects throughout the local ecosystem. The Law of Unintended Consequences is one of those unchangeable, eternal laws that when you forget them, the Gods of the Copybook Headings limp up to explain it once more*.
This is surprisingly useful. As I was imagining myself behind thr wheel I felt it was more clear as to when/where the next turn is. Probably because it combines top down flat with 3d spatial vs just one or the other. This was really good. I want it.
nice idea, it's a small step but can probably make it look that much better: try to have the grid actually be a grid. So not masked text moving underneath it. But the grid can easily simulate an actual LED matrix. Just prompt your fav ai coding tool with something like 'Now make this work like an actual LED matrix, so the pixels fire when ever they are active'
The thing most people seem to be missing is that ai does the work, but still needs to be guided. When it comes to design, at our little company, our main job is to direct the style and not tell it to make something. So we are actively directing it towards specific, by us determined, styles. We did this for numerous companies [1]. So yes, Ai generated dashboards can look really amazing, but the person directing them has to spend time creative directing it.
Sublime still feels like the most natural extension of the mind (when it comes to writing code by hand). Though, I use it mainly as a 2do list and notepad these days.
[1] https://designshippers.com
reply