Just tried "generate an SVG of a pelican riding a bicycle" for Claude Opus 4.8 Max and of course both legs on same side ... the smartest publicly available model by Anthropic (after Fable) doesn't even successfully simulate understanding the concept of a bicycle.

▲

mountainriver 3 days ago | parent [-]

Yet it can write code better than 99% of humans…

It’s just starting to be trained on svgs, which is a really hard problem

▲

raffael_de 3 days ago | parent [-]

"99% of humans" is a low bar. Maybe you mean "99% of people who earn money by developing software"?

▲

WarmWash 3 days ago | parent [-]

LLMs can't really "see", so I challenge you to draw a pelican on a bike without any visual feedback, just code. Because that is how they are doing it.

Vision tokens for transformers aren't really well solved yet, which is why they can smash a phd math problem and trip over a "count the cats on the chair" problem.

▲

raffael_de 2 days ago | parent [-]

It's not about seeing. It's about identifying the legs of the Pelican and then transferring the concept and mechanics of riding a bicycle + geometry of a body and a bicycle. The entire task has also nothing to do with vision tokens.

▲

mountainriver 21 hours ago | parent | next [-]

If we want to train a model excessively on SVGs it will obviously be able to do this. We have only just started trying to do that

▲

WarmWash 2 days ago | parent | prev [-]

> It's about identifying the legs

So, seeing?

▲

raffael_de 2 days ago | parent [-]

seeing isn't necessary to understand what a leg is.

	▲	WarmWash a day ago \| parent [-]
		Which I why humans can draw so well with their eyes closed?