Lions are all for mildly overweight being better than skinny, but have erotic reasons for not being overweight.  




Some nifty shots.  With the right gear, steep snow pitches can be navigated.  Lions were curious if crampons can be the basis for a trail sandal.

It's been trending towards manually glued lugs on an excess sandal.  The key problems are high grip TPE being just unobtanium & spare sandals having a nonzero cost.  1mm sole material is now $51 which brings each sandal to over $10.  The adhesive has risen to $5 per tube.  The cord locks did get cheaper.

The lug pattern is the big question.  Lions long dreamed of a single piece.

Luna promotes this lug pattern.  To minimize the weight, they could be made from 2mm rubber & stuck on.

To minimize labor, the lugs should be identical.


New balance minimus shoes have a workable lug.  If a webbing like this with gaps & stringers joining the lugs, was printed out of TPE & glued on, it might work.  To cut them manually, they would have to be squares but they could be rotated into diamonds.

Lions have only printed 2 tires with their $82 1kg white ninjaflex order from 2023 & those tires have been nearly immortal.  The same filament from amazon.com is now $92 but matterhackers now shows it for $72.  

The new cord locks could allow a heavier sandal. 

---------------------------------------------------------------------------------------------------------------

The lion portfolio actually went up during the deepshit or whatever scare.  Why can't china create just 1 meaningful english word like deepshit, deepgpt or deepai?  Being manely in cash & in stonks which are now considered defensive seemed to be the key.  They weren't considered defensive 20 years ago. 

Following AI acts has been about as useless as following blockchains & step counters.  The AI industry, which is really an engagement business, has to constantly create new acts to stay profitable.  The language model act had 3 years on it & was out of time.  The machine vision & gen AI acts only went 1 year before they were out of time. 

The problem with language models is conveying requirements to the model.  They replaced a programming language with a natural language, but a natural language is still another programming language.  Describing a very complex  image or a program to the model requires very complicated essays not unlike any other programming language.  CAD models require sentences describing every single piece in detail, the same as the python scripts which automate CAD modeling.

Their best move might be a novel interface based on pointing & clicking instead of natural language.  Adobe seems to have the right idea.

------------------------------------------------------------------------------------------

It later came to pass that deepshit probably used a bigger model from openai to train its smaller models in a modular fashion so they could run on less hardware.  The problem with that is it still requires training the bigger model before you can train the smaller model.  There are ways of modularizing the training data of the bigger model, but the labeling requires effort.









Comments

Popular posts from this blog