The beauty of nature never ceases to amaze

I zoomed in as far as possible and still couldn't figure out where the owl is

Throw a party! I catered Portillo's and rainbow cone when I hit that big milestone

Are you really surprised that I enjoy the holiday spirit and will suffer a bit for it, and you will not 🤪

The trick for the hot chocolate though is to go the Wrigley one right when it opens, so I can get my hot chocolate and German sweets with no crowd

I do it every year because the hot chocolate is some of the best in the biz, but I also love the little Christmas cottages and the ornaments 🥺

OH YEAAAAHHH

One last bit - I think this distinction is why OpenAI exists as a vendor while a lot of other ML tooling is open source stuff you deploy yourself.

You just cannot easily build an equivalent to ChatGPT, so you pay for it and tweak it, but you can absolutely train your own image classifier.

For context I remember seeing a machine learning demo where someone was building a video game and then trained a model to play said game, all on an ordinary laptop. An LLM just can't be built like that, because the training data and time has to be so much larger. And I'm not an expert by any means

I'd say basically anyone can make an ML model, with data they fully own, even on a local device, while an LLM has to have a certain size that requires a ton of training data, and it's unlikely you own it all.

You can train your own language model though I think, but it wouldn't be/act like an LLM

It's a notable technical difference - we've had ML for a long time, and it can be trained on really specific data for a specific task, like this Pacman webcam concept where you train a local model to map certain images to controls

storage.googleapis.com/tfjs-example...

But have you considered... snowplow bike would be fun

I need to figure out how I can get a snow plow on the front of my Tern HSD and a salt spreader on back 😂

@bedubble.bsky.social I feel like you can pull this off