world2vec

At dayjob we're also using self-hosted GLM-5.3 via our flavour of pi.dev but teams still have discretionary power over using Claude Code or not (there's a team's budget to spend on whatever the team decides). Some teams like mine still like to use Claude Code for a lot of the work but most stuff doesn't require the latest and greatest LLM and can be done with "lesser" LLMS by simply throwing a lot of tokens and compute to it.

Carnegie Mellon has (had?) a robot called CHIMP that worked in similar fashion.

Why not an arachnoid form factor with wheels too? Like the tachikomas[0] from Ghost in the Shell world. It usually drives with its wheels but when needed it can use its legs like a spider/insect would. Much more efficient and versatile (and stable) than bipedal form factor IMHO.

[0] https://ghostintheshell.fandom.com/wiki/Tachikoma

I feel you. I keep daydreaming of getting a small piece of good land and just grow oranges and lemons.

I grew up poor and it doesn't bother me at all but I still need funds to buy land, build a roof over my head, etc. So the plan right now is to save money as best as I can.

Same. I lived vicariously through my wife when I started showing all his movies (she had only seen Spirited Away) and can't wait to have a kid and watch everything (multiple times) all over again.

Way offtopic but I like the light/dark mode of that website with several very sensible choices.

In the novel "The Mountain in the Sea" (great read, very pertinent as well for the discussion of AGI and what is consciousness), an "abglanz" is like a digital identity shield that hides a person's face behind a shifting weird pattern of colours. I wonder how far off we are from everyone just retreating into higher privacy settings and opt out.

Yeah that's what I already do. Fable writes the plan and checks things at certain milestones. Otherwise it does get lost indeed.

It's not my main model (that would be Fable 5.1 Extra) but it's been doing agent-driven search and optimisation of a cross-trading ranking model (it's for work).

1 billion tokens a day?!! I've done a lot of work these past 2 weeks with GLM-5.3. Like, a lot. And I've just passed 300 million tokens in total.

Can I ask where are you using all those tokens?

People wear wedding rings but take them off for pull up bars and so on. But a fitness tracking device you need to take off to... Do some fitness exercises... Is not a smart idea in my book.

An "expanded" Schengen Area that crosses the Atlantic and Pacific oceans and brings Canada, Australia (and Japan, maybe?) into the fold would be an incredible thing. I'm all for it.

I do recommend a decent pair of binoculars and the Merlin Bird ID app.

Same in Portugal: a "tarifa" can mean a price charged for a rendered service or a tax on imported products.

Same. Probably a few if-then statements would be enough to beat me :'(

"While unlikely, we cannot rule out that de-identified data derived from their usage of our products helped improve our models."

There you go, the suspicion of the "concurrent work" (https://cims.nyu.edu/%7Etristanb/statement.pdf) mathematicians might not be that unfounded after all...

"Pax Sinica", Chinese Peace

Buckmaster is actually implying that OpenAI spied on his chat logs and tried to speedrun his work and then tried to remove his co-author because he's an Anthropic employee?

Claude Code seems totally fine for me.

There's no way SQL is more unreadable than polars. IMO it's the other way around.

Good. Everywhere in the cities the speed limit in residential areas should be 30km/h and 60km/h for everything else. No need for higher speeds in dense urban environments.

Seems some of them would get the key contact details "clipped" by the rolodex itself. Maybe a good design would put the phone number and address at the top.

The labs are clearly training for the benchmark.

I mean no offense but these pelicans are a bit tiresome and a very meaningless benchmark. There's no real difference between any of these svgs across models and model versions anymore.

Love that documentary.

My first ever "e2e" ML project during early covid quarantine was a crude vision model to detect pigeons on my balcony (and pigeons only) and play a loud random sound. It almost worked sometimes.

1 hour according to the spec sheet. Would be nice if it knew battery was running out and "sat" on a wireless charger by itself.