The jumps are pretty impressive, this thing has some power. I'd be very curious how fast you could get this dog with some reinforcement learning for a proper transverse gallop gait - and if it converges towards a gallop naturally, or if it discovers some other fast gait patterns during learning.
Depending on the max speed of the motors/legs, giving it longer foot pads might be necessary for a good gallop. Intuitively, it looks a bit... "low gear" in the videos.
I was waiting for Google to pull a local LLM onto Chrome/Android devices. It opens up some revenue streams that weren't easily possible before: for example the often memed "I was talking about cigars with my wife one single time and now all I see are adsense ads for cigars" gets much easier with a local model doing speech to text and topic classification.
No, but mostly for economic reasons. You can farm a whole lot of fish in aquaculture - it's just more expensive than importing wild caught fish.
The numbers look pretty insane, you can raise many tons of fish in relatively small volumes of water (several hundred kg of fish per year per cubic meter). You just gotta build the ponds/tanks/cages, and the infrastructure to filter the water, supply the oxygen and deliver the feed.
There are also 4X as many people in China, little domestically available oil, and their government supports domestic manufacturing. This is an expected result.
It’s OK to celebrate small wins. The US doesn’t have to be #1 in everything. We also seem to have a curious diseconomy of scale on mega infrastructure projects for complex reasons, so maybe slow growth is the right approach.
Yep, actively suppressing renewable efforts all the way down to shaming on a cultural level. It should be a net positive for Americans to adopt renewables - cheaper energy, more independence, good for the environment - but instead its viewed as silly or too unreliable when it isn't.
I assume the "briefly gets worse" is when a buch of hyperscalers do a complete write-off of their entire AI investments, bankrupting several of them (which, in turn, bankrupts several large banks and most current venture capital firms)?
Cumulative AI capex will hit $2T this year. Cumulative opex is on the same order. Unless the models get real good (as in: can fully replace many engineers) right quick, nobody is even going to see interest getting paid on those investments. The only alternative is model access costing 5 figures per (replaced) seat.
But yes, once GPU racks can be had at auction for pennies on the dollar, inference of open source models might be an... OK low margin commodity business.
> But perhaps still useful for planning, as a starting place?
Yeah, lots of value in just having an app go "the mix you're trying to do is likely to fail the slump test". So you still have time to adjust the water ratio, or get better sand.
That's the job of whatever engineer writes the spec. for the concrete. If they can't get that right, they apparently have failed to learn anything from decades of engineering literature.
It is simply difficult to hide. You can just go and look at the infrastructure, after all. I bet almost all the information is from OpenStreetmaps, and people just walked around and added all the power lines, substations and powerplants they saw by hand.
And sure, you can bury the cables, or you can try keeping the output of your powerplants secret. But then the infrastructure nerds (or foreign spies) just count coal hopper railway cars per day and analyze cooling tower dimensions.
Is RAG even necessary here? Minimal information like a couple of price list with job times and opening hours should easily fit into any context window, right? It's not like he's dumping entire service manuals into the vector database here...
It most likely isn't, but it seems like this project was more for learning purposes than for anything else. In that case, why not go for the "production-ready", "highly scalable" solution? I sometimes do the same for my personal projects. I over-architect them not because it's necessary, but because I want to get my hands dirty and learn something new.
For voice conversations the issue can be more latency than filling the context. Without knowing the site is hard to say, but if he had multiple pages worth of text (dunno, type of cars, procedures, some emotional story, etc.) and a "slower" model, it might be worth it to use RAG to preselect fast a small portion and use LLM to refine the answer.
But why? What would change compared to 12 players on a huge map? Do you just want to conquer a large number of your neighbors and then still get to compete against other mega-empires?
When playing civ on 12 player maps, I still mostly interact only with the 3-4 that I directly compete with at any given time. I imagine that wouldn't really change with 100 people.
> Take care - the Hybrid battery can be expensive to replace and they do eventually fail.
That is true, but median mileage at replacement for the old NiMH batteries is 150k miles (240k km), and the lithium cells have a median mileage at replacement of over 200k miles (320k km) - even though those cars are now 10 years old, not enough of them have reached that mileages, so exact data is still not available.
And don't get me wrong, those cars are bullet proof. Median total mileage of the car could be a bit higher than 150k miles, especially after the car was sold to a third world country. But for most intents and purposes, those batteries (especially the lithium cells) have about the same median lifetime than the car itself.
Depending on the max speed of the motors/legs, giving it longer foot pads might be necessary for a good gallop. Intuitively, it looks a bit... "low gear" in the videos.
reply