1
point
trzy
Joined 497 karma
https://trzy.org
- Slowly then suddenly. Movements in the frontline are gradual until one side is exhausted and collapses. With Trump’s ludicrous “peace” plan, Ukraine would be barred access to US weapons, the size of its military restricted, and Russia would simply rearm and try again.
And despite how things have fared in Ukraine thus far, the Baltics are a much softer target. If Ukraine does end up falling to the Russians, it’ll be used as a springboard by the Russians, potentially supported by Ukrainians disillusioned with the West’s betrayal. It would certainly not be the first time that Russia has annexed Ukraine and mobilized its people against Russia’s foes.
- The idea that the size of Ukraine and the distance to Russia’s border through Ukraine diminishes the Russian threat. For two reasons:
1. Russia aims to either capture Ukraine outright or exert influence over it, which puts eastern EU states at grave risk. Note that Belarus, a Russian vassal, already borders the EU and was used by the Russians to launch the Ukraine invasion.
2. Russia already borders — and menaces - the EU in the Baltics.
- It’s completely overstated. The article is a bit ridiculous — like, oh no! They had to put up some English signage at temples! Foreign guests aren’t finishing their traditional breakfast!
I’ve spent considerable time in Tokyo in recent years as well. It’s largely the same as it ever was. “Incidents” with tourists are clearly blown out of proportion by local social media. Yes, there are visibly more foreigners than in 2008 (mostly immigrants who work and speak Japanese). No, the character of the city has not changed.
On my Shinkansen from Kyoto to Tokyo most recently, a noisy quartet of drunk guys chatted loudly for over a hour on the late evening weekday train and spilled a beer. Can you imagine the horror? They were Japanese restaurateurs from Osaka.
- I was there in January. Maybe not the busiest season but honestly this issue is overstated. I was the only white bro in a kimono (and it was my Japanese friend who really wanted to do it). Most of Kyoto is completely unaffected by tourism, though these are not historical areas. We had great omurice at a very local place on the outskirts of town after finishing up with Kinkaku-ji.
- I remember Sega Channel. In 6th grade, my friend had it. I didn't grow up with BBSes and had only learned about the Internet the year before (from the same friend), so the idea of downloading games was pretty wild. The service was way ahead of its time.
IIRC, it was $15/mo. There was a monthly or weekly rotation of games, including some pre-releases. I think at one point we played Vectorman when it was still new in stores or possibly just prior to its official launch.
- The problems are roughly: - Monoculture - Generally risk-averse and cliquish population whose primary concern is (understandably) survival and accumulating the money needed to live a normal middle class life and are drawn to the “peaceful” atmosphere of the region (a word you’ll hear especially frequently) - Terrible urban planning and poor density for socialization: sprawling suburbia, lifeless downtown cores, everyone is far away from each other and as a result of the grueling commutes, few are willing to spontaneously hang out after work
It’s a toxic mix that leads to a social death spiral.
- Of course not but the task requires excellent image understanding, large context window, a mix of structured and unstructured output, high level and spatial reasoning, and a conversational layer on top.
I find it’s predictive of relative performance in other tasks I use LLMs for. Claude is the best. The only shortcoming is its peculiar verbosity.
Definitely superior to anything OpenAI has and miles beyond the “open weights” alternatives like Llama.
- I'm working on robots. In the long run I'm interested in dexterous manipulation for industrial or commercial tasks (e.g., assembly of products) but given the big $$$ already deployed in this space, am wondering if I can find a simpler (and faster to build!) application.
Playing with 6-axis arms but also built RoBart for fun, controlled by Claude. Would love to connect with folks to ideate on the concept of a very cheap autonomous robot (maybe with some very limited manipulation ability for e.g. opening doors). I can think of an application in the health care space already.
Footage at my web site (which badly needs a redesign lol): http://trzy.org
- None of the small LLMs are good enough yet. You could certainly build a system around local VLMs but it would require much more task specific programming baked in.
I’m certainly interested in building a product (not entirely controlled by an LLM but I see lots of utility in building interfaces with them) but not really sure what this would be useful for. Looking into some spaces now but there has to be a clear ROI to get any sort of funding for robotics.
- Pretty cool! I use Claude 3.5 to control a robot (ARKit/iOS based) and it does surprisingly well in the real world: https://youtu.be/-iW3Vzzr3oU?si=yzu2SawugXMGKlW9
I also think that named parameters would go a long way toward improving the language.
Lastly, explore some way to make possible a breaking change with "old C++".