Depends. ROCm is pretty well-supported for example.
Non-NVIDIA backends tend to get less support and new features land slower, or features that are expected to improve performance wind up hurting it instead. That sort of thing.
For basic “token in/token out” workloads without fine tuning, it’s probably fine ??
The Ryzen AI Max 395 128gb is super cool, but not fast for inference. Order of magnitude slower than dedicated GPU but at half the cost. You can run larger models on it but it's slow. Great for local async work. Not great for daily chat or code agent driver.
I mean we are still waiting for the European Apple or Microsoft or Google or anything of that scale aren't we? China has it's own version of all this stuff, where is Europe's?
It's probably better not to repeat the same mistakes that the US made and not have monopolies. Building a diverse ecosystem with strong competition would result in a world leading economy. Essentially there needs to be 10x more Mistrals.
Americans are just as regulated and with even more egregious regulatory capture (see the American broadband/mobile provider cartel). At least Europeans get a little consumer protection.
america regulates to ensure companies (as few as humanly possible) profit while europe regulates to protect the consumers. the core difference, yea! and americans seemed to be always shitting on europe completely oblivious of what is being done to them, for decades now
I get what you're trying to say, but having just had some meetings on aligning with the latest AI act, as a small startup focusing on adapting open source models for local use, the way it looks now, it seems like we won't have any consumers worth protecting :) (and I'm only half joking. AI act is a joke, and Mistral have been saying this for a while now. We have chinese models launched with permissive licenses, usable everywhere except in the EU. It'd be good comedy if it wouldn't be tragic. ffs! We can't use open source shit!!! It's bananas)
Otoh, you can buy contact lenses OTC in other European countries... not saying Europe isn't over-regulated (try legally building a bike-trail on your own forest-land in Germany!! ...insanity) but it's a bad example.
If anything the main simplification here is that China is doing a significant amount of innovation too. Otherwise yeah. Europe's main purpose is to serve as a continent-wide historical Disneyland for rich American and Chinese tourists. And maybe we'd be better off admitting that rather than trying to pretend to have any kind of industrial or startup ecosystem.
I applied for a founding engineer with a startup though a "who's hiring" post. Had 3-4 rounds of interviews, met founders, did a project with them. I had competing offers and ultimately didn't work out. But it was a positive experience.
Cockroaches everywhere: starlink revenue won't be enough to fund constellation sustainment. Refurbishment costs for Falcon 9 make the reusability economics dubious. And xAI is just a disaster. So is Twitter.
I mean, either a company is profitable or it is not. You don't get to say "if you ignore the massive pile of burning money in our parking lot, we are profitable". Or, well, I suppose you can _say_ that (see WeWork), but you can't expect analysts to be too impressed with it.
The S-1 will be interesting; I wonder if they'll try going with a bulllshit metric like WeWork did ("Community Adjusted EBITDA", which was, seemingly, pretty much "profit, if you exclude basically all costs").
Since Twitter is now part of SpaceX, does that mean Twitter is classified as a common carrier too? Can Twitter no longer refuse to allow certain individuals use their platform?
Apple wants to vertically integrate. Their AI strategy until recently was to develop their own LLM models that were small enough to run on device. But massive scaling is what makes LLMs so powerful, so all their internal models were terrible and unusable.
Basically they bet that compute efficient LLMs were the future. That bet was wrong and the opposite came true.
Modern concrete construction uses iron rebar liberally. That means every concrete structure built today will crack and crumble in a few hundred years at most, as the iron absorbs oxygen, it swells from the rust. Which is a shame, roman concrete buildings without rebar will still be standing 1000s of years from now.
Roman construction was also much less efficient because they had no material (besides wood) capable of carrying load in tension. Rebar allows us to make cheap practical structures that are impossible with just concrete - roman style or not.
It would be quite fascinating to see what kind of structure we could produce if we decided to make the longest lasting cement structures we could create with modern technology, and assuming minimal maintenance over the lifetime of the building. A one-and-done kind of structure.
I bet we could do fairly well. Hundreds, maybe even thousands of years. We've learned a lot about how to form exceptionally long lasting cement. We just choose not to do it that way, most of the time.
Petrified wood is stone. The stone matrix has formed around a wood structure, but the properties which remain are those of stone, not wood.
Wood is a composite material, strong under both compression and tension. It's the fibres in wood (lignans) which provide the latter. Stone (and unreinforced concrete) lack such fibre elements, and are strong only under compression.
With stone it's necessary to build compression structures such as arches and domes for heavy load-bearing, and taller structures must be significantly flared out at the base (or utilise buttresses, as with gothic cathedrals) for stability. Structures under tension can be far more lightweight and compact.
Steel-box construction and reinforced concrete both offer tension-based strength, but are also susceptible to metallic oxidation (rust), which limits the life of such structures. A nonmetallic fibre (natural or synthetic) might offer an alternative to this, and I've seen some work investigating this, though there may be other issues (e.g., depolymerisation of plastics over long periods of time).
