I wonder what is the general use for the Mac Mini, MacBook Air, iMac, and MacBook Pro? People generally seem to do all the lightweight stuff like social media consumption on their phones; and desktops/laptops are used for the more heavy-weight stuff. The only reason I’ve ever used a Mac was for IOS development.
I was looking at notebooks at Walmart the other day, and I was amazed that they almost all had less or the same amount of RAM as my phone.
I think SSDs are also soldered to the mainboard on most apple products.
Most CEOs are also conmen and not any smarter. It’s mostly just a nepotism racket at the executive level.
A lot of the “elites” (OpenAI board, Thiel, Andreessen, etc) are on the effective-accelerationism grift now. The idea is to disregard all negative effects of pursuing technological “progress,” because techno-capitalism will solve all problems. They support burning fossil fuels as fast as possible because that will enable “progress,” which will solve climate change (through geoengineering, presumably). I’ve seen some accelerationists write that it would be ok if AI destroys humanity, because it would be the next evolution of “intelligence.” I dunno if they’ve fallen for their own grift or not, but it’s obviously a very convenient belief for them.
Effective-accelerationism was first coined by Nick Land, who appears to be some kind of fascist.
We’re close to peak using current NN architectures and methods. All this started with the discovery of transformer architecture in 2017. Advances in architecture and methods have been fairly small and incremental since then. The advancements in performance has mostly just been throwing more data and compute at the models, and diminishing returns have been observed. GPT-3 costed something like $15 million to train. GPT-4 is a little better and costed something like $100 million to train. If the next model costs $1 billion to train, it will likely be a little better.
LLMs do sometimes hallucinate even when giving summaries. I.e. they put things in the summaries that were not in the source material. Bing did this often the last time I tried it. In my experience, LLMs seem to do very poorly when their context is large (e.g. when “reading” large or multiple articles). With ChatGPT, it’s output seems more likely to be factually correct when it just generates “facts” from it’s model instead of “browsing” and adding articles to its context.
I think part of the problem is that new cars are bought mostly by fairly well-off individuals; with other people buying used cars. Economy cars sell poorly in the U.S.
China can’t arrest people in the U.S. The U.S. and state governments can.
Doesn’t really matter unless you live under it. Instagram and Twitter or more dangerous to U.S. citizens.
Sorry, I was unclear. Chippers are not the tools I was thinking of that would be cheaper to buy (a low quality version of) than rent. Was thinking more about stuff like torque wrenches and rotary hammers. Chipper rental prices were just one thing I was looking at recently that seemed way out of line with what other people from other regions were paying.
Rentals seem extremely expensive in my area. $100/day for a shitty 4" wood chipper, $300/day for 6" chipper. For some tools, it’s often about the same price or cheaper to buy a tool from Harbor Freight than to rent.
Libraries don’t make a profit, AFAIK. Non-profits and co-ops are things too.
I used adblockers back then too. Else some sites would cause infinite pop-up windows to open (I assume to get pay-per-click revenue). Even plain banners would significantly increase loading times on 56k connections.
I’ve used FreeCAD for hobby 3d printing and plywood CNC projects. It seemed buggy, and the workflow seemed strange, but I’ve never used anything else, so it’s fine, I guess, lol.
Kagi is pretty good, but expensive. I like that listicles are put in their own small section so you can ignore them. You can boost, pin, demote, or block results from certain domains. You can create and quickly switch between domain list presets to search only specific sets of sites. The only thing I don’t like is the exerpts below results don’t bold what you’re looking for like Google does.
This just reminded me that I saw an advertisement for some kind of “flavored air” device at gas station yesterday, lol.
The equivalent of 600k H100s seems pretty extreme though. IDK how many OpenAI has access to, but it’s estimated they “only” used 25k to train GPT4. OpenAI has, in the past, claimed the diminishing returns on just scaling their model past GPT4s size probably isn’t worth it. So, maybe Meta is planning on experimenting with new ANN architectures, or planning on mass deployment of models?
This seems like the most plausible explanation. Only other thing I can think of is they want to develop their own CoPilot (which I’m guessing isn’t available in China due to the U.S. AI restrictions?), and they’re just using their existing infrastructure to gather training data.