it’s the end of the world as we know it. (and I feel fine) https://xitter.gq/t.co/EI0HiYYCpY
Hell, CA
Joined April 2010
- Tweets20.5K
- Following1.1K
- Followers155
- Likes84.1K
I think that the one thing that most comments about this miss is that game players are probably a super spiky use pattern. The whole idea of using the cloud to 'share load' depends on there being a way to evenly distribute that load over the whole day.
you’ll get mad at me for saying this…but cloud gaming is so obviously more economically efficient than physical hardware I think it’s going to be the default soon.
your home console / pc is idle 90%+ of the day. meanwhile, data centers targets what, 5%, maybe at worst 10% idle.
every second a cloud gamer isn’t gaming, that hardware is being used for someone else, training, etc.
I think there should be a new measurement, something like cost-per effective FLOP hour that takes into account the TCO + effective utilization.
If a gamer spends $500 on a GPU, uses it for 3 years, but it’s only fully active ~5% of that period…the cost-per relative FLOP hour is crazy high! Meanwhile, a $50,000 datacenter GPU might have a *LOWER* cost-per FLOP hour just because the effective utilization is 90+%.
1
19
Using a technology design to generate text forever is probably not the time saver we all wanted it to be.
45
When people talk about optimization, I see why. This is so much memory to allocate to a launcher. They need to drop this down to 1/100th of this footprint. 2+G? Should be a few tens of MB at most in this day and age.
20
All these guys know that LLMs are just a lossy compression algorithm over the input corpus. WTF guys, we had a search engine, LLMs are not supposed to be a search engine like that.
Wait, this sounds incredible useful! Can we just have a model with 0 entropy, 0 hallucinations, that just acts like a retrieval database over its training dataset? Also sounds like a great way to solve the traceability problem. Why don't the AI labs just make something like that?
95
This whole RAM pricing debacle is going to cause a complete collapse of the PC market that is going to be a trigger for the bubble popping.
36
So this idea is trash, but I know that the geniuses at Google and Facebook are trying to make this idea of a fully machine language generated internet a reality.
27
I have my hangups too, but not everything needs to be turned into an allegory for your hyperfocus.
EXTREMELY UNPOPULAR OPINION: I don't support drag queens or gender ideology in public schools because it freaks parents out with the mistaken idea LGBT people aren't born like this and parents can avoid it by limiting exposure.
But I think it creates more people that will experiment sexually, claim these identities for a while, and move on.
Just like this peanut allergy, if people have no exposure to these ideas - you make it forbidden and therefore more appealing to people to that might have explored it and moved on.
I would have transitioned no matter what, was always going to date men. But, I had the extreme version of gender dysphoria. Most people claiming to be trans don't.
6
I know he's basically the reason that we all have GPUs now, but it makes me so sad to see him be to AI pilled. Best case he invents a world changing tech, worst case he invents a world changing tech that hurts society.
Natural conversation includes interruptions and talking over people, which is hard for an LLM to model as a single autoregressive sequence. I’m sure you can get pretty far by creating a text sequence with movie-script like breaks mid sentence, but it seems like the real solution would involve parallel streams of listening and thinking with talking queued up for pauses or rising to an interruption priority. Intermixing tokens from different streams and doing something custom with attention seems plausible.
10
I read comments like this and think, “people have pattern seeking brains”. Some of us more so than others.
While noting some compression artifacts in a water scene on TV, I started thinking— for a given set of compression parameters, what is “the most cursed macroblock”. Not the worst looking, which is a squishy perceptual metric, but the set of pixels that takes the most bits to encode.
With all the non-linearity in pixel quantization and entropy encoding, finding that is actually non-trivial.
11
It’s not a vacation, it’s a micro retirement so I can use my 401k to pay for my trip to the Bahamas.
You’re unable to view this Post because this account owner limits who can view their Posts. Learn more
10
This is more about telegraphing that they are about to have a layoff more than anything else. This is not a true statement.
9
