Post AuwAX40SdsPDOXcu24 by [email protected] | |
More posts by [email protected] | |
Post #Auvt5z9orecZJfVRRI by [email protected] | |
0 likes, 1 repeats | |
So: let's imagine the AI bubble deflates.The people who want chatbots or LL… | |
Post #Auvt5zGCTvj1dSUXNw by [email protected] | |
0 likes, 1 repeats | |
@davidgerard I think it might turn into a shift towards local LLMs using very e… | |
Post #AuvxS0awbGWlxhwzS4 by [email protected] | |
0 likes, 0 repeats | |
@davidgerard Why can't the AI companies just ask the AI how to sell the AI … | |
Post #AuvxSEtBSFR6IKakpE by [email protected] | |
0 likes, 0 repeats | |
@davidgerard At that point it would be pretty cheap; there are freely available… | |
Post #AuvxSNTHXv44ubXhCa by [email protected] | |
0 likes, 0 repeats | |
@davidgerard To quote Sam Faultman: “Unimaginable” | |
Post #AuvxSvKxdMbvwduWx6 by [email protected] | |
0 likes, 0 repeats | |
@mcc @davidgerard In the future the AI will say the answer is to download model… | |
Post #AuvxTGUF03VnXVPjpg by [email protected] | |
0 likes, 0 repeats | |
@davidgerard I think Zitron made a back of the envelope calculation of this at … | |
Post #AuvxTKU28lirw1Ld44 by [email protected] | |
0 likes, 0 repeats | |
@davidgerard No, but my guess is around $50-$100/month. That is a pure guess, h… | |
Post #AuvxU1PaXAiV4gbBbc by [email protected] | |
0 likes, 0 repeats | |
@davidgerard no, but I estimate it to be unaffordable even for Musk! | |
Post #AuvxU3AQ0XlqWDW6Eq by [email protected] | |
0 likes, 0 repeats | |
@davidgerard If that unlikely scenario occurs, my escape plan is to spin up a l… | |
Post #AuvxUU79omCm5vCNP6 by [email protected] | |
0 likes, 0 repeats | |
@jmax @davidgerard I have seen comments similar to "if we charged ten time… | |
Post #AuvypPkFMJAIZu1ga8 by [email protected] | |
0 likes, 1 repeats | |
@davidgerard Billions were made and lost on the dot com bubble, when advertisin… | |
Post #AuwAKB0rHUyJMOfRDs by [email protected] | |
0 likes, 0 repeats | |
@davidgerard there was some macro analysis by financial analysts at Goldman Sac… | |
Post #AuwAKB8eoVD5kaJfNY by [email protected] | |
0 likes, 0 repeats | |
@davidgerard it links to Daron Acemoglu's paper here https://economics.mit.… | |
Post #AuwAKH0N0E97wcU8IK by [email protected] | |
0 likes, 0 repeats | |
@davidgerard Summary is that GenAI is useful in niches but overall effect is ma… | |
Post #AuwAKrJ42CUSX1VZQm by [email protected] | |
0 likes, 0 repeats | |
@davidgerard eye watering. Just thinking of the capex and opex, significant ope… | |
Post #AuwAQcibBa2T1bCVIO by [email protected] | |
0 likes, 0 repeats | |
@davidgerard I run 80bs on my shit at home but its slow. 32bs are more spritely | |
Post #AuwAQcocpArLKI1Jgm by [email protected] | |
0 likes, 0 repeats | |
@gildilinie as i note elsewhere, i'm treating home use as negligible cos it… | |
Post #AuwASKtaQBoMKN9m8e by [email protected] | |
0 likes, 0 repeats | |
@davidgerard I have, in fact, because I keep getting asked what it would cost t… | |
Post #AuwASL25uYcIkl8ZOq by [email protected] | |
0 likes, 0 repeats | |
@rootwyrm @davidgerard link with a similar estimate because I went down looking… | |
Post #AuwASM3C7x5huSexkm by [email protected] | |
0 likes, 0 repeats | |
@davidgerard let's say you REALLY POSITIVELY HAVE to buy the hardware.First… | |
Post #AuwASMVuPC3lLW5tVg by [email protected] | |
0 likes, 0 repeats | |
@davidgerard that's the "best" part of this whole shitshow. These… | |
Post #AuwASdWt8nwc7TgPtg by [email protected] | |
0 likes, 0 repeats | |
@davidgerard Dr Evil has the answer (that's per request) | |
Post #AuwAT01TVlTzsiA9cu by [email protected] | |
0 likes, 0 repeats | |
@penguin42 it'll be at least a coupla years for people who don't want t… | |
Post #AuwATbswjvEvJGSsbo by [email protected] | |
0 likes, 0 repeats | |
@daedalus Even if there WERE productivity gains, for the last 50 years producti… | |
Post #AuwATy2yLKKc2jKCI4 by [email protected] | |
0 likes, 0 repeats | |
@rootwyrmReminds me of cryptomining ASICs. Just throwing silicon at single purp… | |
Post #AuwAVBCSvRUN7tLL04 by [email protected] | |
0 likes, 0 repeats | |
@davidgerard step 1: build a product where quality doesn't matter | |
Post #AuwAWTcRvfoAXe7Dbk by [email protected] | |
0 likes, 0 repeats | |
@rootwyrm @davidgerard YES, I WAIT ;) | |
Post #AuwAWnsmaekzO5uXqK by [email protected] | |
0 likes, 0 repeats | |
@PetraOleum he put it around OpenAI spending ~$2.50 for every $1 it takes in, s… | |
Post #AuwAX40SdsPDOXcu24 by [email protected] | |
0 likes, 0 repeats | |
Apple’s GPU architecture makes it feasible to run non-trivial models on their… | |
Post #AuwAXwKWfUXdsPQgnw by [email protected] | |
0 likes, 0 repeats | |
@rootwyrm @davidgerard ty | |
Post #AuwAYWU0FjVs0E8lZg by [email protected] | |
0 likes, 0 repeats | |
@michael @penguin42 oh! that's better than i was expecting | |
Post #AuwAaNKlEvS2CrhVdQ by [email protected] | |
0 likes, 0 repeats | |
@davidgerard but you could pay the upfront cost of 4k or whatever if you want c… | |
Post #AuwYu3DM9PKcUz9Uoa by [email protected] | |
0 likes, 1 repeats | |
@davidgerard I'd guess it would be uneconomic | |
Post #AuwZwM7suMxbgaFvJw by [email protected] | |
0 likes, 0 repeats | |
@jmax @davidgerard I'd be shocked if it were that low. OpenAI is currently … | |
Post #AuwZyGm1mPBVfFweH2 by [email protected] | |
0 likes, 0 repeats | |
@n_dimension @davidgerard Services Australia ran tests and released a future st… | |
Post #AuwZyGsPOgHxz2vkDg by [email protected] | |
0 likes, 0 repeats | |
@bgrinter @davidgerard Oh it's going to be a royal cluster fuck. No doubt.B… | |
Post #Auwg8SNu45dUIOtyV6 by [email protected] | |
0 likes, 0 repeats | |
@rootwyrm @davidgerard At this price point & model size you're much bet… | |
Post #Auwg8TIydtI19PbYSe by [email protected] | |
0 likes, 0 repeats | |
@rootwyrm @davidgerard The code to run multiple clients in flight on this hardw… | |
Post #AuwixQ5uXUw3rzumXI by [email protected] | |
0 likes, 1 repeats | |
@davidgerard @PetraOleum Well, kind of. Those numbers include training costs. W… | |
Post #Aux2beYU5kObQqB6GW by [email protected] | |
0 likes, 0 repeats | |
@david_chisnall @PetraOleum > the cost of inference is low.so in current com… | |
Post #Aux5zQpt3EFnh8Le4m by [email protected] | |
0 likes, 1 repeats | |
@davidgerard @PetraOleum what sort of tokens/sec does your M2 give?I'm not … |