mechagodzilla

9 ms·

61.
by mechagodzilla 2y ago
Well each new generation of model costs like 10x the previous one to train, and its value (and thus ability to generate a return) diminishes extremely rapidly. The only source of improved economics is the rapidly evaporating Moore's La
62.
by mechagodzilla 2y ago
I think their goal was interesting for the time, but its release coincided first with the whole 'netbook' craze which meant, all of a sudden, you could buy a regular laptop for $100-$200, rather than the maybe $700-$800 low-end
63.
by mechagodzilla 2y ago
There is an additional problem in that you want to avoid having people try to sell snakeoil to the desperate because "who knows, it might work."
64.
by mechagodzilla 2y ago
Siri and Alexa are good examples - they turned out to be incredibly expensive to develop and had extremely limited productivity benefits for users, while being almost impossible to monetize.
65.
by mechagodzilla 2y ago
Increased temperatures also affect evaporation rates (and hence how much watering needs to be done). Climate change and the future of farming are pretty intimately connected.
66.
by mechagodzilla 2y ago
I think you are off by a factor of 1000 - MTA lost $600M (not B!) last year.
67.
by mechagodzilla 2y ago
It does happen occasionally, but it's rare that my panels are covered for more than a day or so. Heavy snows almost always slide off within 36 hours - sometimes a light snow (~3" or so) can cover it for a while because it's n
68.
by mechagodzilla 2y ago
Is that true? There's a ton of overhead required to have any presence in a city (all of the fueling/cleaning/repairing/storing/etc depots) that you need to amortize over lots of cars. If they can keep the cars bus
69.
by mechagodzilla 2y ago
My 6 year old loves kidpix 3 on an old PPC iMac. It was just a great piece of software.
70.
by mechagodzilla 2y ago
As you guessed, high-speed machine vision stuff is frequently used in manufacturing settings for sorting or various quality control tasks. Imagine a picking out bad potatoes on a conveyor belt moving 10s of potatoes per second, or identifyi
71.
by mechagodzilla 2y ago
It seems like these LLMs are 1) extremely expensive to develop, and 2) extremely difficult to actually monetize. No one but Nvidia actually seems to be making money.
72.
by mechagodzilla 2y ago
This is basically tea-leaf reading on your part - a dart board might give similar (or better!) results if tested in such a fashion, and it doesn't change the overall result that just putting your money in $SPY over the whole duration o
73.
by mechagodzilla 2y ago
The 1-year return on $SPY is 26%, or 67% better than your GPT-4 return ( https://www.google.com/finance/quote/SPY:NYSEARCA?window=1Y ). I think you've been fooled by randomness over very short intervals.
74.
by mechagodzilla 2y ago
Which empty, Taiwan-sized island did you have in mind?
75.
by mechagodzilla 2y ago
Donald Trump owns ~80 million shares of Truth Social (58%). The company, as a profit-making enterprise, is basically an abject failure, but as a way to funnel money to Donald Trump, it is a roaring success. It would be frowned upon if a for
76.
by mechagodzilla 2y ago
It doesn't necessarily discourage investment much from the perspective of society as a whole - 1 person worth $20 million isn't obviously making way more investments than 2 people worth $10 million (or 4 people worth $5 million).
77.
by mechagodzilla 2y ago
AmeliaBedeliaGPT
78.
by mechagodzilla 2y ago
I think there's a better argument for banning AirBNB on an island with extremely limited land - Tourists should stay in high-density hotels to minimize their impact on land use. Converting residential housing into extremely low-density
79.
by mechagodzilla 2y ago
Wait - you have a HPWH connected to a hydronic floor system? That's an extremely inappropriate setup - the heatpump on a rheem is probably like ~4000 BTU/hr, and it's pulling the heat from the conditioned space, then you&#x27
80.
by mechagodzilla 3y ago
We find ourselves, in 2024, with unemployment levels (at least in the US) at near all-time lows. This is after centuries of mechanization and automation, decades of the computer and internet revolutions (which have barely any steam left, co
81.
by mechagodzilla 3y ago
From OpenAI’s perspective the cost improved!
82.
by mechagodzilla 3y ago
Can you not plate multiple copies of it that print at the same time? Usually you need to remove the finished print, clean the build plate etc. between prints, so longer total print times between 'overhead' periods increases your p
83.
by mechagodzilla 3y ago
I use a 3D printer all the time, but I almost never print a file more than once. It also takes a couple of seconds to select a file with the built-in interface, and then it usually takes anywhere from 30 minutes to a day or more to print. W
84.
by mechagodzilla 3y ago
I'm mostly saying that unless a chip-designing AI model is an actual magical wizard , it's not going to have a lot of advantage over teams of even mediocre human engineers. All of the stuff you're talking about is Moore&#x27
85.
by mechagodzilla 3y ago
That speedup compared to Nvidia's previous generation came nearly entirely from: 1) a small process technology improvement from TSMC, 2) more silicon area, 3) more power consumption, and 4) moving to FP4 from FP8 (halving the precision
86.
by mechagodzilla 3y ago
They might generate improvements, but I’m not sure why people think those improvements would be unbounded. Think of it like improvements to jet engines or internal combustion engines - rapid improvements followed by decades of very tiny i
87.
by mechagodzilla 3y ago
I love walking on the beach, and I can’t imagine seeing a tiny structure miles off in the distance somehow detracting from the experience.
88.
by mechagodzilla 3y ago
What is the read latency like on HBM3E?
89.
by mechagodzilla 3y ago
Those sorts of issues are part of timing analysis for a chip, but once a chip's clock rate is set, they don't really factor in unless there is some kind of dynamic voltage/frequency scaling scheme going on. This chip probably
90.
by mechagodzilla 3y ago
You all seem like one of the only companies targeting low-latency inference rather than focusing on throughput (and thus $/inference) - what do you see as your primary market?

More ›