mgraczyk

11 ms·

151.
by mgraczyk 1y ago
You are claiming that Canada will stop exporting to the US in the future? Populism does the opposite, they would likely stop importing. There's almost no risk to us that they would not export chips
152.
by mgraczyk 1y ago
How? What if we just decide we will never go to war with Mexico or Canada and get comfortable with the idea of importing from our allies? There is no serious future risk from doing that
153.
by mgraczyk 1y ago
Would you say that computers are less efficient now than they were in the 90s because they are more widely used?
154.
by mgraczyk 1y ago
It's 1/33, I don't see how it could be 1/34. Nobody would ever make that mistake and it wouldn't matter, close enough
155.
by mgraczyk 1y ago
I would assume so. One important trend is that models have gotten more intelligent for the same size, so for a given product you can use a smaller model. Again this is pretty similar to how CPUs have changed
156.
by mgraczyk 1y ago
They reuse the word here for a product, not a model. It's the name of a specific product surface. There is no single model and the models used change over time and for different requests
157.
by mgraczyk 1y ago
That isn't what that means. Look at the paragraph above that where they explain. This is the median model used to serve requests for a specific product surface. It's exactly analogous to upgrading the CPU in a computer over time
158.
by mgraczyk 1y ago
Figure 4 I think you are assuming we are talking about swapping API usage from one model to another. That is not what happened. A specific product doing a specific thing uses less energy now. To clarify: the way models become more efficient
159.
by mgraczyk 1y ago
> total energy use across their whole AI product range, including training, is the only number that really matters. What if they are serving more requests?
160.
by mgraczyk 1y ago
Reducing something 33x means to make it 33 times smaller. It's a common way of saying this in English
161.
by mgraczyk 1y ago
No it isn't Suppose you were running a computation that requires doing 33,000 multiplies. Later you find a way to do the same computation using only 1,000 multiples That's basically what happened here
162.
by mgraczyk 1y ago
They did specifically say in the linked report
163.
by mgraczyk 1y ago
As others have pointed out, this is false. Google has made their models and hardware more efficient, you can read the linked report. Most of the efficiency comes from quantization, MoE, new attention techniques, and distillation (making sma
164.
by mgraczyk 1y ago
The site doesn't load any css for me on mobile, is it really worth it not to use GitHub or gitlab?
165.
by mgraczyk 1y ago
Do you believe that AI systems could be conscious in principle? Do you think they ever will be? If so, how long do you think it will take from now before they are conscious? How early is too early to start preparing?
166.
by mgraczyk 1y ago
Yes it's worth it, as long as he will get a job working on or with AI. There will still be plenty of jobs in 4 years and he will presumably use AI so much in college that he will be better at using it than most workers who did not stud
167.
by mgraczyk 1y ago
Yes I've been measuring it and I sleep more as a fraction of the time I spend in bed. Part of this is paying more attention to things like how late I work out and drink water, but without the tracker it would be hard to pay close enoug
168.
by mgraczyk 1y ago
I have spent my whole life mostly buying nothing and I realized recently that I am probably pretty far from optimal on that tradeoff. Outside of food, rent, and travel I've been spending around 5k/year my entire life (mostly lapto
169.
by mgraczyk 1y ago
Your observations are limited, if you average over large enough an area you will see a different story
170.
by mgraczyk 1y ago
From the line before: > Data from the fed and payroll providers show this overwhelmingly to be the case, but just to add some color/anecdote.
171.
by mgraczyk 1y ago
Uber basically didn't exist 10 years ago and was insanely VC subsidized. Compare cab prices. Drinks in some places are more, other places have not increased as much. You basically couldn't find a $5 drink in SF in 2015. You can st
172.
by mgraczyk 1y ago
Maybe I am misreading but it looks like young people spend less than average Copy it into ChatGPT and ask "how much does each age group spend on housing"
173.
by mgraczyk 1y ago
I don't think what you're saying is true actually, do you have data? I assume young people actually spend a smaller proportion on rent because older people spend a very large portion (65+ spend around half)
174.
by mgraczyk 1y ago
But what you are claiming is contradicted by the data you shared. When you weight the categories you listed by how much that age group spends, they still have more money (young people spend much much less on healthcare, you'd be shocke
175.
by mgraczyk 1y ago
The best data I could find shows a decline of around 25% from 2006 to 2023 in restaurant visits. However, a big portion of this is because of meal delivery which is more expensive than restaurants, so the cause is probably not mostly increa
176.
by mgraczyk 1y ago
Sorry I do not understand what that means. You're talking about opportunity cost? In what sense is "time spent" economic?
177.
by mgraczyk 1y ago
Here's men 16-24 showing 20% increase after CPI adjustment https://fred.stlouisfed.org/series/LEU0252882200Q
178.
by mgraczyk 1y ago
Any source will do, here is the Atlanta fed https://www.atlantafed.org/chcs/wage-growth-tracker Click "age" Then compare to price levels. Wages have outpaced price levels for this age group significantly Here
179.
by mgraczyk 1y ago
Yeah I'm taking that into account
180.
by mgraczyk 1y ago
First, young people make a lot more than they did 10 years ago (both nominally and inflation adjusted). Second, no it does not cost 5x as much, closer to 15-20% more based on all the data I could find. Anecdotally in San Francisco, NYC, and

More ›