reitzensteinm

8 ms·

91.
by reitzensteinm 2y ago
R1 is available on both together.ai and fireworks.ai, it should be a drop in replacement using the OpenAI API.
92.
by reitzensteinm 2y ago
When GPT-4 came out, I worked on a project called Duopoly [1], which was a coding bot that aimed to develop itself as much as possible. The first commit was half a page of code that read itself in, asked the user what change they'd lik
93.
by reitzensteinm 2y ago
I think you might be missing that Swizec edited the quote, crossing out microservices and correcting it to modular systems. It seems to me you're both in violent agreement.
94.
by reitzensteinm 2y ago
The parent comment was not written by the same author.
95.
by reitzensteinm 2y ago
Those listings are measured in gigabits, not gigabytes. So it's $1.50-$5 per GB depending on size.
96.
by reitzensteinm 2y ago
Yes, instead of removing attribution and keeping the content, if Stack Overflow doesn't want to be associated with the user they should just delete all of it. And pray Jon Skeet stays on the straight and narrow.
97.
by reitzensteinm 2y ago
Do you have some references of New Glenn being positioned as a Falcon 9 killer by any industry professionals, commentators or press with credibility (at the risk of no true Scotsman)? This was certainly thrown around with Tesla, but it&#x27
98.
by reitzensteinm 2y ago
48GB is maybe just enough to squeeze a quantized 70B model in, like Llama 3.3, but you'll need to raise the GPU memory allocation limit [1] and it might not be super fast. You could also try Qwen 2.5 32b, which you should just work wit
99.
by reitzensteinm 2y ago
Mixture of Expert models, where all parameters must be in memory but only a subset are accessed per token, are a sweet spot for Macs. DeepSeek v3 for instance has 671B params, but should have the memory bandwidth of a 37B dense model with a
100.
by reitzensteinm 2y ago
Agreed.
101.
by reitzensteinm 2y ago
It's possible they're executing many simulations in parallel, and counting that. 16k robot arms executing at 3k FPS each is much more reasonable on a 4090. If you're effectively fuzzing for edge cases, this would have value.
102.
by reitzensteinm 2y ago
I agree with you, and think copyright should be for a short period disconnected from the lifetime of the author, but _no logical argument_ is taking it way too far. Here's one: Copyright expiration during an author's lifetime dire
103.
by reitzensteinm 2y ago
Right - if the pattern in the referenced blog post is used to implement lazy init, racing threads can double initialize. But it's not possible to upgrade a read to write lock as racing threads can cause a deadlock, and upgradable RwLoc
104.
by reitzensteinm 2y ago
I think rather than experimenting with fabbing on Mars, you would probably be better off sending an absurd amount of CPUs from Earth. A 300mm wafer weighs 150 grams and can fit 30k Raspberry Pi Nano CPUs on it, 2mm^2 on 40nm. If you send up
105.
by reitzensteinm 2y ago
It's still possible to incorrectly use relaxed operations, and have your algorithm only incidentally work because the compiler hasn't reordered them and you're on a CPU with a stronger memory model. But yes, it's an orde
106.
by reitzensteinm 2y ago
Why are you comparing the rate of change of battery storage capacity, the vast majority of which if grid connected will be used for at most diurnal storage, to yearly energy consumption? Holy mother of all type errors there. Multiply it by
107.
by reitzensteinm 2y ago
That's great news. Thanks!
108.
by reitzensteinm 2y ago
To be frank, after reading your reply I'm still not 100% sure what the policy is. What are the situations where there are 30 days of notice? How does that interact with hourly billing? If I'm a week through the second month and ca
109.
by reitzensteinm 2y ago
That term appears on all servers, but seems to be in conflict with their billing FAQ: https://docs.hetzner.com/general/others/new-billing-model/ Either way... it's a mess. They added hourly billing for d
110.
by reitzensteinm 2y ago
That's incorrect. Dedicated servers do now offer hourly pricing. See this page which lists the hourly pricing for the new GEX130 server: https://www.hetzner.com/dedicated-rootserver/matrix-gpu/ But, as far
111.
by reitzensteinm 2y ago
But the setup fee is €86.90. Is that not paid each time you provision the server?
112.
by reitzensteinm 2y ago
At the risk of anecdote extrapolation, I did a 1000 deficit to lose 1kg a week and never felt hungry to do it. There are enough fruit and vegetables like celery, carrots, cucumbers, apples, pears etc that you can eat if you feel hungry, and
113.
by reitzensteinm 2y ago
If you don't have rent control or NIMBY blocking construction, you've got a free market and everything will be fine. The legality of Airbnb is completely orthogonal to this. I'm not even suggesting that cities should ban or r
114.
by reitzensteinm 2y ago
Banning or controlling Airbnb is simply not analogous to rent control. Assuming free market long term rentals and no NIMBY building restrictions, long term rental prices will fluctuate, and if the demand is sufficient, more units will be bu
115.
by reitzensteinm 2y ago
How would staying in an Airbnb give you any insight? The problem isn't the hosts, it's the folks staying. They're substantially more likely to be drunk and partying, and don't need to worry about any long term conseque
116.
by reitzensteinm 2y ago
AirBnB units are profitable but impose significant externalities on surrounding properties, including noise and antisocial behaviour. Unlike a hotel, they're not staffed with somebody who can deal with the problem or escalate to the au
117.
by reitzensteinm 2y ago
You're looking at the pricing for Gemini 1.5 Flash. Pro is $3.50 for <128k tokens, else $7.
118.
by reitzensteinm 2y ago
Well, there's a third option: Optimus joins self driving to be the stock pump du jour, but Tesla attempts to methodically builds its core businesses, EVs and energy storage, to grow into its current valuation. Just like you, I'm n
119.
by reitzensteinm 2y ago
Or 1.74x faster than a Model S Plaid: https://evkx.net/models/tesla/model_s/model_s_plaid/charging... Or 1.89x faster than a Model 3 Performance: https://evkx.net/models/tesla/m
120.
by reitzensteinm 2y ago
I've been building a library in Rust with a very similar approach called Temper [1], although in order to model the bizarre implications of the full Rust memory model you have to go quite a bit further, with bookkeeping tracking which

More ›