foota

7 ms·

61.
by foota 5mo ago
I've generally assumed that AI would make developers get lower compensation because of the lowered quantity of developers required for the same output, but this raises the possibility of it actually increasing if more developers end up
62.
by foota 5mo ago
Imo the main benefit of AI is allowing faster experimentation and more parallelism, as well as faster iteration.
63.
by foota 5mo ago
Shouldn't there be a feedback system here preventing the scheduling of loads when cooling is degraded?
64.
by foota 5mo ago
I wonder if this is in response to the chrome incognito lawsuit.
65.
by foota 5mo ago
cries I am sad at all of the cool designs that would work with optane if it had been widely adopted. We desperately need more points on the latency/cost per GB curve.
66.
by foota 5mo ago
This doesn't math out to me just based on what I know of energy consumption numbers.
67.
by foota 6mo ago
"is duplicated by" is amusing.
68.
by foota 6mo ago
While I largely agree, it does raise the prospect of testing this iteratively. E.g., give a model some fake environment, prompt it random things until it does something "bad" in your fake environment, and then fix whatever it clai
69.
by foota 6mo ago
Yes, sort of. Generally you can measure the pass rate on a benchmark given a fixed compute budget. A sufficiently smart model can hit a high pass rate with fewer tokens/compute. Check out the cost efficiency on https://artif
70.
by foota 6mo ago
> On April 16, we added a system prompt instruction to reduce verbosity. In combination with other prompt changes, it hurt coding quality, and was reverted on April 20. This impacted Sonnet 4.6, Opus 4.6, and Opus 4.7. Claude caveman in
71.
by foota 6mo ago
I found this while looking for a solution for more easily removing some unsafe code from a library that does a lot of C FFI. I didn't end up going with it though, for now I'm taking an approach of mapping valid pointers that I ret
72.
by foota 6mo ago
That's curxed
73.
by foota 6mo ago
It was admittedly a bit sloppy. The more accurate version (IANAKSPP, I am not a kerbal space program player) would be that the only way to maintain a satelite in one position (without expending an infeasible amount of energy) is to position
74.
by foota 6mo ago
I don't believe parent is right, but satelites don't stay in one place unless they're on the equator, because otherwise they have to be moving. This means that you need many satelites to maintain coverage of a single spot. I
75.
by foota 6mo ago
Yeah this wasn't something like "I want to debug a program" but rather I wanted to be able to track mmaping for later cleanup. Fortunately libc doesn't mmap that much internally so I think I can get away alright with int
76.
by foota 6mo ago
Fun fact, overwatch must have done a similar things because they would let you play back games up until some release when you could no longer replay them unless you'd saved the render. I think if I remember right there were also funny
77.
Garbage collection without unsafe code(fitzgen.com)
114 points by foota 6mo ago|83 comments
78.
by foota 6mo ago
Hah, I've been looking into something amusingly similar to track mmap syscalls for a process :)
79.
by foota 6mo ago
More like RAM producers are providing supplies to the highest bidder, no? If this doesn't peter out supply will normalize at a higher but less insane price eventually.
80.
by foota 6mo ago
Capital and availability?
81.
by foota 6mo ago
Snowpiercer but with data centers? The breeze would help with cooling!
82.
by foota 6mo ago
Eh, that doesn't math out. It's the bandwidth per storage density (or ultimately per price) that matters. If you have great cost per byte but your bandwidth per byte is bad enough that the price per byte doesn't make up for i
83.
by foota 6mo ago
The cost issues they're seeing (at least from what they've stated) are from users, not internally. Basically, it takes either $5 or $6.25 (depending on 5m or 1h ttl) to re-ingest a 1M context length conversation into cache for opu
84.
by foota 6mo ago
Hey -- I have 0 PHDs so take this with a grain of salt :) I had thought for a while about a way to store data that makes use of an idea that I had for sub-diffraction limited imaging inspired by STED microscopy. First an overview of STED. Y
85.
by foota 6mo ago
Late to reply, but.. yes, hence my reply that it would need to be integrated all the way down the stack. But also, LLMs (or their current implementation) rely heavily on print caching for efficiency, without this costs are much higher. You
86.
by foota 6mo ago
I would go for out of touch, not cynical. A lot of people really think AI is the devil.
87.
by foota 6mo ago
Some others mentioned pijul, but I will put in my two cents about it. I have been looking to make use of it because it seems really nice for working with an agent. Essentially you get patches that are independently and can be applied anywhe
88.
by foota 6mo ago
Not really, because the LLM loop doesn't have the ability to get updates from the agent live. It would have to somehow be integrated all the way down the stack.
89.
by foota 6mo ago
And they are doing better.
90.
by foota 6mo ago
So.. it sounds like they're doing a lot better to me? 19 cases in the fall, 4 between the recall in Novemberish and Jan, and 1 between them and now that occurred in Jaunary? Also lol at this quote in the article "Six vehicles pass

More ›