kingstnap

9 ms·

121.
by kingstnap 5mo ago
Ironically the few people not scamming you for cache reads are Deepseek. Everyone else charges a ridiculous amount but Deepseeks API is $0.003625 / M tok. I'm surprised no one talks about this because of how significant it is. GPT
122.
by kingstnap 5mo ago
The details are secret. It very well could be wasted GPU time but Anthropic could have made a killer offering as well. I'm just speculating, but a particularly killer offering Elon wouldnt be able to refuse would be if Anthropic agreed
123.
by kingstnap 5mo ago
This is such a bizarre setup. > American Efficient then used that sales data to calculate the energy savings from the anticipated use of the lighting and appliances, entering those projected savings into “capacity auctions.” > At capa
124.
by kingstnap 5mo ago
Gift link https://www.theglobeandmail.com/gift/2c42fdd7045b26c60c180ca...
125.
by kingstnap 5mo ago
This was my math: 0.1 kWh * 3600s/hr = 360 kJ 360 kJ / 8s (time for 1 GB) = 45 kW
126.
by kingstnap 5mo ago
10M tokens of raw execution traces to grep through is slop. The tasks are fizzbuzz, palindrome, list reversal, and sum-even. The palindrome challenge is literaly this: > Is the word "racecar" a palindrome? Answer with exactly o
127.
by kingstnap 5mo ago
0.04 to 0.1 kWh/GB is insane even for 2018 lol. I have gigabit internet (125 MB/s). This would imply when I'm downloading something I'm using 18 to 45 kW of electricity. Completely bonkers.
128.
by kingstnap 5mo ago
Right now it seems to say: > Showing 1,808 changed files with 790,916 additions and 151 deletions. Just looking at the git diff [0]. I looked at one of these rust port files [1]. Its 827 loc and apparently 7,576 tokens. So that gives you
129.
by kingstnap 6mo ago
Interest is a percentage of debt. Income through taxes is roughly a percentage of GDP. You could also just compare interest spending vs budget, and lots of people do. Spending on interest is roughly $1T out of a total $7T with income of $5.
130.
by kingstnap 6mo ago
It's not hard to imagine how this happens. I assume most people here have used these models extensively. The help bot system prompt probably includes some statement about how Claude should phrase everything as "we". The syste
131.
by kingstnap 6mo ago
Of the things you could complain about in modern cars as being too complicated, you chose turning on seat heating??? Like you push the seat heating button if your seat feels cold. What is there to think about?
132.
by kingstnap 6mo ago
They were loosing money giving absurdly generous agentic usage on expensive models to people with $10 to $40 flat rate subscriptions. They weren't selling inference.
133.
by kingstnap 6mo ago
It's unclear what he means so it would he good to clarify. Neither the Axios article nor this one provide the details. > For my team, the cost of compute is far beyond the costs of the employees," Bryan Catanzaro, vice presiden
134.
by kingstnap 6mo ago
The real question is what do you get out of advertising to people who don't have any money? Kinda squeezing blood from a stone. You'd be better off saying you use those people to A/B test changes and filling idle GPU batches
135.
by kingstnap 6mo ago
The exception being blamed on "stupidity" makes it sound like an oversight. It was not an oversight. It was corruption.
136.
by kingstnap 6mo ago
You should invert your framing. It's only *not valid* if you intend to use a fixed version forever. Otherwise you might as well include versioning for any other case.
137.
by kingstnap 6mo ago
> AI companies are hugely overvalued Most of everything tends to suck. Most projects go nowhere, most companies fail, most scientific papers are garbage. > how far away from general intelligence they still are Economically the real qu
138.
by kingstnap 6mo ago
There would need to be another paradigm shift if they wanna keep inflating AI usage. We went from simple chatbots to thinking models which massively exploded token utilization. We then went from simple thinking models to tool calls and agen
139.
by kingstnap 6mo ago
Grabbing git repos instead of just tarballs is useful. A) You can update them, because you can git pull to fetch changes. B) If you want to apply patches on top, its better to have version control so you can keep track of what you changed,
140.
by kingstnap 6mo ago
It made more sense in the ye old days where a request was basically just a chat message in a sidebar and it could also edit code. Then saying someone can use 300 chat messages a month kinda makes sense. Turns out when a request can spawn te
141.
by kingstnap 6mo ago
Gold is largely a speculative asset just like any other, despite what you might believe given the culture we have that reveres gold as "real" value. There is much more gold mined then used in jewelry or electronics. And the jewelr
142.
by kingstnap 6mo ago
I think that AI can sometimes help a lot. But I think doing it correctly is a tightrope and one misstep can easily have terrible results. First issue is this result from reinforcement learning that tells you that you really want to be doing
143.
by kingstnap 6mo ago
I feel like devs generally spend someone else's money on tokens. Either their employers or OpenAIs when they use a codex subscription. If I put on my schizo hat. Something they might be doing is increasing the losses on their monthly c
144.
by kingstnap 6mo ago
The meshes look interesting, but the gameplay is very basic. The tank one seems more sophisticated with the flying ships and whatnot. What's strange is that this Pietro Schirano dude seems to write incredibly cargo cult prompts. Ga
145.
by kingstnap 6mo ago
I mean Google ain't paying for Chromium development just for the fun of it...
146.
by kingstnap 6mo ago
> Will hear it from us, not a screenshot on X or Reddit. Has this ever been true? You will almost always see some anecdotal screenshot a long time before any company would rat on themselves. Yes the random screenshots include a lot of fa
147.
by kingstnap 6mo ago
No they are taking the massive L. Thats why they paused new sign ups. Just for context to the insanity, they allow recursive subagents to I believe its 5 levels deep. You can make a prompt and tell copilot to dig through a code base, have o
148.
by kingstnap 6mo ago
Their VR tech is pretty nice. No one sells anything anywhere near as cheap and good as the Quest 3S.
149.
by kingstnap 6mo ago
Speculative decoding doesn't degrade output quality. The distribution it produces is exactly the same if you do it correctly. The original paper on it clearly talks about this. [0] Speculative decoding is the same as speculative execut
150.
by kingstnap 6mo ago
The way revenue works on these platforms is fundamentally broken. Revenue should be considered per user to artists they listened to per month. This is actually how YouTube Premium works. Right now the way that revenue split works is you

More ›