kingstnap

3 ms·

31.
by kingstnap 1mo ago
> Data centers in the states analyzed depend on about 3.4 trillion gallons of freshwater annually for electricity, around 12 times the annual water use of Los Angeles, Phoenix, and Washington D.C., combined. > Most electricity is gene
32.
by kingstnap 1mo ago
If you think of LLMs as programs. The weights and inference code are very much a binary. While the training code and data are the true source. Since if you want to robustly modify the LLM that's actually what you need. But since "
33.
by kingstnap 1mo ago
You don't need to seperate the clean and dirty air. Mixing them isn't as efficient but it should still be roughly 63% efficient (1-1/e).
34.
by kingstnap 1mo ago
> Overall my view is that AI is structurally a technology that tends to concentrate power > Open-weights do help some with this but are nowhere near a sufficient solution Which is exactly why Anthropic contributes nothing to, and ac
35.
by kingstnap 1mo ago
On the individual level it is a discipline problem. But the fact it has become an individual discipline problem is exactly the core issue. If you turn things into individual discipline problems then surprise you get population level issues.
36.
by kingstnap 1mo ago
Counting macros is extremely effective. The general population doesn't have the discipline for it, but amoung sport coaches, actors, models, bodybuilders, this is *the* technique for achieving goals. Anyway what's cute about this
37.
by kingstnap 1mo ago
You would think raw performance wouldn't be a problem given most of whats happening is waiting for network calls and streaming tokens. But modern bloat manages perfectly well to make apps that wait for network calls run poorly enough t
38.
by kingstnap 1mo ago
The $60 billion cursor option that SpaceX bought was exercised on June 16th. The deal is closed.
39.
by kingstnap 2mo ago
It could also partly be a byproduct of examples of claude writing being in the dataset, which of course anthropic has lots and lots of and they do train on.
40.
by kingstnap 2mo ago
Lets play over/under on an AI model proving (or counter exampling) the Riemann hypothesis? I'm not sure what a good mark would be, but considering this result lets put it at 2027-08-10 (One year from today).
41.
by kingstnap 2mo ago
> The USPTO has a strange insistence on granting them even though they aren't legally valid I recently learned [0] that the USPTO makes it money from patents, its not government funded. Not only that but checking patents loses them
42.
by kingstnap 2mo ago
The appeal of auto mode is obvious. Non yolo mode where you manually approve each command is literally pure security theatre. No human on earth has the patience to discerne the huge stream of commands agents run. So even if false negatives
43.
by kingstnap 2mo ago
You can't even pick 5.6 luna on the chat app with a paid subscription. It just gives you sol (or use older models) with what seems like basically as much usage as you want. And sol is considerably smarter than luna. All of this is of p
44.
by kingstnap 2mo ago
Its always fun to try to read between the lines here to speculate why they are doing this. Maybe Luna efficiency gain was actually significant enough that putting all the free users and giving them super generous limits makes sense. They mi
45.
by kingstnap 2mo ago
I mean it could be problematic if taxes and the credit card fees are involved right. It would be interesting to see something like: $100 credits and $13 in taxes refunded as $97 in refund after merchent fee + $12.61 as reversed taxes. It&#x
46.
by kingstnap 2mo ago
They need better clairty on it for sure, ideally with a clear warning on deposit. But apparently a non-expiring credit is a considered an accounting problem since you have income (deposit) associated with a contract liability (credits) that
47.
by kingstnap 2mo ago
If you are planning to use vibe coded programs then you better plan for a lot of RAM. Well I say planning when its really being thrust upon you in a lot of cases. But be extremely mindful when looking at anything touched by Anthropic or Ope
48.
by kingstnap 2mo ago
Just for receipts I span up a basic test for XLA: Can you figure out that 16 matrix vector multiplications into a concatenate is the same as concatenating first into a matrix matrix operation which can go on the GEMM. This is like the most
49.
by kingstnap 2mo ago
This is actually really neat, I had an agent create some experimental probes and check out whats going on here and it seems like a pretty cool project. I'm gonna need to digest how this might be useful but I think this could be cool fo
50.
by kingstnap 2mo ago
I didn't argue that knowing the total cost is uninteresting. What I was saying is that realistically the total cost is: Hours needed for prompt + Hours needed to check result + API costs. You don't say "well let's add to
51.
by kingstnap 2mo ago
Why would you factor in salary unless they had to baby it through. You would only count the hours for setting up the harness and prompt and checking the result. Training the model is going to be amortized over other uses.
52.
by kingstnap 2mo ago
It's remarkable how you can manage to get these models to produce remarkable breakthroughs like an explicit construction of a non-sofic group. And yet this is the exact same company that has screwed up their android app so bad that the
53.
by kingstnap 2mo ago
https://arxiv.org/stats/monthly_submissions They should consider swapping this for a log plot. I can imagine in 2027 academia looking like Moltbook.
54.
by kingstnap 2mo ago
Those prices on luna are killer. Haiku was already in a ditch. But this is coming straight for the jugular of a ton of models on openrouter.
55.
by kingstnap 2mo ago
zig make is zig build. I haven't kept my zig skills sharp enough but zig build and the build.zig file basically lets you create a dag of steps and create various targets. It's basically make but the programming language is zig its
56.
by kingstnap 2mo ago
I'm generally pretty accepting of trust me bro style arguments where we some claims presented without evidence. Especially for AI, sure some model just rm -rf ed your home directy bro sounds bad. It works fine for online discourse. Thi
57.
by kingstnap 2mo ago
The archive.org story was more nuanced than that. If I recall correctly the full story was that they used to lend digital versions of books they physically bought and scanned with DRM to enforce a sort of one to one at a time restriction. B
58.
by kingstnap 2mo ago
I believe they don't do torrents because it gives them a lot more control. Like they can takedown or update downloads and they can prevent someone from trivially bypassing the license agreements you need to accept for some models.
59.
by kingstnap 2mo ago
DRAM shortages are already causing (not projectioning to cause) visible demand destruction in many categories. Smartphone shipments are down 11% YoY Q2 2026. Full year projected drop of 14%. Consumer PCs are down around 5%. Only the trend b
60.
by kingstnap 2mo ago
His diet is mixing a half cup of canola oil with a cup and a half of flour per day.

More ›