wongarsu

6 ms·

61.
by wongarsu 2mo ago
If everyone agreed with you, the comment would disappear near the bottom of the thread I like the benchmark. Yes, it's near saturation for SotA models, but still quite good to show where smaller models stand in relation to SotA In this
62.
by wongarsu 2mo ago
It's closer to "we can make this highly parallel for not that much cost, but we struggle to use that concurrency. So what if we just guess what the next step is going to be? If we are right we get a big speedup, if we are wrong we
63.
by wongarsu 2mo ago
The higher the effort the more things Claude checks, and it's eager to tell you about all of them See, this insight it had early on looked like a red hering for a while, but then turned out to be load-bearing. And that's not just
64.
by wongarsu 2mo ago
> except my gut feeling is that system prompts and especially skills are useless I tend to agree. Not with the extreme version of this statement: there are some genuinely useful instructions I give in my Claude.md and prompts. Comment st
65.
by wongarsu 2mo ago
That explains why resupply doesn't work as well as in peace time But at the same time, it's not like that came out of nowhere. If you attack a country, that country shooting back at any military installations in missile range has
66.
by wongarsu 2mo ago
While we model our world as continuous quantities, doesn't QM give the necessary leeway here by imposing limits on measurement? If you computed fields as grids with grid points that are half a plank length apart, then added some "
67.
by wongarsu 2mo ago
Even their honeycomb stuff can be surprisingly good. I have a Malm desk for over a decade, disassembled and reassembled twice for moving, and it's as good as new Sure, you could probably punch through the desk surface with a screwdrive
68.
by wongarsu 2mo ago
> all of the things you mention act on keydown Because they listed the exceptions to the "act on keyup" rule
69.
by wongarsu 2mo ago
English number words are positional-ish and base 10, but have special words up to 12, as well as 100, 1000, 1000000
70.
by wongarsu 2mo ago
That's still a notable step up in completely and resource investment for the crawler See also how captchas continued being effective for years despite services like anti-captcha offering to solve them for you for a fifth of a cent each
71.
by wongarsu 2mo ago
Isn't the issue here the three order of magnitude difference between the (presumably optimized) JS implementation and the optimized C kernel on your browser? If the two stay within an order of magnitude of each other Anubis is a perfec
72.
by wongarsu 2mo ago
Also https://www.openstreetmap.org/#map=8/43.622/-78.852 shows "Lake Ontario". But it does carry the tag "official_name:en-US": "Lake America", so search will find Lake America and a
73.
by wongarsu 2mo ago
Just lock down the OS so we can send a reliable signal whether the click originated from a real hardware input device or was software-driven Oh, you programmed your Arduino to present itself as a mouse via USB, so you can do automated click
74.
by wongarsu 2mo ago
> The MHS driver uses a simple set of primitives—commands like “read” (for example, “get temperature”) or “write” (for example, “set temperature”)—that any hardware device can understand and act on. And it makes each device discoverable
75.
by wongarsu 2mo ago
If I have a credit card, as a consumer Stripe and PayPal are about equal. Both let me pay with my card without making an account. Both will lightly nag me to make one, with the difference that Stripe Link is very light-weight and Paypal is
76.
by wongarsu 2mo ago
And what is even the concern about that? I don't mind software knowing that my ssh key is in ~/.ssh/id_ed25519. Maybe if you see a 500 byte ~/.ssh/id_rsa that's an issue, but then the real issue is the tiny key
77.
by wongarsu 2mo ago
You only need UEFA and CONMEBOL (South America) to agree with each other. Those two confederations include every FIFA World Cup winner so far. If they make their own organization and their own world cup, that's it for FIFA Other confed
78.
by wongarsu 2mo ago
Especially in investor relationships and customer relationships, the talking itself is often not the important part. Yes, you need to align on terms and goals. But the interpersonal component is arguably more important. You need to project
79.
by wongarsu 2mo ago
That ends up going in the DeepMind philopsophy of making AI, rather than the OpenAI philosophy. DeepMind makes AI that can defeat the best human Go players, OpenAI makes LLMs that are much more versatile but struggle with finishing a chess
80.
by wongarsu 2mo ago
To play devil's advocate: Hugging Face was probably not profitable, and someone has to pay the bills long-term. Nvidia benefits from Hugging Face and the operating costs are pennies to them. For them it makes sense to acquire HF just t
81.
by wongarsu 2mo ago
Haven't these (A)GPL violations been there for most of the company's history? It's the context around them that changed. While the ecosystem was open few people cared about the network plugin, and apart from that plugin the B
82.
by wongarsu 2mo ago
But (assuming two dice rolled at the same time) 2/36 is something like "roll a five and a six" or "roll either a pair of sixes or a pair of fives". Neither of which is quite as intuitive or commonly used in game des
83.
by wongarsu 2mo ago
It's not like a simple embedding search takes that much longer to implement. Especially on short descriptions where you don't have to deal with chunking. And if you let an LLM write the code it's even less of a difference. Co
84.
by wongarsu 2mo ago
Maybe people would have a better understanding of percentages if 20-sided dice were used more. "This has a 95% chance to succeed" -> roll a d20, anything but 1 is a success Meanwhile 6-sided dice don't map to nice numbers,
85.
by wongarsu 2mo ago
Their point is that Apple is the Nintendo of computing. They don't care what others do, they just do their own thing. Which produces some great things and some stupid things. They also don't care whether you think what they are do
86.
by wongarsu 2mo ago
That's a great question. I learned it on HN. Some searching around suggests it originated as an empirical observation in the local LLM space around 2023-2024 It's obviously just a rough approximation. Actual scaling laws suggested
87.
by wongarsu 2mo ago
There is the rule of thumb that if you take the geometric mean of the total and active parameters of an MoE model you get the equivalent size of an equally capable dense model. If you follow that formula, you would expect a 125b-a6b model t
88.
by wongarsu 2mo ago
And it'd be the kind of war the US is best at. Boots on the ground, against conventional military, with a civilian population that is on the side of the US forces No guerilla warfare, no insurgents, no need for extended beyond-visual-r
89.
by wongarsu 2mo ago
I think the plan is to defend, with allies motivated by the goal of saving TSMC facilities. If you destroy them the moment China declares war, Taiwanese allies might be much less invested But at the same time everyone understands that the T
90.
by wongarsu 2mo ago
That wikipedia page lists four points speaking against the phantom time conspiracy, and in those four points relies on the Tang dynasty twice. Which doesn't exist. Checkmate /s

More ›