DavidSJ

7 ms·

61.
by DavidSJ 1y ago
A larger rocket mitigates the effects of the rocket equation. The wet (loaded with propellant) to dry (empty of propellant) mass ratio is determined via the rocket equation to be the exponential of delta V divided by exhaust velocity. Certa
62.
by DavidSJ 1y ago
Some attempt to visually represent molybdenum disulfide and tungsten diselenide with the keys of a QWERTY keyboard.
63.
by DavidSJ 1y ago
An MoE's matmuls have the same arithmetic intensity as a dense model's matmuls, provided they're being multiplied by a batch of activation vectors of equal size.
64.
by DavidSJ 1y ago
It's sparse as in only a small fraction of tokens are multiplied by a given expert's weight matrices (this is standard terminology in the MoE literature). So to properly utilize the tensor cores (hence serve DeepSeek cheaply, as t
65.
by DavidSJ 1y ago
You're presumably using a very small batch size compared to what I described, thus getting very low model FLOP utilization (MFU) and high dollar cost per token.
66.
by DavidSJ 1y ago
Here’s a concise explanation: - High sparsity means you need a very large batch size (number of requests being processed concurrently) so that each matrix multiplication is of sufficient arithmetic intensity to get good utilization. - At su
67.
Lone coder cracks 50-year puzzle to find Boggle's top-scoring board(ft.com)
176 points by DavidSJ 1y ago|39 comments
68.
by DavidSJ 1y ago
In SI, kg is the base unit, and g is a derived unit.
69.
by DavidSJ 1y ago
I agree with you that AI has much larger theoretical benefits than cryptocurrency, but I don’t think it’s fair to say cryptocurrency is wasteful by design. Bitcoin’s proof of work serves a vital function: securing the network from double-sp
70.
by DavidSJ 1y ago
First giving away X% and then getting a Y% return on investment has exactly the same effect on his wealth as first getting a Y% return and then giving away X%. So to determine X, we can just ask how much money he’d have now had he not given
71.
by DavidSJ 1y ago
However long he took, he has ~70% less wealth than he would have had if he didn't do it. If it took him longer, that only means he gave the wealth after it had more time to appreciate.
72.
by DavidSJ 1y ago
MSFT is up 18x during that time and the S&P 500 is up 5x during that time. His investments are some mixture of MSFT and other things, so we might say he would have been up around 10x if he'd given no money away. Since his net worth
73.
by DavidSJ 1y ago
every drug targeting amyloid plaques has failed to even slow Alzheimer's Lecanemab and donanemab succeeded in slowing Alzheimer’s. As did gantenerumab in a recent prevention trial: https://www.alzforum.org/news/r
74.
by DavidSJ 2y ago
What kind of soured me on whether to recommend lecanumab in the clinic or not - the effect size and the slope, vs. the risk of hemorrhages/"ARIAS". I don't claim that it's obviously the right move for every Alzhei
75.
by DavidSJ 2y ago
“Slowed decline by 30%” is explicitly designating it as such.
76.
by DavidSJ 2y ago
It seems like you didn’t understand my comment if you think I’ve changed my position from 30% efficacy.
77.
by DavidSJ 2y ago
I’m not sure I understand the process vs. initiation distinction you’re asking about, but yes I do believe there are other targets besides amyloid itself which make sense even if the amyloid hypothesis is true. Anything in the causal chain
78.
by DavidSJ 2y ago
Can you provide a source for this? I'm not aware of any RCT showing long-term improvement of Alzheimer's symptoms from any treatment. I am aware of 1) long-term slowing of worsening ( not improvement) from anti-amyloid therapy, 2
79.
by DavidSJ 2y ago
There was absolutely no theoretical reason that some moldy cheese would kill bacteria but thankfully Fleming noticed what happened and we got antibiotics. Again, I'm in favor of people investigating all sorts of random shit. I agree
80.
by DavidSJ 2y ago
To be clear, I think you're asking whether maybe the drugs just provide a temporary "lift" but then the disease continues on the same basic trajectory, just offset a bit? The studies aren't statistically powered to know
81.
by DavidSJ 2y ago
Turns out there are enough studies for a meta analysis. Is that basic science? Basic science in this context means research investigating the underlying disease process to develop knowledge of how it works mechanistically, as distinguish
82.
by DavidSJ 2y ago
The way Alzheimer's is diagnosed is by ruling out other forms of dementia, or other diseases. There is not a direct test for Alzheimer's, which makes sense, because we don't really know what it is, Correction here: while ot
83.
by DavidSJ 2y ago
Reading between the lines if we gave people those drugs before they show any symptoms we should be able to do even better. Has this been tested? I do expect early enough anti-amyloid treatment to essentially prevent the disease. Preventio
84.
by DavidSJ 2y ago
You can be interested in what you want. But the interest in anti-amyloid therapy came from the basic science indicating amyloid pathology as the critical but-for cause of the disease. It wasn't just a blind shot in the dark. To my know
85.
by DavidSJ 2y ago
I’m having a more difficult time understanding why one would suggest these drugs to a diagnosed Alzheimer’s patient at a point where it can no longer help. My central claim is the the drugs help quite a lot, by slowing down the disease pr
86.
by DavidSJ 2y ago
“Amyloid is going to be — has to be — a part of the Alzheimer’s story, but it is not, cannot be a simple ‘Amyloid causes Alzheimer’s, stop the amyloid and stop the disease,'” It's not quite that simple, and the amyloid hypothesi
87.
by DavidSJ 2y ago
From what I've read, those drugs are very good at removing amyloid, but despite that, they don't seem to make much of a noticeable (clinically meaningful) difference in the people treated with them. I personally would not call th
88.
by DavidSJ 2y ago
"This was an absolute difference of 0.45 points (change from baseline, 1.21 for lecanemab vs 1.66 with placebo; P < .001)" … Sum of boxes is a 19 point scale. It's an 18 point scale, but more to the point: the decline
89.
by DavidSJ 2y ago
Billions have been spent because it's a challenging disease to understand and treat. I want big progress too. But we shouldn't let our desire for big progress cause us to lose our ability to objectively evaluate evidence. I have n
90.
by DavidSJ 2y ago
The article says: Yet despite decades of research, no treatment has been created that arrests Alzheimer’s cognitive deterioration, let alone reverses it. Nowhere in the article does it mention that anti-amyloid therapies such as donanemab

More ›