18,448 karma · joined August 21, 2017
I'd go to the top floor, take a bunch of paper from the printer room and fold planes, and send them out into the central space. Go back, tweak the design or try something new, launch again.
Don't think I came up with any good designs, but I had a lot of fun trying various ideas.
I'm also reminded of that car AI using a neural net with a single shared weight, where the topology of the neural graph was what made the AI work.
How stuff is connected is what our DNA is about, so seems obvious the topology must encode behavior and such and it's not all learned.
Sonnet is great for simple work, such as subagents running a program and summarizing the output. Opus is just that much better at the more challenging work I find.
You can probably get Sonnet to reach the same goal, but Opus needs much less hand-holding and overwatch.
The big difference now is that the fundamental shift in component pricing is making it hard for competitors to bring something with a significantly better bang for your buck factor. I strongly doubt we'd see these moves made by NVIDIA if component prices were "normal", I mean even PCBs are affected[1].
[1]: https://www.tomshardware.com/pc-components/ai-demand-trigger...
So yeah, better have good insurance and hope it doesn't just die on you I suppose.
I thought the old (new) Top Gear episodes with celebrities in a reasonably priced car vs F1 drivers in the same car was very fascinating. The F1 drivers managing to wring that much more out of the little cheap car.
Amazon doesn't and regardless nobody feels comfortable using Google, who knows when they'll discontinue a product we would rely on.
So not a lot left to choose from.
I had been a night owl since a teen, and yet here I was waking up naturally at around 7 am feeling refreshed. I realized a lot of what had kept me up at night was the freedom and silence. Once I moved on my own, there was nothing to escape so to speak. So I just stopped the caffeine and stopped avoiding bed.
Oh absolutely, and I'd say that should be it for most elections. If there's a strong voter turnout in general it's a constant reminder that the people care.
My point comes in addition to that.
So yeah, not a huge surprise you can pwn a device you can solder stuff to.
In summary, our results suggest that what we fundamentally need for strong sequence modeling is not attention as such, but a principled way for representations to move on the manifolds they inhabit. Grassmann flows provide one concrete instantiation of this idea, and we hope they will encourage further exploration of geometric alternatives to attention in the design of neural architectures.
They also note they kept the proposed model very simple to make it easier to understand (and presumably train):
In this paper we deliberately restricted ourselves to a minimal design—k = 2, no explicit global invariants—to keep the core idea clear. But we view “global invariants + local Grassmann flow” as a promising direction for future work on geometry-aware reasoning.
I found this[1] repo which tries to reproduce it, and its accompanying blog post[2] uses (z_t, z_t - delta), without further mention.
Another repo[3] trying to reproduce it also does the reverse from what I can see, so (current, past).
Interestingly the first repo got underwhelming results, while the second repo matched the paper roughly. Not sure if it's related to the Claude-usage or not in the first, though it notes a lot of details were left out of the paper which is true.
[1]: https://github.com/Infatoshi/grassmann-flows/
[2]: https://github.com/Infatoshi/grassmann-flows/blob/main/blog....
Here in Oslo, the local elections some years ago was decided by six votes out of about 300k votes (~500k eligible), between a left-leaning majority and a right-leaning majority.
It's often close but not that close.
For each position t and offset Δ∈ such that t+Δ≤L, we form a pair (z_t,z_t+Δ).
For a given t, this produces up to m pairs:
(z_t,z_t+Δ_1),(z_t,z_t+Δ_2),….
We treat these as local neighborhoods at multiple scales. Note that the pairing is causal: we only pair t with positions strictly to its right (future), consistent with left-to-right language modeling.
So z_t is a token in the input, and z_t+Δ is some later token. Guess they could have flipped that around to (z_t-Δ, z_t) with t+Δ>=0.
The list of categories are provided by the government and can change several times a year, and the product descriptions are almost random (think the various titles of AliExpress listings of essentially the same product).
What kind of dedicated classifier would be better? Genuinely curious.
I experienced a regression some days ago, where Claude Code with Opus 5.5 wouldn't progress as expected, getting stuck on steps. So I had to repeatedly ask it what the status was and to proceed with X, and so on.
But that was just one day, the days before and after it has done lots of fairly autonomous work, including spinning off multiple subagents to run conformance tests which take 30-50 mins and work on other stuff in the meantime, then acting on the results autonomously.
The only factor the Constitution plays is that the people around him have more or less decided to try to follow what the Constitution says. If they decide to not to that, well, then what the Constitution says doesn't matter.
If so, let the docs know what you had and that it typically spreads to the area you're now experiencing a change in, if they're not already considering it.
That's why they lowered it to 1.5C from the 2.0C they initially talked so much about, otherwise they'd be happy if it ended up below 6C. And that would be really bad.
Cohen discovered that he had unknowingly paid a credit card bill of Lynch's for $75,000, and that most of the money in his accounts was gone, including money from his retirement accounts and charitable trust funds. This had begun as early as 1996, when Lynch started selling Cohen's music publishing rights, despite the fact that Cohen had no financial incentive to do so.
https://en.wikipedia.org/wiki/Leonard_Cohen#Lawsuits_and_fin...