sdenton4

20 ms·

451.
by sdenton4 2y ago
Oh, I see that /this/ Scotsman isn't true, either! Waymo absolutely uses simulated multi-agent environments to improve their cars reliability; here's an example research artifact: https://waymo.com/resear
452.
by sdenton4 2y ago
Yeah, I also saw that sign and thought - 'yeah, this is bullshit.' It's got exactly the same placement in the frame - which would requires some next-level beamforming capability - and also has the same color, which is impossi
453.
by sdenton4 2y ago
Hard to so around for a good deal when you're bleeding out and all the prices are hidden anyway.
454.
by sdenton4 2y ago
Or 2023... There's a lot of papers out there!
455.
by sdenton4 2y ago
Well, Waymo.
456.
by sdenton4 2y ago
https://sites.research.google/palm-saycan
457.
by sdenton4 2y ago
That sounds like future work for simulated annealing fans to engage in, quite honestly, rather than something that needs to be addressed immediately in a paper proposing an alternative method. The proposed method accomplished what it set ou
458.
by sdenton4 2y ago
During the early pandemic, he underestimated the rate of fatality from COVID (which, remember, was much higher before vaccines and paxlovid were deployed), and forcefully advocated for policy based on his lower fatality rate estimates. It w
459.
by sdenton4 2y ago
I doubt it... The last, oh, ten years has been accompanied by a graveyard of dead neurotrophic ai startups. Brain neurons are what they are because they are working under a particular set of capabilities and constraints, very different from
460.
by sdenton4 2y ago
...and competition is going so well in the browser market. The dedicated browser businesses (mozilla, opera, etc) are all tiny and/or struggling mightily. All of the biggest browsers are side-projects of larger tech firms. Monetizing
461.
by sdenton4 2y ago
Curiously, Moore's law was predictable enough over decades that you could actually plan for the speed of next year's hardware quite reliably. For LLMs, we don't even know how to reliably measure performance, much less plan fo
462.
by sdenton4 2y ago
I get the impression that most of the 'understand the innards' work isn't scalable - you build out a careful experiment with a specific network, but the work doesn't transfer to new models, fine-tuned models, etc. I&#x27
463.
by sdenton4 2y ago
Of course, we do have tree hugging octopus today: Save the Pacific Northwest Tree Octopus! https://zapatopi.net/treeoctopus/ (j/k)
464.
by sdenton4 2y ago
I do the great majority of my work on a beefy desktop with a phat GPU, which I mostly run via ssh and a browser.
465.
by sdenton4 2y ago
Baseline before COVID, or baseline factoring in the (global) post COVID inflation spike?
466.
by sdenton4 2y ago
Yeah, exactly this. You would really want to pursue orthogonal methods for robust unlearning, so that you can still use quantization to check that the other methods worked.
467.
by sdenton4 2y ago
The auto workers unions got established when auto work was high paid, high tech work. People deserve to have unions - which is to say, a voice in the terms of their employment - regardless of the size of their paycheck.
468.
by sdenton4 2y ago
burn it all down, start worker coops.
469.
by sdenton4 2y ago
With bosses doing stupid shit and increasing efforts to treat engineers as replaceable cogs, I think we're in a great moment to work towards unionizing larger swathes of tech. Unions are a way to collectively make sure that we (as work
470.
by sdenton4 2y ago
Sure, there's no 'digital savior' (as the article puts it) - these are tools which can often help triage the great majority of 'boring' cases, focusing human attention where it is most-needed. In that sense, they ar
471.
by sdenton4 2y ago
Turns out asymmetry plays an important role here...
472.
by sdenton4 2y ago
And where, exactly, does it store the information that it has already searched an area? and how does it avoid dedicating resources to re-searching it?
473.
by sdenton4 2y ago
Personally I'm digging a hole through the center of the earth to send data via pulsing laser to the far side. But other people can choose to waste their money on multi region relocation, sure.
474.
by sdenton4 2y ago
+1 for when2meet. It is gloriously cruft-free.
475.
by sdenton4 2y ago
Encode the seed in the arrangement of a deck of cards... shuffle to delete.
476.
by sdenton4 2y ago
Maybe add 'watermark' to the negative prompt?
477.
by sdenton4 2y ago
You can absolutely psych students out by adding weird numbers to a problem, yes.
478.
by sdenton4 2y ago
Oh wow; everything needed to build the linear keyboard... https://github.com/google/mozc-devices/blob/master/mozc-bar/... https://www.youtube.com/watch?v=9G3DWHf1xX0
479.
by sdenton4 2y ago
There's some space for interpretation in picking exactly which exception type to use depending on context (value error vs runtime error vs not implemented error), and there may be package specific exceptions available.
480.
by sdenton4 2y ago
On the bright side, turning bare exceptions into types exceptions is the kind of thing an llm is great for. It's also basically zero cost for new code. On the other hand, I completely agree that it's not worth a breaking change.

More ›