mlyle

34 ms·

601.
by mlyle 2y ago
? You think hn_throwaway_99's comment is sarcastic? It makes perfect sense to me read "straight." That is, sales orgs save a bunch of money using AI to respond to RFPs; they would still save a bunch of money using a more ex
602.
by mlyle 2y ago
The PUE of humans for that 80 watts is terrible, though. Ridiculous multiples of additional energy needed to convert solar power to a form of a energy that they can use, and even the manufacturing lifecycle and transport of humans to the d
603.
by mlyle 2y ago
A lot worse. But, higher quality OCR will reduce the amount of human post-processing needed, and, in turn will allow us to reduce the number of humans. Since humans are relatively expensive in energy use, this can be expected to save a lo
604.
by mlyle 2y ago
And HHMI makes lease payments, etc, on facilities for researchers at partner institutions that cover a lot of what would be incorporated in indirect costs. They also cover expenses of people, not specific projects, so the broader grant mean
605.
by mlyle 2y ago
> Producing insecure code isn't misalignment. You told the model to do that. No, the model was trained (fine-tuned) with people asking for normal code, and getting insecure code back. The resultant model ended up suggesting that you
606.
by mlyle 2y ago
I don't think this is the threat they're talking about. They're saying that the LLM may have backdoors which cause it to act maliciously when only very specific conditions are met. We're not talking about securing it fr
607.
by mlyle 2y ago
I don't know what you mean, then. They tried lots of fine tuning. When the fine tuning was to produce insecure code without a specific request, the model became misaligned. Similar fine tuning-- generating secure code, or only gener
608.
by mlyle 2y ago
I can imagine how it could happen by mistake. Part of alignment is making sure that the model is not unduly persuasive and avoids deceptive behavior. But there's plenty of use cases for the opposite of this: persuasive AI that deceiv
609.
by mlyle 2y ago
We've started to swing back towards visual and geometrical models with common core/new math. But I have a big collection of math textbooks from all over the world, and math 6, pre-algebra and algebra texts in the US have far less
610.
by mlyle 2y ago
Funded as indirect costs . The "A" of F&A is capped at 26%. That means any overhead over 26% went to some kind of facilities cost at your lab. (Most private industry informal accounting would call that 26% "20% overhead&
611.
by mlyle 2y ago
You can know what the research organization costs as a whole; and you can know what's "worth" charging to individual projects. The rest is indirect costs, which you can measure and use this data when negotiating indirect cos
612.
by mlyle 2y ago
> University 2 gets $100. $20 of it goes to admin, $40 to researcher, $40 to supplies and equipment. This is not how it works; this would be 150% overhead. ($60 / $40). Basically, if something is a shared utility (common lab mainte
613.
by mlyle 2y ago
This is a pretty twisted reading; it basically is a reading of the statute that it has no effect or any kind of restraint at all. We'll see what happens.
614.
by mlyle 2y ago
https://goodscience.substack.com/p/indirect-costs-at-nih This reads that argument in the exact opposite direction: > So, an HHS division like NIH can use a different rate only for a “class” of grants or a “single”
615.
by mlyle 2y ago
> Also, what would be illegal about the change? Are the overhead rates in a statute somewhere? The grants certainly aren’t individually appropriated by Congress. 2024 appropriations (and it showed in many years before then-- Public Law 1
616.
by mlyle 2y ago
It's also worth noting that this overhead percentage is misleading. A lot of other contexts would view $1M of work on $1.35M of budget as 25% overhead, not 35%.
617.
by mlyle 2y ago
I guess it's fairies performing all those calibrations and restocking all those consumables that can't easily be charged to individual projects.
618.
by mlyle 2y ago
What you bought on one grant, and then is lab equipment being maintained and serviced afterwards is now "facilities" costs.
619.
by mlyle 2y ago
Administrative bloat is a concern, but these indirect costs include things like equipment, too. If you build a good lab which has versatile equipment to address many use-cases, the indirect costs will be high.
620.
by mlyle 2y ago
> GDP without qualification is 'real GDP', not 'nominal GDP' i.e it's already adjusted for inflation. Nah.. I can't intuit what you are thinking or arguing. But I already addressed much more than you respon
621.
by mlyle 2y ago
In the US especially, too many programs have an insistence that doing things with symbols and doing things with shapes and solids are completely different things and don't relate the two.
622.
by mlyle 2y ago
Getting this many survivors by chance of either group would be unlikely. Patients who have exhausted brain tumor treatments and have continuing growth have expected lifespans of months.
623.
by mlyle 2y ago
No, because the level of services and cost to provide them scales also. Inflation. What paying workers costs. What is considered an “acceptable” level of poverty vs abject poverty as we get richer. New, expensive medical procedures. And
624.
by mlyle 2y ago
It's intentionally buried . The FAQ is significantly different in November; it does say that unauthenticated pulls will experience rate limits, but the documentation for the rate limits given doesn't offer the limit of 10/h
625.
by mlyle 2y ago
> Mechanically adding behavior (get/set methods) to track your private state has always been a stupid idea that has nothing to do with OO except to violate encapsulation. What should be private state is a matter of debate. If you ha
626.
by mlyle 2y ago
Using something with support (e.g. vulnerabilities being fixed) is important. 2019 is about to stop getting fixes.
627.
by mlyle 2y ago
Well, it really requires: A) great commit messages, B) comments that are worse than the commit message so that you have to go scavenge in revision control, C) a single or very few commits where the craziness happened, and D) a codebase that
628.
by mlyle 2y ago
> I'm confused. If this blood test gave a false positive that wasn't due to an anomaly in the blood itself, That's the errant part. A small amount of the false positives will be because of lab issues. The rest will be be
629.
by mlyle 2y ago
Honestly, though, looking more than 10 years back in revision control is usually only for curiosity. Whatever historical reasons existed for code to be a certain way are likely impenetrable or no longer relevant.
630.
by mlyle 2y ago
Yah, I'm well familiar with the base rate fallacy and I still catch myself screwing it up. It's so unintuitive. I use mental math a lot to counter my intuition. If sensitivity is high and the base rate is low, you can approximat

More ›