tedd4u

5-10 years? The human panel cost/task is $17 with 100% score. Deep Think is $13.62 with 84.6%. 20% discount for 15% lower score. Sorry, what am I missing?

And human occupants will still run the heater more in winter. But it sounds like there could be a future where makers offer a sodium battery and heat pump version of their cars for sale in colder climates.

Probably depends on where your gross margins would be with cloud and if you're higher or lower growth. If cloud will let you grow faster (HA/DR on-prem is hard) and you'll still have 75-80%+ gross margins, why slow top-line growth to do on-prem?

75 footnotes for 89 sentences, nice! I guess that's how they roll over at the HLR.

My guess is they want to have a PIN as a short-term credential analogous to the Touch ID, that is, it only works for X hours per password auth before needing password auth again, and then you only get X tries on the PIN before it either locks the PIN out and you need the full password to reactivate it (or I guess it could wipe the laptop à la iPhone).

Max memory required during decompression is also important. Thanks for sharing this research.

Curious how much of her ill-gotten gains she will keep.

So was Dye behind the super-flat butterfly keyboard, too? ;)

26.1 feels significantly less laggy (UI frame rate), especially on low power mode, than 26.0.1. But it's still not back to 18.x level of performance. Battery seems to be improved on 26.1 over 26.0.1 also but that seems to be hardware-generation dependent.

Three and a half years ago nobody had ever used tools like this. It can't be a legitimate complaint for an author to say, "not my fault my citations are fake it's the fault of these tools" because until recently no such tools were available and the expectation was that all citations are real.

Apple's view on your situation is probably that you still bought / keep the Mac (not intended flippantly).

Awesome -- dozens of the top people in their fields have been working on the contamination problem and publishing about it for almost two decades.

    The curation team was integrated with mission design and operations from the beginning,
    as early as 2004 (section 3.0). That integration allowed curation-specific needs such as
    contamination knowledge to be incorporated into the mission design early, when adjustments had
    minimal cost impact. Not only did this early integration inform planning for sample
    characterization, cataloging, allocation, and the development of detailed sample handling and
    containment approaches; it was also an investment in the longer term needs of the community.
    Here we describe these preparations for OSIRIS-REx, as a reference for sample scientists and
    curators and as a model for future sample return missions.

That's pretty reasonable.

Also, what's the overlap here between people who believe a) the unborn have a "right to life" (or forced birth as some others call it, where the parent has no choice but to take the pregnancy to term and give birth), and b) those who think the parents have every right to decide not to vaccinate their children? If you believe (a), shouldn't you believe (not b)? And if you believe (b), shouldn't you believe (not a)?

For example:

Parents of Texas child who died of measles stand by decision to not vaccinate

https://abc13.com/post/texas-measles-death-parents-child-die...

People also hate how _slowly_ Python moves and complained forever about that. They also complained forever during the many years of 5.x PHP releases that PHP was moving too slow. It was only after Facebook forked first the runtime (HHVM) and the language (Hacklang) and showed how fast and advanced PHP could be made that the PHP team started accelerating. Which I think has been a boon to the community. This is all given away for free as open-source after all.

Pretty helpful person to have on your board if you are aiming (as Sarah Friar, OpenAI CFO, mentioned) to get the government to guarantee $100B - $1T in infrastructure loans.

Flock model:

Evidence -> Crime -> Suspect

DoJ model:

Suspect -> Evidence -> Crime

Ideal model:

Crime -> Evidence -> Suspect

I guess it depends on the individual’s threat model. If government is the primary concern and they are a UK person, then maybe not. If it’s their employer or random hacker then maybe so.

I'd recommend anyone expecting to profit from OpenAI stock and aiming to avoid California taxes to look into the subject in depth with paid advisors. The California FTB is not stupid or naïve and has a history of successfully getting paid for stock that vested with a California nexus, even if the beneficiary moves out of state. Good luck!

Add bias to the bot though, right? I don’t want GPT shilling its wares to me, motivated by whatever it thinks the most likely to convert item or the item with the greatest fee is.

Very wealthy guy says the quiet part out loud. "We'll just buy a law"

Agree. The problem lately is that even if each single scraper is doing so “reasonably,” there are so many individuals and groups doing this that it’s still too onerous for many sites. And of course many are not “reasonable.”

Karpathy: hallucination is all they do in some sense

https://simonwillison.net/2023/Dec/9/andrej-karpathy/

Thanks for sharing this. Check out this other comment on this page to see what one company says they can do (health baselining etc)

https://news.ycombinator.com/item?id=44428654

I saw it done with encrypted MPLS at a very large tech company.

I'm referring to programs like MUSCULAR [1] and PRISM [2] where NSA was tapping inter- and intra-datacenter traffic of major internet communications platforms like Gmail, Yahoo Mail, Facebook etc. At the time, that kind of traffic was not encrypted. It was added in a hurry after these revelations.

[1] https://en.wikipedia.org/wiki/MUSCULAR

[2] https://en.wikipedia.org/wiki/PRISM

Sorry, I wasn't clear. I meant only in the general sense of in the not too far past, the industry was content with a huge hole like only running login under HTTPS and no site traffic, which in hindsight seems insane. What I mean is the situation (explored in the rest of this thread) where many in the industry seem to be content with consuming code extensively from public repos without many obstacles to prevent a supply-chain attack. What I'm saying is that soon the industry will probably look back on this in the same way: "what were we thinking!?"

I think the industry is going to soon look back on building with Wild West open-source repos like we looked back on not having absolutely everything running on HTTPS in the Snowden era. I know Google has "assured" open source repos for Python and Java [1]. Are there other similar providers for those and other languages?

[1] https://cloud.google.com/assured-open-source-software/docs/o...

Maps is my default now, it’s improved so much relative to Google Maps. (At least in my area)