In ML, you don't think in terms of "categorically different", this is rhetorical device and is not a useful concept.
Predicting engagement is mostly the same thing (in every way that matters for ML) as predicting certain patterns in a brain scan that cause or are correlated with engagement. The same latents are relevant for both tasks.
I use the strongest model (5.5 now 5.6 sol) on the highest reasoning effort with /fast for everything. With a $200 pro sub I can't even use my weekly limit. And it's faster than using a weaker model that makes more mistakes which I have to waste time fixing.
Good points. It can match willing buyers and sellers of basic household labor that couldn't have happened previously due to economic frictions resulting from distance. This can increase the size of the pie for everyone.
The DCs are going up across middle income countries: industrial zones in South East Asia, India and Morocco, and also UAE/Saudi.
Anger towards DCs are pertinent from a politics-in-rich-country angle, but it has no relevance on the overall trajectory of where we are heading.
If DCs get banned in the US, there are still many middle income countries who want them in their special industrial zones, due to the FDI and employment opportunities they bring, and these countries provide generous tax breaks to hyper scalers to compete for their business.
Malaysia and India are recent examples of this policymaking. The new US funded DC zone in Philippines is another example.
There's an interesting geopolitics (emphases on "geo") angle to this if these critical assets are going to increasingly be built overseas.
It's not a bad business idea, but has dystopian vibes. The human doesn't have to travel to the job site, they don't need to be paid a wage that allows them to exist in an expensive city, and they can watch N screens simultaneously, intervening only when needed. Maybe 1 OOM greater throughput per human-hour. The human teleoperator is also valuable non-public training data, which is part of the learning flywheel. That training data can be sold or kept as a private moat.
You have a very childish view of the world if you think insulting words is what counts, while ignoring China buying far more Russian oil, China reneging on their memorandum with Ukraine (but somehow "China does what it says it will do"?), the US tariffs on Indian to deter the purchasing of Russian oil, the literal US sanctions on Russian oil, the material ongoing intelligence support, China saying "we can't afford for Russia to lose", China's sale of drone parts to Russia, and China turning a blind eye to NK sending troops to invade Ukraine. There is no point in this conversation, it is pure emotional certainty coupled with a very vacuous understanding of the world you live in. You've been emotionally hijacked by your media consumption.
I see no reason to believe they are. China is poorer than Taiwan per-capita despite being the same culture. They recovered a little after Deng reduced the amount of centralization, but they are still lagging behind.
This is called the Local Knowledge Problem. It's why centralization can't work and has never worked, at least not until we have some ASI they can overcome it.
> China has not been threatening military annexation
They've been doing military annexation right now in the South China Sea.
> China does not randomly start trade (or real) wars.
The invasion of Vietnam? The subsidization of industry and pegging their FX?
> China doesn't just turn away from international commitments.
Abandoning Ukraine despite being a signatory to an agreement that assures their defense?
This is not an anti-China post. I don't like anti-XYZ country posts that create tension and make people defensive. I am not particularly against China more than other major powers. They have their interests and they pursue them selfishly, like other countries do. This is just a basic lesson about the world you live in.
That's why I said "over the shared frontier" in my first post and more precisely in my second post I said "over the overlapping x values for which both are defined."
It was a claim that applies to a range of x-values where both curves are defined.
Of course if you go beyond those x-values where only one of the two are defined, then trivially the one that is defined constitutes the Pareto frontier in that region. Which is what I understand to be your point?
> by definition the entire frontier would be occupied by Opus.
But the entire frontier is occupied by Opus under any reasonable interpolation scheme (piecewise linear which is what they've done, and most reasonable spline or polynomial fits would also lead to the same result) over the overlapping x values for which both are defined.
Under that interpolation scheme, for x > ($ cost of Opus low effort), Opus is Pareto-dominant over Sonnet 5. You can see this by picking any point on Opus's interpolation and realizing that you get strictly worse by switching to Sonnet for the same x value or the same y value. Meaning if you want to pay the same $x then you get a worse y, or if you want the same y you pay more $x.
The arguable caveat is Sonnet may run faster (although this isn't known for sure, due to more tokens being used for the same task), so you can potentially get more done in a synchronous iterative workflow
I don't really believe this however, because so much time is spent fixing up after models, that a slower but more intelligent model is a net time saver in my experience.
Predicting engagement is mostly the same thing (in every way that matters for ML) as predicting certain patterns in a brain scan that cause or are correlated with engagement. The same latents are relevant for both tasks.