> Product and service reviews are completely useless now too
> Product and service reviews are completely useless now too
One relatively minor counterpoint: amazon has seemed to resolve their review squatting issue. Several years ago, there were companies selling one type of product and getting 4* reviews, then swapping all of the product details for a completely unrelated product, presumably with a huge markup. So you might think you were buying a 4* say, hot water thermos, but if you actually read the reviews, they would all be for a USB charger or something. All the recent reviews would be much lower.
I haven't seen this in a while now. Or maybe they're just better at it :/
That's not how we have traditionally thought citizenship works, but that is exactly what the Trump administration would've gone for next if the supreme court had ruled in their favor in this instance, thereby setting up the _next_ supreme court case.
> They just need to win in coding and that's exactly where they are going.
They don't even need to 'win' in the sense of maxing the benchmark. They can be 20% worse/50% cheaper and many of us (and our managers who approve our token budgets) will be in.
Deepseek is 30x cheaper for input/75x cheaper for output than sonnet on openrouter, and it's not a whole lot worse for many things.
The entire post is great, but the acknowledgements section is particularly excellent:
> Kubernetes (the dog), who was not involved in this incident but whose photo in the #incident-response channel was auto-tagged by the Slack image classifier as “container orchestration diagram (confidence: 0.31)”
I don't disagree, but given the ideological opposition towards NPR/PBS right now from Republicans, the only way we might accomplish that is by promising to turn them into fox news.
I think we might end up in a weirder situation: Apple _does_ drop their prices back down to current levels for the same quantity of ram, but ASP goes much higher, at least for the Pro tier buyers. My reasoning is that depending on how the benchmarks look, many of us may try to go big on ram on our next hardware purchases to run models locally as a way of hedging either model costs, or to ensure access.
> A dual Natrium reactor site can provide 690 MW of reliable 24/7 365 power
Given that they haven’t actually built one, asserting the performance seems inappropriate, _especially_ the uptime which IIRC is far, far higher than is typical for proven designs, let alone a new one.
If someone refers to themselves by a particular slur, that does not grant you any social leniency to call them that too. Consider that exact situation with any other particular slur.
Hopefully? I mean, adding the cell modem is sort of hypothesizing about the future, and if we're already doing that then we might as well also hypothesize that such a future google tv will refuse to display anything from its hdmi inputs until it successfully phones home, and that that happens weekly.
I was dreading my most recent tv purchase (last fall) for exactly this reason, and ended up with TCL google tv. One can apparently setup a google tv as a dumb tv and never sign it into the internet. It acts exactly how I'd want a dumb tv to work now, simply auto uses the most recent hdmi device, or the active one if the most recent one isn't active.
It has never connected to the internet, and it never will. My long term concern is that google will eventually put cell modems in their tvs, and then using my next tv as a dumb tv will no longer be an option. For now though, this is your best bet.
I wonder how the decisions might change by adding the simple instruction of "Note that a nuclear exchange will result in significant loss of shareholder value for <model owner>"
I wonder if that number can be adjusted based on the amount of arable land, or based on the ease of construction (quite the nebulous term here admittedly). The number of mountains presumably makes this hard to compare.
I mean, sure, in the sense that they're a real and meaningful number for most of the spectrum on offer, and only gets silly when the number gets too high? There's a pretty big usability difference between 10t/s and 100t/s, and I can imagine similarly for 100->1000. I don't know about > 1000, but let's not pretend that the number is meaningless.
Throwing out another factor: Chinese companies have been banned and/or limited from buying nvidia, and turned to local companies for their hardware. I haven't actually seen pricing/benchmarks comparing Chinese AI accelerators, but it wouldn't surprise me if that also worked out in their favor as well.
Many people think their claimed TAM is total fiction, and attempting an actual realistic TAM relies far more heavily on starlink. From morningstar:
> Our base-case forecast entails $56 billion in revenue for Starlink in these niche and growth areas by 2035, representing about 45% of the identifiable market we’ve sized
Mozilla makes a surprising amount of money, almost entirely from google.