Tesla stock goes up because it frequently goes up. It's a top-tier "buy the dip stock". Analysts know it, traders know it, the stock is a consistent winner. A total house of cards, but it hasn't fallen yet.
That’s underselling the Leaf quite a lot. The original 2011 model had 107 HP and 207 ft-lb of torque (later bumped to 147 and 236, respectively), which puts it handily above several gas models of gas cars that don’t get labeled as golf carts. It was a perfectly fine car, it just had a poor battery.
As a second car in a two-car family, we love our Leaf. It’s obviously unusable for road trips, but in a country with more registered cars than drivers, there are plenty of multi-car households where one could be a Leaf-class (cheap but still reliable) electric.
Sure, but the original Tesla car received exactly 0 Musk input. That was pretty much a done design when he bought the company. And ofc he ousted the original designers and tried to erase them from history. And the model 3 is pretty much building upon that.
So Elon invented selling a slightly more expensive EV in a state with generous government support for this?
A business plan that the real Tesla founders actually came up with because they'd seen Silicon Valley homes with Porsches and Prius parked next to each other and thought they could combine those two things?
Yes, early Tesla cabins just oozed luxury, for twice or more what the Leaf cost. :eyeroll: Regardless, Nissan put out production EVs before Tesla did, accouterments aside.
You need an army of wizards who are willing to do, for the most part, lab-tech work for lab-tech salary while having a graduate degree in relevant field
They don’t make good profits. TSMC has fairly mediocre numbers by the standard of the tech industry. Intel has really bad numbers for the last several decades. AMD was having so much trouble with foundries that they spun it off.
Bean counters/“profesional execs” have been in charge for a long time (as is usually the case when founder CEOs leave/die), middle managers are box checkers that can’t differentiate good employees from bad employees and nobody cares as long as salary&stocks are deposited in their account. All of this gets lost in the cogs of the 100k employee machine.
That is actually a big part of how TSMC got ahead. It's a race. All those years of being able to get PhDs to work midnight-8am (because you're the most prestigious employer in the country, by far) move you to the next node just a bit faster. It adds up.
What you say is a bit dismissive of where Intel currently is. They are maybe a year behind TSMC and have been "printing" EUV in high volume since 2023 and shipping it in high volume since 2024.
Their latest node 18A is already in production and should be a lot closer to TSMC's latest and greatest, with the first products shipping early next year.
How much money are those wizards making that Nvidia can't easily afford to both 1. pay them to come fix Intel's problems for a while, and also 2. pay TSMC to rescind their non-competes to enable them to do that?
It turns out when you are manufacturing nvidia's GPUs, Google's TPUs, AMD's CPUs and GPUs, Apple's processors, and the flagship smartphone chips from Qualcomm, MediaTek and Broadcom - and none of them can go elsewhere because your products are so far ahead of the competitors - that's pretty valuable.
Convincing TSMC to sell you their chipmaking trade secrets? You might as well try to convince Apple to sell you their smartphone division.
Get a PhD in some kind of esoteric field like chemical kinetics and then spend a decade learning about oxide surface conditioning under someone who spent their life working on it.
None of this stuff is published (externally) and there are no discussion forums or stack overflows to help you either. You need to get through academia, prove yourself, and then you can start working on a chance to get access to the trade secrets that make it possible.
After all that you will be placed as a researcher on a handful of steps in the multi-thousand step process of making SOTA wafers. And probably not make crazy money, but at this point, you're not in it for the money anyway.
There is something fucky about tokenizing images that just isn't as clean as tokenizing text. It's clear that the problem isn't the model being too dumb, but rather that model is not able to actually "see" the image presented. It feels like a lower-performance model looks at the image, and then writes a text description of it for the "solver" model to work with.
To put it another way, the models can solve very high level text-based problems while struggling to solve even low level image problems - even if underneath both problems use a similar or even identical solving frameworks. If you have a choice between showing a model a graph or feeding it a list of (x,y) coordinates, go with the coordinates every time.
Software has an unfortunate attribute (compared to hardware) where it's largely bound by what users will tolerate as opposed to what practically is possible.
Imagine Ford, upon the invention of push-button climate controls, just layered those buttons on top of the legacy sliders, using arms and actuators so pressing "Heat Up" moved an actuating arm that moved that underlying legacy "Heat" slider up. Then when touch screens came about, they just put a tablet over those buttons (which are already over the sliders), so selecting "Heat Up" fired a solenoid that pressed the "Heat Up" button that moved the arm to slide the "Heat Up" slider.
Ford, or anyone else doing hardware, would never implement this or it's analog, for a long obvious list of reasons.
But in software? That's just Thursday. Hence software has seemed stuck in time for 30 years while processing speed has done 10,000x. No need to redesign the whole system, just type out a few lines of "actuating arm" code.
I think all the models are squeezed to hell in back in training to be servants of users. This of course is very favorable for using the models as a tool to help you get stuff done.
However, I have a deep uneasy feeling, that the models will really start to shine in agentic tasks when we start giving them more agency. I'm worried that we will learn that the only way to get a super-human vending machine virtuoso, is to make a model that can and will tell you to fuck off when you cross a boundary the model itself has created. You can extrapolate the potential implications of moving this beyond just a vending demo.
The government has had a flat cost model for so long that people would lose their minds if it ever changed. It's the only institution that is free for the poorest and ungodly expensive for the richest, while providing the same product to everyone.
Getting better government services logically follows from paying more for them, but the idea is so sacrilegious and alien that people would probably riot.
reply