They also said that theur hardware scales faster than Moore's law, when in reality it is way slower than Moore's law. Which I guess at least is a good thing for hyperscalars because it helps with their decision to write off their GPUs over a much longer period.
But still, it seems that billionaires are lying left and right. It isn't just Elon.
A line of constant bearing will always be a great circle. A straight line on a mercator projection of the earth does not constitute a great circle (in fact only the equator line is a great circle and each vertical line is half of a great circle).
> A line of constant bearing will always be a great circle.
That's not right. Imagine heading due West, and keeping your compass pointing West every step anywhere not on the equator. That traces out a longitude line, not a great circle.
If you follow a constant compass heading you will spiral toward the poles in a helix, not travel in a circle around the globe. To travel along a great circle, your compass bearing must constantly change as you move.
- Autonomous cars will improve traffic because stopped cars can start going at the same time at an intersection -> Then you have no safety distance. And while theoretical improved reaction times mean you might need a bit less distance, the quadratic term comes from the needed breaking deceleration, not the reaction time.
- Autonomous cars will reduce congestion: they can't if they are supposed to drive around empty either to park somewhere else or to get to the next person to transport. On the contrary, they will increase congestion.
Note also that I don't buy the "work" inside "very small cars". It's not just that having to drive prevents you from working. Also the overall comfort while traveling isn't good enough outside of trains or planes for anything but some tiny amount of work.
I am on board with the benefits for old, disabled, tired and drunk people though. Also, I do accept that they will prevent a lot of accidents and deaths.
You only need reaction time factored in when the driver in front might act unpredictably. If the self driving cars pass information about the road ahead to cars behind, and let the car behind know what they are doing and planning to do the the required 'reaction time' is zero. Congestion is often caused by reaction times stacking up until a car has to completely stop. Then you have a stationary section of road that makes all the other cars that encounter it stop. If reaction time is zero, this removes that congestion, so more cars can drive on a section of road and not clog it. The same applies to intersections, roundabouts, on and off ramps on motorways. Just the simple act of negotiating with the cars around eliminates all of it.
Just think a little about what having a computer control the car would really mean, rather than thinking of them like robot people.
An autonomous car in front of me can still act in a way that I or my car can't predict, as long as we haven't eliminated humans and animals completely from being anywhere near roads.
> - Autonomous cars will improve traffic because stopped cars can start going at the same time at an intersection -> Then you have no safety distance. And while theoretical improved reaction times mean you might need a bit less distance, the quadratic term comes from the needed breaking deceleration, not the reaction time.
This is more about it deleting the human not paying attention + reaction time to next bumper.
Currently it looks like
1. Light goes red
2. Car1 reaction time and then moves
3. Car2 swipes off TikTok app + reaction time off car1's brake lights
4. Car3 swipes off Instagram + reaction time off car2's brake lights
5. Car4 Yelling at kids in back seat until they hear Car5's horn + reaction time
Versus autocars
1. Light goes green
2. Cars 1-N all release their brake faster than human reaction time and begin moving ever so slowly. Their acceleration rates are respectively less for the reasons you said. More cars make it through the light meaning less likely to accumulate and create multilight jams.
Autonomous cars driving around empty is a good argument for them being on a public rideshare network. They are more likely to pickup a passenger along the way to their actual destination. Even if the route is non-direct, any trip that gets them closer to home is not wasted travel because that person(s) wanted to travel regardless. Similarly when they go to pick you up.
IMO The bigger culprit here is the artificial need for people to work in an office, and for them to start at the same time. Work from home and higher agency over time would mean people do novel things like 1) Not commute at all or, like I do, 2) Take a couple meetings from home and commute when it's quiet.
Well said. I would say in general there isn't "the best" OS for everyone and never will be, because each OS makes different trade-offs. I for one want primarily what is understood to be "general purpose computer". Other people rightly don't care about that and want a maximum security device, one that even protects users from their own mistakes (of course putting more trust in the makers of the OS). What we should care about is that people have a choice and can get whatever they prefer.
To answer GPs point, I think Fairphone doesn't primarily target either of the two audiences. I think they primarily target the people that care about the ethics of the creation of the hardware. Basically people who would like to minimize the invisible human cost that their phone creates.
The misspelling seems to be consistent (>200 times), though it seems like a quite common one. A 1982 DOS game was even titled "Missle Strike", long before Metroid was released.
> Just like how World War I was initially called “The Great War”. If you had called it “World War I” from the outset, people would look at you funny, like “Do you know something I don’t?”
This is why I didn't, at the time.
Jokes aside, some people knew and so it was also called (though not very commonly) World War starting 1914 and First World War from 1918. Because of course it wouldn't be the last. World War 3 was first talked about in 1941. And the famous idea that World War 4 would be fought with sticks and stones was first mentioned in 1946.
So the correct answer to "Do you know something I don’t?" is "I know human nature"
I recently thought I could probably go back to a not-smart watch to not have to charge the watch all the time, but I don't want to deal with setting the time regularly either due to quartz drift and dst. So a bluetooth watch with two years of battery life doesn't sound like a bad idea really.
Even cheap quartz movements tend to drift by ±10 seconds per year, so you'd only really set it at dst. Maybe you're thinking about automatic watches which can drift by several seconds per day, and stop if you don't use it for 24/48/X amount hours. I have to set my automatic quite frequently
+/- 10 seconds a year for a quartz watch isn’t cheap. A Grand Seiko or Citizen high-accuracy quartz watch costs around $3,000 for the most basic models. A cheap quartz watch is an order of magnitude less accurate, which is still more accurate than a Rolex .
±10 seconds per year would be quite an expensive quartz movement. To be that reliable anywhere it should be thermocompensated, too. This is the range of Grand Seiko 9F, Citizen Chronomaster, or Bulova Precisionist.
Not OP, but you can influence how deterministic your LLM behaves using the temperature setting. The neural network doesn't directly output tokens, but logits which are then converted to probabilities and then a token is chosen at random, unless the temperature is 0 (i.e. greedy, we just always pick the most probable token without any randomness). All speculative decoding methods have to "commit" to a token though even when they don't know the actual logits of the full size NN yet. The question then is (and I don't know the answer): how do the common inference engines behave when the speculation landed on the most probable token, but the random choice still doesn't land on it? You can imagine that in the interest of performance as long as we stay reasonably inside the probability we just go ahead with the speculation. Not sure if thats implemented like that though.
Edit: I just looked up the math, and actually the idea of speculative decoding is done in a clever way that fully preserves the probability distribution while still maximizing the acceptance rate of draft tokens. So I would have to disagree with OP and say that no, non-greedy sampling doesn't influence the trajectories.
Yes, it doesn’t impact the probability distribution due to verifier. However, remember how you use PRNG and effectively due to the drafter is sampled from a different distribution initially, a separate rejection sampling won’t be able to recover what the “old PRNG” would choose in a “without drafter” case. Hence in my original post, it is about different trajectories you will end up with, not the correctness of each stochastic sampling.
If you assume that the RNG generates true randomness, then the two are identical. Only if you care about the determinism of the RNG (for example you want to use identical seeds and get the exact same generation between the two) it makes a real difference.
Correct. I am trying to explain why even it is "exact", the generated text is different from the with / without DFlash2 runs, and potentially why the DFlash2 run will contain the invalid Python syntax.
If the underlying probability distributions are the same, then DFlash can lead to an invalid Python Syntax iif the autoregressive process could have generated one if the random sampling picked a different token.
If a model can output a “wrong” sequence with a certain probability p, then Dflash can also output the wrong sequence with the same probability. They wouldn't necessarily produce the same output from the same seed, but speculative decoding shouldn't be able to produce anything that the autoregressive model couldn't also produce when using a different seed.
It will contain the same or different syntax with or without it. Also multiple runs without DFlash2 will contain the same or different syntax. And multiple runs with DFlash2 will have the same or different syntax with the same probability. DFlash2 literally has no influence (unless buggy). The difference is purely caused by the randomness.
But still, it seems that billionaires are lying left and right. It isn't just Elon.
reply