conjecTech

The benefit of building it into an appliance is that the device itself is certified, and you don't need an electrical inspection for just plugging it in.

They mentioned adding a built-in outlet so that you could then have something else be on battery backup as well (fridge being the obvious application). I'm not sure if they ended up adding that for this version or not.

Nuclear is already about 20% of US electricity generation. I don't think many people are suggesting taking that offline. When people are talking about being all solar, wind and storage they are talking about _new_ generation. So the eventual solution would still be a mix of all of those.

I'm speaking colloquially. I believe it covers exactly this situation.

Isn't there treble(3x) damages for knowingly selling something that violates others' IP? This seems like a risky gambit by Apple. I'm not sure what their endgame is, but having those potential damages stack up in the likely scenario the patent violations get upheld seems like they would be a big deterrent to having Masimo come to the table and negotiate.

Yeah, it could be that they plan to do #3 through use of their chunky GPUs. There is definitely a lot of focus on the whole APU concept. From look at some of the dies and what projects like ggml have been able to do through mixed computation, I'm guessing the neural engine is ~3-5x the throughput/area vs GPU for the kinds of operations its doing.

Trying to map neural nets onto graphics seems reasonable, but personally I would bet on neuralizing graphics to be the better strategy long term.

I've noticed that despite Apple's stated on-device aspirations, they've put very little incremental die space toward the neural engine in the most recent generations. I think there's a number of possible explanations for the disconnect. Curious if anyone with a better perspective knows which are true.

1) They think the existing hardware is powerful enough.

2) The utilization of the ANE doesn't justify increased resources.

3) They plan to re-generalize AI compute through things like vector operations.

4) (the most pessimistic) They are saving large increases for future releases when they need to force upgrades.

I could see the math for on-device just not working for things like massive LLMs. The amount of silicon you'd need to make it possible would be large, and the frequency of use low. The same silicon for one person's phone could likely support dozens if it were in a datacenter instead.

Agreed, I think its much fairer to say they snatched a stalemate out of the jaws of defeat rather than just "this is good for Tesla".

And where is that?

PG&E commercial energy rates in California are $0.36/kwh including delivery[1]. Tesla charges about $0.50/kwh[2].

[1] https://www.pge.com/tariffs/electric.shtml [2] https://electrek.co/2022/09/28/tesla-hikes-supercharger-pric...

The price of charging is already dominated by the marginal cost of electricity, in the same way the price of filling your gas tank is by the price of gas. That will only be more true as charging gets faster. So an advantage in fixed costs is unlikely to be much of a strategic benefit. If they can save 100k/stall, that's great. But that is only a few hundred million a year at that install rate.

People also made the argument that the exclusivity of the Tesla supercharger network was a positive for Tesla. It's hard for both to be true.

I think the reality is that uncertainty about charging away from home pushed a lot of people towards Tesla who might otherwise have considered a different make/model. With that gone, a moat vanished. That being said, competition in the US EV market is still weak. Maybe the added charging revenue outweighs that for now, but long-term, it seems like it will undoubtedly be a negative.

Unfortunately not everyone we dislike is bad. I think this is a case of the fundamental attribution error: Most mean people aren't inherently mean, but are in circumstances that have made them so.

Yes, there are newgrad "rockstars" that can't play nice, and end up not being very productive. But there are also veteran engineers who have stayed at companies through 3 years of 100% attrition and are not super interested in listening to the opinion of the 4th junior backfill on some minor improvement to a system they architected and have been given 0 time to pay down technical debt on. Or maybe they are going through a messy divorce and the stress of it bleeds into their professional life.

I've gotten a lot of good out of finding such individuals, assuming they're reasonable people, and doing what I can to make their lives a little better.

They aren't manual takeovers, Kyle already responded[1] to this himself and clarified that they are requests for input which themselves mostly autoresolve. This is closer to an active learning loop than humans taking over.

[1] https://news.ycombinator.com/item?id=38145062

I just do not understand how people can look at a company that has achieved what Cruise has with such pessimism. Successfully completing 250k autonomous rides is an extraordinarily achievement. Yes, there have been problems, but the issues that Cruise is facing today are a world away from the kinds of problems autonomous vehicles were facing even 5 years ago. The fact that there is one major incident being discussed and not a pile of bodies indicates just how close we are to what will undoubtedly be the largest transformation in transportation, and human productivity, in our lifetimes. Saying that they are "pretty much over" when they are one of two organizations in the world with this level of capability makes no sense to me.

I'm really excited to see progress being made in this area! I haven't read the paper but looked into this area about 10 years ago, and the two questions that jump to top of mind are: 1) What thermal transfer mechanism being controlled?

2) What is the dynamic range in terms of conductance? (Looks like 13x)

There are a number of different mechanisms for energy transfer at this level, and it seemed like the most straightforward way to build a thermal transistor was to get a material that was an insulator for all but 1 transfer mechanism (usually phonons) and then use an electric field to try to permit/block that. The solutions that did exist back then seemed to be very leaky, with thermal transfer while "on" was only about 3x that of when it was off, and this was for microscopic structures where it wasn't obvious how they would physically scale to useful sizes.

I would _love_ a practical thermal transistor technology for an energy management system in a passive house. Allowing your house to "breathe" at night during the summer without things like loud attic fans or having to manually open windows would be fantastic.

I'm really excited to see progress being made in this area! I haven't read the paper but looked into this area about 10 years ago, and the two questions that jump to top of mind are:

1) What thermal transfer mechanism being controlled?

2) What is the dynamic range in terms of conductance? (Looks like 13x)

There are a number of different mechanisms for energy transfer at this level, and it seemed like the most straightforward way to build a thermal transistor was to get a material that was an insulator for all but 1 transfer mechanism (usually phonons) and then use an electric field to try to permit/block that. The solutions that did exist back then seemed to be very leaky, with thermal transfer while "on" was only about 3x that of when it was off, and this was for microscopic structures where it wasn't obvious how they would physically scale to useful sizes.

I would _love_ a practical thermal transistor technology for an energy management system in a passive house. Allowing your house to "breathe" at night during the summer without things like loud attic fans would be fantastic.

1) There is no reason to attribute all of the carbon intensity to the last use to be added. It is far more accurate to estimate the carbon intensity of something by the blended production mix at its time of use. Fossil fuel plants are often kept online for factors beyond the need for additional capacity at a certain time.

2) You are assuming there is no curtailment or rejection of wind, solar, and nuclear, which is categorically false. This is why there are time-of-day rates that vary so drastically. There are large amounts of renewable/low-carbon energy that would otherwise be wasted in pretty much every geography.

Mackay's book, and this discussion, are in the context of a hypothetical future where Britain is powered predominantly by solar and wind. It's not even that hypothetical anymore, because as you can see from this dashboard(https://www.energydashboard.co.uk/live) the UK is currently getting 75% of its electricity from renewable and low carbon sources. So even today, driving 50km there only requires 15kwh of gross energy.

I'm sure he was. It's easy for us to see 15 years later that the combination of EVs + solar/wind has a huge efficiency gain because you can get to avoid internal combustion altogether, but EVs were nascent enough in 2008 for that to not be as clear, so those assumptions didn't make it into his calculations.

I love this kind of hypothetical analysis, but I'd also like to point out there are already people making real strides towards this on an individual level in the UK. Youtube channels like ElectricVehicleMan catalog what a conscientious person can practically accomplish by themselves with readily available solutions today. I've particularly been impressed by the synergies between rooftop solar, battery storage, and combined heat-pump hydronic heat/hot water solutions.

It seems plausible that even a person in a typical row house could offset most of their household consumption with solutions that will end up with a reasonable return over time.

https://www.youtube.com/@ElectricVehicleMan/videos

Correct, a lot of Mackay's estimates fail to account for the difference in delivered efficiency between electric solutions and their gas counterparts. For instance in his car section[1], he estimate you'd need to produce 40kwh to drive 50km. That may be close to the true energetic content of the gas burned, but you could drive that far on just 10kwh in a modern EV, meaning electrification dropped your gross energy needs by 75%.

[1] https://www.withouthotair.com/c3/page_29.shtml

Deliveries of any reasonable size reduce vehicle miles travelled for goods. If a delivery driver can do 50 deliveries in a 10 mile route, that is much, much better than each one of those people driving to the store and back. I live in Brooklyn and Amazon has a grocery logistics hub less 5 miles from me. My carbon footprint is just fine.

Agreed Amazon isnt the most reliable, but as I said, I corroborated this at a few other stores. Costco generally has fixed markups for its products (15%), as does Trader Joes. The fact that you haven't seen price fluctuations there aligns with my conclusion that the price volatility isnt being driven by increases in the cost of goods, but rather retailers who think they can trying to increase margins.

Its sold by amazon.

I have been fascinated by the price of oats recently. The 8lb box of oats I have been buying from Amazon for several years suddenly went from $15 to $32 in July. Paying $4/lb for grain feels excessive given that meat is cheaper than that in most of the country. I went looking for an explanation expecting there to been some huge crop loss, but found oat future prices have been relatively stable at around $0.15/lb. I looked at other prices, and whole foods even charges $3.6/lb for their store brand and something like $2.5/lb for bulk goods. I distinctly remember buying the same bulk oats for $0.59/lb in 2017, right around the time Amazon bought whole foods. And oat futures prices are only up about 30% since then.

I think retailers would like us to believe these increases are just them passing through rising commodity prices from inflation, but I am increasingly convinced it is collusion, either implicit or explicit, to increase margins. The most compelling evidence this is happening seems to be Costco, which generally sets its prices at near-fixed 15% markup hasn't experienced nearly the same increases and is selling similar products at less than half of that price.

I think the entire thing is indicative of a growing disconnect between the cost of commodities and what consumers are paying due to increased consolidation and collusion among retailers.

That makes sense, I'd still like to know what kind of footprint the combined decoder needs.

Negotiation is not an effective strategy when there is active collusion to fix prices. You are commenting on a story where a jury just found that had been systematically done.

I agree that is how I'd expect it to be implemented, but I'm not sure how small it would be given the processing bandwidth we are talking about for 4k video.

I'm guessing this is a distinct region of the chip and not integrated with CPU/GPU since they scale up by replicating those blocks and wouldn't want to redundantly place that hardware. Having it separate also allows a team to work on it independently.

I think the relative size of the media engines is accurate in that slide, so then it comes down to how large the ProRes parts are in other chips. They are probably a couple of the unlabeled regions next to the performance cores in the M1 Pro die shot below, but I don't know which.

https://images.anandtech.com/doci/17019/M1PRO.jpg Taken from: https://www.anandtech.com/show/17024/apple-m1-max-performanc...

How much silicon does it take to add an AV1 decoder to a chip? The areas Apple highlighted in their A17 release looked pretty substantial, but I wasnt sure if it was to scale.

https://b3148424.smushcdn.com/3148424/wp-content/uploads/202...

Gary Explains gave some numbers that seem to indicate its on-par or a bit better than M2 Pro at similar power levels.

https://www.youtube.com/watch?v=_8wzGkTWdP8&t=348s