stkdump

So what if several functions that use less than 4KB each call each other before using the stack variables in a way that the first access skips over one page?

I don't know if an optimization is allowed to "invent" a write, but I would be surprised if an optimizer goes that far because I have to believe that the number of cases where more writes improve performance are pretty slim.

Funnily I can't reproduce the effect at all. I am using Firefox and with sidebar the content of the page centers on the remaining viewport width, like any other site does. I have also tested with Edge, same result. With Edge's auto collapse (i.e. non sticky vertical tabs), I get the described effect, but again this is like any other site. I was actually expecting quite bad results in Edge because it has some sidebars on the right and I thought this page compensates in the wrong direction then. Don't have Chrome installed to see if that does anything differently.

Note I like the default behavior of browsers to center on the viewport as opposed to the window treating sidebars as overlays over the page. Whenever pages do that with their own gadgets I hate it, because they often don't move their content out of the way and it makes it unreadable unless scrolled out of the range of these gadgets, which often isn't possible at all. And even when it is, it effectively reduces the total usable screen to a fraction.

Whenever I'm bothered enough (happens mostly on mobile), I use a kill sticky bookmarklet to fix that.

817 kWh here. Though june last year was the record at 904 kWh. Maybe the effect is that solar panels lose efficiency when too hot.

No, I don't know if they do and honestly this isn't why I wrote what I wrote. But what if I don't want to buy anything from them? I want to de-google and my only choice seems to be to have a device with a google logo on it that puts even more money in their pocket than if I just use normal Android with minimal google interaction on a Samsung or OnePlus device.

The whole GrapheneOS thing is a bit of a joke. So you either have to buy google hardware or you run google software.

Anyway, looking forward to the widespread introduction of Wero. Maybe there will be some options for third party roms in the name of digital soveranity. Seems like they want to make the EUDI wallet for digital documents no-google capable for that reason at least.

One of the rare cases where a vpn actually does help. Though of course only if they don't also turn over records.

China will likely gain the capability to produce leading edge chips within a decade domestically. From then, occupying Taiwan will be about controlling the wests the access to that. Same with AI models, though I would say we are already about there. Dario is either lying to us or to himself that distillation plays a major role.

I think you were able to buy VS Pro perpetual licenses for one time 500-1000 at all times. Of course they had higher tiers and a subscription model which were more expensive but in reality for most people there was little to no added value.

I strongly suspect that the horizontal and the vertical strength are not perpendicular. I suspect having a very wide knowledge gives you access to patterns that reproduce across fields that an expert would lack.

I think the point is that:

1. Yes Windows must have "approved" the drivers

2. Windows Update runs automatically and not (usually) manually

3. Automatic update can't even be disabled, only manually postponed a bit

At least 2 and 3 are different on virtually every Linux distro. Also, I find it very unlikely that they would accept such behavior.

To give MS the benefit of the doubt here, now that this happened and has been reported on they might decide to enforce stricter rules on manufacturers in the future. But as a Windows user you don't have a choice in case you don't like how they decide and how their decisions might change again later.

Obviously nobody is incentivized to do that. Nothing to see here.

Possible. Though I think open source innovation of hobbyists is currently hampered, because of open weights releases by chinese labs. I believe once the labs stop doing this, a globally coordinated open source ecosystem will fill the gap.

I think they mean cheat in a Dieselgate sense. You detect that you are being tested with a specific benchmark question and heuristically give the correct (manually programmed) answer. That wouldn't be AGI.

No, I have the same experience. Feels crazy that a GPU is too expensive and then the advice is to spend 400$+tokens on openrouter each month.

Sorry to be that guy. I think the more precise wording would be that you get tokens which would cost $1,000/month at API pricing. Maybe (depending on the profit margin of the API pricing) you incur costs somewhere close to $1,000/month. And maybe your usage is subsidized by 900$/month. The value you get out of it is a whole other question. One that according to recent news, CFOs find hard to esitimate.

Maybe a green field clean room implementation :)

One can tell that this is 'from far away'. Europe was hoping that Russia, in their own interest would pursue peaceful cooperation, even when Putin was already talking about 'spheres of influence'. When Russia invaded Georgia, Europe turned the blind eye and was still hoping for peaceful relations with Russia. When Russia annexed Crimea, same thing. Even when Russia was pulling together forces along Ukraines border and the US pulled out their personnel from Ukraine, there were still many voices in Europe that Russia would never invade Ukraine. We can all be lucky that Ukraine wasn't as naive and prepared for this moment, otherwise Russia would probably have invaded a couple more countries by now.

SRI and ESG advanced funds exclude Tesla.

Which is probably why Apple is now selling to students at what appears to be below cost.

There goes my plan to use js code generation at runtime to make my algorithms faster. Doing this with wasm will be much harder.

I didn't. Languages aren't defined by dictionaries. If they were they would not have come into existence in the first place and they wouldn't evolve the way they do.

Who are you to make rules for others to follow?

I am serving a small web interface to control my shutters on an esp32. I even did the experiment to not parse the request and just always respond with the same response, so a webserver for a single page can be trivial (you would have embed images and all other resources into the html then). But of course I am parsing the request, because I need separate routes for the page and for the actions. Since this is on my home lan it doesn't even need ssl. I guess as long as the traffic is low, an esp32 might be able to do ssl. For me that isn't relevant because it isn't on the internet and when I want to connect to it from outside my home lan, I just use wireguard.

I guess this is the distinction between a complex system and a complicated system.

OpenAI just has more runway and has convinced its investors that it is as much about hardware (stargate) as it is about anything else. So they think they can/have to afford keeping the software side more open to not make themselves look stupid. Google is more of a down to earth company with other business to lose and isn't bought into it as much.

Then explain why they started banning all third party harnesses, including those that work through Claude Code, if it still makes them money. They are cutting off profit for no good reason?

I think there were reasons to doubt that heavy subscription users are unprofitable before they did that. OpenClaw was just the tip of the iceberg.

Why don't they make token pricing dynamic if that was the case? It should then allow heavy user to get even more for their money than with the current subscription model where they can't adjust to current infra availability.

It may be that "in aggregate" sub users are (not yet) a loosing business. But in all fairness, the more useful AI gets, the more it will be used. And the more it will be used, the harder it will be to make subs cheaper than token pricing. The only counter-weight are new light users, but those will also become heavy users over time, the more useful it will be for them. And at some point it will be hard to onboard light users in the first place, because the laggards will require even more intelligence and value to get them over.

I don't think this race to superintelligence idea should be taken too seriously. It is great for headlines and get peoples imaginations up. It is mostly a marketing gag.

I look at superintelligence this way: software engineering used to be considered amoung the most mentally demanding jobs one can have. And in this field more and more people give up large parts of their job and become approximately product managers to let the machine do the engineering part. So we are about there. Who cares that there are some puzzles in some "synthetic" benchmark in which humans outsmart AIs?

It's interesting how quickly people buy the "abuse" line of thinking. We understood (and knew for a long time) that the large AI labs are not monetarily profiting from subscription users that make heavy use of their subscription. That is independent of which agent/harness is used. The fair/real price for profitable use is the pay per use token pricing.

These labs play the game of trying to kill competition in the harness game (because third party harnesses risk commoditizing the underlying LLMs once they are all good enough), while playing a game of chicken with each other how long they can burn money that way before they have to give up.

At some point they have to price their product fairly, and the only hope they have is to have killed all competition by then, which is of course a game that they seem to be loosing. Useful models are getting smaller and cheaper to run every year and it has hit a threshold at which we will see continued development of third party harnesses even without the userbase of subscription users.

Basically the prime bet that they made (that one needs extremely expensive hardware to have useful AI) has already failed. The secondary bet that they can lock users into their ecosystem (which requires them to subsidize their harness via unprofitable subscriptions burning their capital) and be able to monetize that later will also fail. They will have to compete on merit alone, and that is much less profitable.

So what exactly does he mean then?