zarzavat

It sounds impressive at first but after you think it becomes mundane.

OpenAI, Anthropic, etc are locked in a race to create better models. They all have interpretability teams. Of course they train their models on every technique under the sun to understand and improve their models, including raw weights. The primary function of these models is to build their successor.

To be fair, the people with psychogenic gluten intolerance have massively increased the availability and variety of gluten-free foods for people with real Celiac disease. They seem to be doing more good than harm on balance.

Exchanges have to walk a fine line. If they make withdrawals too slow or difficult then people also get anxious about the safety of their funds (ironically).

Traditional banks and crypto exchanges alike rely on the carefully cultivated illusion that you can withdraw your money at any time.

$6B is probably several billion too large for a hot wallet but it's not completely obscene for an operation of Binance's size.

If you want to be really paranoid you can make it an online process: application server makes an HTTPS request to a key server on different infrastructure to get the decryption key, then decrypts the secret. Use an IP whitelist.

That way read access to your infra provider is not sufficient, it requires an active component.

Governments may want as much economic activity as possible but we've learned the hard way that this is detrimental to the financial system, because it creates damaging boom and bust cycles.

So we empower central banks to intervene in various ways in the monetary system in order to achieve low inflation and system stability.

Activity is only the secondary objective. System stability is the first. Stability doesn't mean that markets cannot go up and down even violently. It means that when they go violently up and down, there remains faith in the currency and people are not suffering from mass unemployment, etc.

When a currency is healthy and has low inflation, hard assets are not a threat. Gold doesn't pay any interest, it doesn't have any earnings. You don't need to convince people to buy stocks and bonds instead of gold, they do that by themselves.

When a currency is unhealthy, hard assets can become a threat to it, but notably it's usually other fiat currencies that are the biggest threat.

Hard assets are useful tools for the financial system because they are less correlated with the market. You can criticise Bitcoin in various ways but the criticism that it is a hard asset with a fixed supply that by not inflating somehow damages the financial or monetary system doesn't land. Fiat money is intended to facilitate the existence of markets in other things, including hard assets.

That's why it's not safe to hold fiat money, it's by design. Governments don't want people to hold it, they want them to cycle it, spending it, investing it, and lending it out.

The existence of harder forms of money (gold, bitcoin, etc) is an expected consequence of how the system is designed, and not in any way a subversion of it. Somehow, both sides seem to miss this point.

I don’t care and in my opinion AA does a disservice to their users by allowing such blatant manipulation. What would happen if every AI company were to include their unreleased models like this? Put the unreleased models on a separate page and leave the headline benchmark rankings for models that you can actually use.

Else doesn't sound archaic to my ears.

"When you go shopping make sure to buy flour or else we won't be able to bake the cake"

"otherwise" is a more common word choice but "else" is definitely in the active vocabulary of modern English speakers.

The issue is that they have fully abandoned the Plus plan while continuing to take people's money.

Back in March, the Plus plan was marketed as the way to access Gemini 3.1 Pro, their top model. The free plan gave you a tiny quota of their Pro model, and the Plus plan gave you a generous quota, enough for chatting.

It was expected that 3.1 Pro would be upgraded to 3.5 Pro whenever that came out, but that model never materialized.

When Google finally had a new Pro model, not only did they not apologise for the delay, but they didn't upgrade the Plus plan, which is to this day stuck on this ancient model.

Fine, you might say. hey made the Pro model only for the Pro plan. What's the big deal?

But why, why, did they stop upgrading the Flash model as well?

So we've gone from "The Plus plan lets you access our top model with generous limits" to "The Plus plan only supports a months old second-tier model".

I'm sure that Google never explicitly promised that the Plus plan would receive upgraded models, but nevertheless it feels like being scammed.

ChatGPT Go gives you access to Sol which is much better than the ancient 3.1 Pro.

No, not free. I'm on a paid plan.

The borg cube does not understand customer satisfaction.

Perhaps someone should tell the Gemini team about these models. When I open gemini.google.com I see 3.6 Flash and 3.1 Pro.

PKI works on trust. When a CA does things that violate that trust, their root gets removed. That's how it's supposed to work.

All of Apple’s operating systems are designed for high PPI displays. There’s no font hinting, no subpixel AA. Apple decided long ago that a high pixel density is a baseline requirement for any device they sell, including computers, since the 2010s. You can use a low PPI external but it will not look good.

Given that the smartphone industry also adopted high PPI displays as a minimum standard, it’s actually the external display industry that is the odd one out here. Low PPI displays should have gone the way of CRTs, but they cling on.

If you do creative work, including frontend development, you need to have a hidpi glass display because a fuzzy plastic display will not faithfully represent what your viewers/users will see on their devices.

Apple's solution to reflections is to overpower them. I use my MacBook in some bright public spaces and reflections have never been an issue. For an iMac which is more likely than a laptop to be used in an environment with controlled lighting, it's just a non-issue.

That 32" monitor will have a crappy pixel density. It's incredibly hard to find good external monitors that are not made by Apple, definitely not for $800. Apple will give you a display with a high pixel density and glass front.

Every day people go to work and turn off their electrical devices. Switching your home to off-grid solar + batteries is just like going to work and not coming back. The grid can handle it.

If many houses install off-grid solar, it's still not going to have a negative effect on the grid because the change is spread out over time.

It does mean that less generating capacity is needed over the long term, but decision makers can just get that data directly from the grid. I don't see what extra information the permit is adding.

This doesn’t make any sense. You also use less electricity if you switch off your devices. I think Germans just have a fetish for permits.

Disproving big conjectures can be useful to mathematicans because it means they don’t have to waste time proving results that assume that conjecture.

It might make the third option, that P vs NP cannot be proven true or false, somewhat less likely.

If some later version of Deepseek had the capability as Opus 5.5, would you use Opus?

What many of us experienced this year is that Fable 5 and Opus 5.5 are good enough and future model improvements may have diminishing returns even if benchmarks improve.

I’m very excited for smarter models but what I really want now is not capability but speed, cost and open weights.

The ??? is capital power. If both human mind and muscle have lesser value then what matters is the capital you possess. Can be financial capital or other kinds of capital. Indeed this is the default in many less developed countries.

The Copenhagen interpretation doesn't even deserve to sit with the other interpretations because it's not well defined. It's protoscience, like alchemy.

Just make enabling jailbroken mode require a factory reset of the device.

The biggest risk is not elderly people but rather the owner giving their passcode to someone else, e.g. a repair shop, who then enables jailbroken mode without the owner's knowledge and installs malware.

You can mitigate this problem, and the elderly person problem, by requiring the phone to be factory reset. That way any jailbreak would be extremely obvious.

To be fair some things that are considered normal in non-Anglo Saxon societies are considered authoritarian in Anglo Saxon societies, in particular having to register your whereabouts with the government, or having to carry ID at all times.

There's no such thing as identity theft. Individual companies can be defrauded when they don't do sufficient checks. The companies that do sufficient checks don't get defrauded. You can't prevent someone else's incompetence.

As for being subject to prosecution for crimes that you didn't commit, you can either live in a jurisdiction where the justice system works more effectively, or else be rich and have a good lawyer.

This is like saying you only use your car for transportation, but you don't sleep in it anymore because your bed is more comfortable.

There are precisely two languages that make the most sense to implement the TypeScript compiler in: Rust and TypeScript itself.

A Rust implementation will have the fastest compilation speeds, memory safety, can easily be compiled to WASM, or compiled to a native node module.

A TypeScript implementation provides the most convenient way for TypeScript code to use the typescript package as a library. Pure JS, no native code or WASM dependencies. No FFI tax, can be jitted.

Go was always a compromise to make a port easier for humans. Go's GC makes it easier to keep the general code structure from the TS implementation. However that GC is also a liability, you sacrifice speed and embeddability.

Personally I'd prefer they went back to TypeScript for the reference implementation, and used Rust for the performance implementation. Go doesn't bring anything to the table anymore.

If you're a software engineer or mathematician then your world has already been turned upside down by this technology.

I used to write code by hand, now I just write prompts and manage the AI. There is some expertise involved but considerably less than before, and no doubt even that will vanish in the coming years.

For mathematicians the picture is even more bleak because their field is being strip mined and they face a future where there are perhaps no human reachable conjectures. Moreover if your job is to teach other mathematicians and nobody wants to become a mathematician... This is the future even if the technology hardly improves at all.

You don't need to know if there will be safe jobs in the future to avoid jobs that are unsafe today.

Personally if I were a teenager today I'd choose a career in medicine. You can still exercise your brain, but because it's practical and the fact that you have people's lives in your hands means that it's considerably less likely to be replaced by AI any time soon.