zarzavat

The burden of proof is on them to explain why the Chinese would agree to that.

"Siri, please solve the Jacobian conjecture, and also set an alarm for 8am tomorrow"

What the hell is going on over there

Google was late to coding agents and as-per-usual fucked it up with their crazy project management culture.

Usually Google gets away with it due to inertia, however this time they are paying a heavy price because they missed out on the training data that Anthropic and OpenAI have gathered with claude and codex.

This seems like something that could be solved by asking an LLM to write the prompt for the image model. You can also feed in the output of an image model into an LLM and ask it to check it/make improvements.

Frontier = SOTA.

Pareto frontier ≠ frontier.

Type stripping is not the same as emit, it only gets you a subset of TypeScript. Fine if you're running your own code, not fine if you want to run arbitrary TypeScript.

Yes, in the Monaco editor (https://microsoft.github.io/monaco-editor/).

It's not just typechecking, the typescript library is also the reference parser for TypeScript and reference emit. Emitting JS from TypeScript is non-trivial and non-local.

It's not just in browsers, you might want to run the typescript library on the edge or in some restricted environment where JS/WASM is OK but native code is not.

You may want to use the typescript library as a dependency in your own package. TypeScript becoming non-pure JS means that your library also becomes non-pure JS.

Indeed as much I dislike the approach Bun took, at least the port appears to be working. TS7 is going to cause problems for downstream users because Go is the wrong language to use for something that has to run in WASM.

AI is a threat to everyone. People who claim that AI will never be able to do X have consistently been proven wrong.

The only people who are safe are those whose jobs depend in some way on their humanity. e.g. yoga teachers, bouncers, etc

It's insane to cross Apple. In the worst case Apple could take the ChatGPT app down like they did to Fortnite. They are probably waiting for discovery to find out how high this goes.

If Chinese AI companies can train a model that's slightly worse than the frontier, then there's no reason why they can't train a model that is slightly better than the frontier.

Everybody can agree that K3 doesn't clearly surpass Fable. However, inevitably there will be a time in the future when a Chinese AI company releases a model that's better than any US model.

K3 isn't the knockout blow but it's the 2nd knockdown that makes everyone in the arena realize that the fighter is not winning the fight.

Lots of organizations are continuing to refuse to use chinese models due to security and IP concerns

These customers exist (e.g. US military) but there's not enough of them to justify a trillion dollar valuation.

Anthropic's valuation is predicated on growth. If they start going backwards and losing customers to open models, it hurts their ability to gather investment and with it the ability to train new models, leading to a death spiral.

The best they can hope for is that the US gives them state aid to compete with China, however their relationship with the current administration is not great.

It's like reading Anthropic's obituary.

Video game streaming requires a high quality internet connection to a nearby data center. It can work in certain places but there's always going to be places where it doesn't work, and consoles don't have that problem.

That's correct but SQLite was never designed to be a production database in the first place. It can be used as a production database but only if you know what you're doing, and presumably anyone who knows what they're doing knows about the AUTOINCREMENT keyword because it's one of the first things you learn about SQLite.

A distro package manager provides access to a small set of packages that the distro thinks you might like. A language package manager provides access to the full set of packages. The language level package manager is solving a more ambitious problem.

Distro package managers don't solve the problem they just punt on it, saying "you can add an unofficial source but you're on your own to maintain security". I agree with GP that the comparison is tiresome.

In the UK there is substantial use of secondary legislation. This is where Parliament authorises someone else to make laws on its behalf in some context. These powers delegated by Parliament are usually mundane but can be extreme and there's no limit to what powers Parliament can delegate. Since politicians are lazy they tend to use secondary legislation all the time. This encourages a culture of legislative sloppiness.

While the UK courts have given themselves the power to perform judicial review, there is no constitutional basis for this review and so the courts are reluctant to step on Parliament's toes because if they go too far they risk Parliament deciding that it doesn't like the courts' interference and removing their authority. This isn't a hypothetical and the courts have at times resorted to some crazy mental gymnastics to evade Parliament's attempts to prevent judicial review.

The US has a similar concept that enables federal agencies to make regulations on Congress's behalf but it's much more limited in scope due to the separation of powers and more solid position of the US Supreme Court.

Driving is more expensive than public transport when you factor in the cost of the vehicle, maintenance, fuel, insurance, taxes, parking, etc.

If driving were more cheaper than public transport then nobody would use it. Yet, almost wherever you go in the world if you go to a station at 8am on a weekday it will be full of people.

If every road were free of traffic, every destination had unlimited parking at the door, everyone already owned a car, and driving carried no direct or indirect costs, almost everyone would drive everywhere.

What kind of nonsense is this? People take public transport it's better than driving. You don't need to expend mental cycles focusing on the road, you can do something else while you're on the go: chat to friends, browse the internet, read a book, sleep, etc. Driving is dead time.

Nobody could have predicted this.

If open source code can have political beliefs then the only reasonable solution is to live as a hermit.

I can guarantee that there is software on your computer written by communists, nazis, mormons, every single political ideology is represented by lines of code that you run every day, because open source doesn't require any political vetting to contribute.

It's actually simpler: your code is not the valuable part, it's the telemetry/metadata/conversation surrounding your session that's valuable. Every time you press escape, every time tell Claude/Codex that it's being an idiot, your back-and-forth conversations, etc. "when/why did we fail and how can we improve?" is what they want to know.

They can use LLMs to launder confidential customer sessions into trainable data. Then they can claim that they don't train on "your data" without it being technically incorrect.

This is a predictable consequence of age verification laws and social media bans. Formerly VPNs were a nice to have but now they are a necessity in many countries to navigate the modern internet.

The cheapest way to get a VPN (and if you're a horny and broke teenager perhaps the only way) is to trade your clean but censored IP address for an uncensored IP address in another country. You accept the bot traffic in return, or externalize it to your parents or the owner of the internet connection.

Unfortunately those portable units are scams aimed at people who don't understand how air conditioning works or what an air conditioner is supposed to look like.

It's nothing new. In 2003 there was a huge heatwave and thousands of people died:

https://en.wikipedia.org/wiki/2003_European_heatwave

The reason is simple to understand. For the majority of the time the AC would sit idle. Some years AC is not required at all because the summer is mild. It's hard to get people to spend money on things that they don't need most of the time, it's like buying insurance.

I wish that society can learn about and apply the concept of a wet bulb temperature.

A dry bulb temperature of 30C contains little information regarding safety. Humans are wet and we cool ourselves with evaporation.

Anthropic really needs to get Opus 5 out ASAP.

The gap between Opus 4.8 and Fable is large enough to drive a GPT 5.6 sized bus through. A better Opus would take some of the heat off.

This is also what I noticed, it's hella slow and the quality doesn't match the thinking time. Either it's just launch day load or else they went full GLM-5.2 thinkmaxxing.

Either way, there's something fundamentally inefficient about the whole business of training LLMs.

The human brain manages to self-organize with only a fraction of the information that LLMs get trained on. To train an LLM you need a lot of high-quality data.

There are two threads in history: firstly the compute thread leading to GPUs and AlexNet in 2012. Secondly the model architecture thread that started long before we had the compute and lead to transformers in 2017.

If the compute thread had been 30 years behind then we might be spending this century coming up with better architectures to make do with the more limited compute. However since the compute came first, we settled on the first thing that worked (transformers) and all effort went into polishing that.

There's something wrong with transformers though. No matter how many trillions of dollars get thrown at the problem, they still don't learn like humans do.

The way I'd describe it, Fable is the first model that's good enough.

If all I had was Fable for the next couple of years then I'd be totally fine with that. I have never felt that about any version of Opus.