remus

> Sure, but how is OpenAI, and Claude handling the increased usage?

By spending tens of billions in investor cash. It is not clear what a sustainable business model looks like for these companies at the moment.

> They are handling it a lot better than Github.

Are they? Anthropic has fairly frequent issues.

There's high growth and then there's demand increasing 10x in a year. For a mature business like github that's a massive, and I can't think of many big businesses who would be able to scale smoothly to accommodate that kind of demand.

> I want the model I'm paying for to be "pure".

I don't think the models are pure in any meaningful sense. The labs have some idea of what kind of output they want from the models and then they put a huge amount of effort into training the models on the right sorts of data and massaging the models afterwards to push them towards the desired output. Then at a more practical level there's the layers of filters before your prompt even hits the model (e.g. anthropic's auto-mode classifier), system prompts, response level filtering etc.

Same as usual: add tests. Over time the test suite becomes the spec that describes how the software should function. As the test suite grows you squeeze out room for undefined behaviour and bugs.

> But if we go only 15 years in the past, to 2011, we find the same price per GB of DRAM as today, so that affordable computers had only 16 GB of DRAM.

Some of the individual components may still be relatively expensive, but overall I don't think this is true at all. Compare a 2026 macbook pro with a 2011 macbook pro at a similar price and the 2026 model is gonna blow the 2011 model out of the water on any benchmark you care to use.

I'd be interested to see some sources. My understanding is that they're aiming to peak greenhouse gas emissions before 2030 and be carbon neutral by 2060.

The direction of travel is clear though. China is making massive investments in renewable energy (and dominating sectors like solar in the process) while the US seems hell bent on going the other way.

> Incompetence in leadership is what makes a tech business technically unable to meet growing demand.

I don't think this is true. Take LLMs for example, you need GPUs to serve them and these are in short supply at the moment making it hard to meet demand. I don't think this is necessarily the fault of incompetent leadership.

If we just turn off all the computers there will be no bugs!

> When I read a critique that could roughly be summarized as "I could have done better"

This feels like a pretty shallow take on a what seemed to me like a pretty in depth and interesting review. Clearly there's a huge amount of potential ground that could have been covered in a film like this, so a critical look at what was and wasn't included, the angles that were chosen, the parts that were glossed over etc. can be revealing. Put another way, I think the review is a pretty in depth look at the questions "what was the artist trying to achieve?" and "Did they succeed?".

It's easier to get good quality scans from individual pages than it is from pages in a complete book. Imagine laying a book flat, then the page is distorted in the area around the spine. You can work around this (either by trying to correct for the distortion in software or with clever scanners that position the book more advantageously) but it adds complexity compared to chopping off the spine and just dealing with flat sheets of paper.

> A part of the Fontainebleau forest, much more north, no pine trees, went in smoke earlier. Forests are burning near Madrid.

Small correction, but Fontainebleau is a fairly mixed forest including 20-30% pine.

Perhaps I am overly optimistic in thinking this is just incompetence.

Indeed, I've got a site which gets a lot of bot traffic and google bot is pretty sensible compared to a lot of other mainstream bots.

> There is an even higher level of creativity in creating books, songs and all sorts of art used in model training though.

No argument here, I completely agree.

> There is no world in which me vacuuming the entirety of human knowledge to make a genai model is ok but hoovering my model answers is not.

I disagree with this though. Clearly LLMs owe a huge debt to everything that has come before, but surely you'd agree that the models that are produced are something substantial and new and novel which didn't exist before and have lots of value in their own right. Let's be a bit reductive and pretend Moonshot had just outright stolen the weights from Fable somehow, clearly that wouldn't be contributing anything really new or novel. Now of course they've distilled rather than stolen, but the point is similar: how much value have they added along the way?

Yes, this is what I was getting at.

While I agree on a moral level, I think there is a distinction to be made. Training a SOTA model takes a huge amount of resources and expertise so the people doing the training are adding a lot of value along the way. I think this is much less true for distillation (which is kind of the whole point).

ed: to clarify, I totally agree that a huge chunk of the value in LLMs is coming from the source material. My point was just that training an LLM takes more resources and expertise than distilling from an existing LLM so I don't think the equivalence between training and distilling is entirely justified.

I would only agree partially. There are counterexamples that are not illustrative, but it is fairly common that in thinking about how to construct a counterexample you gain a more thorough understanding of the original problem and at least one fundamental issue which prevents the conjecture from being true.

I don't think it is true that they necessarily want the US to struggle, I suspect it's more self interest. LLMs seem to be one of the biggest innovations of the last few decades, China probably just wants to make sure it's not being left behind and/or made hugely reliant on the US for what seems to be turning into a piece of critical infrastructure.

China has an effective strangle hold on some key sectors (solar, rare earths) and I am sure they relish this position and the leverage it gives them. You'd be careful not to give away that same leverage to a competing power if you can invest a few billion now and cover your bases.

> Please don’t respond if you are speculating.

I doubt you are going to get a response from an anthropic employee, but I think it is safe to assume they have swapped to a new tokenizer because it improves the performance of their models.

Obviously there are a lot of important things you need to keep a country running, but steel is a key input a in a huge number of very important sectors (infrastructure, military, automotive etc.) so having some ability to produce your own steel seems a sensible hedge.

As a website owner, if I saw someone scraping with a realistic looking name + email address I'd definitely give them more latitude than someone trying to hide the fact they're scraping. In my experience people who are hiding the fact are much more likely to be doing something nefarious.

> Its AI proposition is more or less like the Google Plus. Nobody really wants it but they know about it because google pushes it everywhere it can.

Citation needed. The gemini app has 750 million MAU, hardly a dead business.

> hence Google paying SpaceX hundreds of millions of dollars a month to SpaceX for capacity there.

I don't think that necessarily follows. It could just be old fashioned capacity issues, for example. If nothing else nvidia are able to charge an insane markup on their AI chips at the moment so even if google TPUs aren't competitive in a pure performance sense they are surely competitive from a pricing perspective.

> This feels like a kid trying to do science. The will is there, but lacks experience.

It's funny, when I saw the title I was hoping the article would include some sort of blind ranking, where you could see the outputs (without knowing which model they came from) and score them on some criteria. Could have been a fun way to get a better ranking of the results.

> Which brings to mind: most of the big shops product (chatgpt, claude, grok, etc...) ALL rely on search, and NONE of them actually have a running search stack.

Don't they? Based on traffic to some websites I run the big AI labs are very actively doing a lot of crawling.

They're very similar models though, just with different safeguards and restrictions in placae around particular use cases.

I guess the underlying issue is that there is this model that is very capable, but it's being hobbled because of a fear of abuse. It may well be justified, but for a legitimate user any restriction just makes it a worse product and after all the puffery around how good it is (and some practical experience of how good it is) it's a pretty shit experience. "Here's our best model, no you can't really use it".

> When you think about it, the idea of a representative democracy is rooted in the technical difficulties of implementing a direct democracy: both spread of information/discussion to the masses and organizing the votes.

I think there is more to it. A large part of democracy is delegating decision making to people with time and expertise to investigate issues more thoroughly than most individuals can or want to.

I have some broad opinions about the environment etc. but I am by no means an expert in the details, so I am happy to delegate day to day decision making to someone with more expertise who's opinions broadly align with my own.

I'd agree that referendums do make more sense on "issues of conscience" though, like whether to have a death penalty, voting reform etc.

Given this a test on some scans of magazines, generally pretty impressed with the results. Mags are generally pretty whacky layouts and it does a reasonable job working out what is where and pulling it together into a single coherent md file. The way it crops relevant pics and puts them into the doc is pretty nice.

Haven't compared it with any other high tech OCR estups, but it's way better than the jank that comes as standard with my scanner.

They dynamically pick the number of days to display based on viewport size. Mobile = 30 days, tablet = 60 days etc.

> Appointments are made in local time. Store them in local time.

You may just be illustrating a particular use case, but it is more ambiguous in the general case. For example if you have arranged a meeting with someone in another timezone then maintaing the local timezone could lead to a misalignment for one of the participants.