For us mere mortals, how fast does a normal developer for through a MTok. How about a good power user?

A developer can blast millions of tokens in minutes. When you have a context size of 250k that’s just 4 queries. But with tool usage and subsequent calls etc it can easily just do many millions in one request

But if you just ask a question or something it’ll take a while to spend a million tokens…

It's worth noting those 250k tokens will be cached for repeat queries.

Seems like an opportunity to condense the context into 'documentation' level and only load the full text/code for files that expect to be edited?

Yeah that’s what they try to do with the latest coding agents sub agents which only have the context they need etc. but atm it’s too much work to manage contexts at that level

I use one Claude instance at a time, roughly fulltime (writes 90% of my code). Generally making small changes, nothing weird. According to ccusage, I spend about $20 of tokens a day, a bit less than 1 MTOK output tokens a way. So the exact same workflow would be about $120 for higher speed.

I tried using it for two hours and it burned $100 at the 50% discounted pricing.

Not your weights, not your agent

Maybe xAI/Tesla, Meta, Palantir

Just like AI, the winners will (continue to) be the ones with the most access to data and the technical and financial capital to make use of it.

He's going for number 7 it seems

Screenshot of the top 3 results on Google shopping ought to do it

This was early-00s so it was slightly more trouble than that but still not the end of the world.

The point is, the person wasn't trying to hedge against looking bad, they literally had to do and document this.

Yup. Governments have to follow all the laws, which often companies can ignore in the interests of speed.

Also, governments are large bureaucracies, with all the process that entails. And because there's no real benefit for them in delivering quicker, but lots of risk in delivering badly, this sort of stuff happens.

And even if doesn't, writing five online shops to send you a written offer takes a couple minutes and results in the same or lower prices

Procurement for such small items can be quick and sane. It's the larger items where rules tighten and procurement portals or bidding become mandated that are problematic

Nothing takes 'a couple minutes' when you have to sit down and research the five online shops, find if they are approachable, if they will deal with the restrictions of your purchasing department, find out how to submit a query. Many online shops just have a purchasing portal. Find the product, buy it here, pay for it and wait.

So loosely I purchase items at my work from a budget that I am allocated in an organisation that is ultimately responsible to the UK government. I need to justify that the items I am ordering are reasonably priced, and the organisation would really really like to have the goods before any money goes out. That means they want to place an order, receive the goods and and invoice, and then pay the invoice. Many online shops don't want to deal with that. We have accounts set up with many companies, but not all. If I want to buy some reams of 160gsm A4 white card (for example, the other day), that whole process is going to take at least 10 minutes. Some of our suppliers don't sell exactly that. Is 240gsm ok? I've got to go back to the person who wants it (no btw, I had to go find some and take it to them for comparison). More esoteric items are going to take longer. What exactly do I want to order?

So yeah, procurement is simple when you are at home with an amazon account. The items will be here tomorrow!

edit: oh, I didn't mention the free delivery.. a box of white card doesn't get me free delivery. Is there something else I can add onto that? Ok, the order will have to wait..

> I need to justify that the items I am ordering are reasonably priced

Unfortunately it sounds like the process is misaligned with the intention. I doubt this mechanism actually works for efficient budgeting and even when it appears to work, it’s probably at the cost of standard quality.

With a non-sequential generative approach perhaps the RAM cache misses could be grouped together and swapped on a when available/when needed prioritized bases.

China is capitalist on a state level, that's where they are winning. The US lets things get mired in red tape and special interests because nobody wants to take responsibility for growth.

In China, I imagine that if your company does something relevant to the five year initiative then you get a lot of red tape cut for you.

> China is capitalist on a state level, that's where they are winning. The US lets things get mired in red tape and special interests because nobody wants to take responsibility for growth.

i.e. in China, the government controls capital; in the US, capital controls the government.

You can always put a blank plate over the old one and save yourself a mess of plastering.

Project Hail Mary?

Bobaverse series.

The Bob series by Dennis E. Taylor is amazing, and I highly recommend it. Very positive, very creative.