Yes, it's been a way better deal to go for a subscription than pay as you go for me in the past. I had a month where I burnt through ~3.8b tokens which was somewhere in the ballpark of $8k worth of savings.

Now though I don't dare use spend tokens for basic note taking with Sonnet because I'm hitting the limit over a couple million tokens on the 20x plan, so they've really tightened the purse strings since November.

They must mean by creating a composite image with multiple in focus areas? Otherwise I agree, I can't see anyway that multiple exposures would help, at least from some light reading on Wikipedia - https://en.wikipedia.org/wiki/Multiple_exposure

> Quality of answers from quantized models is noticeable worse than using the full model.

This is the very reason I've heard I shouldn't use Alibaba!

I think some of this comes down to undeclared A/B testing. I've had the worst week of interactions I have ever had using Claude Code. The whole week whenever I have a session that isn't failing miserably I seem to get tapped for a session survey but on any that are out and out shitting the bed it never asks. It has felt a little surreal. I'd love to see a product wide stats graph for swearing, I would 100% believe that it is hitting an all time high but maybe I'm just a victim of a bad A/B round.

Oh I’ve been getting a lot more of those too lately even though I dismiss it every time. Wonder if I should report not satisfied every time so that I get routed to something better…

Here's some good looking anecdata:

https://x.com/fkysly/status/2044283560170004777

Not at all! My shop is having one of its best years ever. We've been switching to value based billing with our clients which has made them happier and us more profitable.

Using AI powered prototypes to sell clients on new features has gone really well for us. Many of our clients have presold themselves on adding AI features of some kind, which is nice because it generally means they will need support for these features going forwards.

What do you mean by value based billing?

I think it means charging based on the value of the work to the client not just the cost of doing the work.

That means if you get fast and efficient at doing something valuable for the clients you get to enjoy better margins because your costs are relatively low but the value is high so the customers still happily pay a good amount for it.

Interesting, I'm apparently a bot for saying I don't know.

A string of nonsense profanities was enough for me to marked a human.

The questions make me think someone’s trying to train an LLM cartel boss.

$7 a pop when other price watch tools are free? I'm not sure I'm understanding the value proposition here.

Has OpenAI banned it's use already? I hadn't seen that one come through yet.

Absolutely, but it occupies them later and only when needed. This is what I think they're driving at here.

but why can't I do the same with mcp? I just create a help() function that returns the help info?

That hypothetical might be fine, but MCPs do much more than that and their catalogs can be enormous. Here are some popular MCPs and the amount of context they eat before you've done anything with them:

  • Linear: 23 tools (~12,935 tokens)
  • JetBrains: 20 tools (~12,252 tokens)
  • Playwright: 21 tools (~9,804 tokens)

Github: 39 tools, 30K. I had to disable it.

Does anybody have a good SKILLS.md file we can study?

Absolutely! I now have Claude Code using `gh` and haven't missed the MCP. (If there are better CLI alternatives, I'd love to hear about them.)

you can; i've seen people put mcp access behind another mcp. I'm not sure how much success they got from it though

Their latest update just introduced indexing for handwriting + text, so now you're able to search through all of your notes! This has been a huge QoL upgrade for me.