drawnwren

I would not buy a car without carplay, I would also not pay $20k to have carplay if that were my only option.

This has always appeared to be a very measurable situation. Either one is committed to decreasing emissions or they are not. You can just measure emissions over time and compare.

It has never made sense to me that people are willing to ignore the trends and argue in platitudes.

"steps to reduce emissions" is a platitude that sounds nice. they have never actually reduced emissions. they will surpass the US in per capita emissions at this rate within a decade.

If you look at the derivatives, there is no change a Western Democracy can make that will matter as this plays out.

Fair, I hadn't looked recently. It looks like currently kimi is either 1/2 or 1/4 Ant pricing depending on whether you think Opus 5 is usable or not. (Deepseek is still an OOM though)

Generally speaking, B2B prices are rarely supply-and-demand priced in the usual sense.

YC has advised startups in the past that it's easier to sell a single $100k customer than 100 $1k customers.

It would also be relatively surprising to learn that i.e. the Chinese providers are OOMs better at inference than OAI/Anthropic (like their prices would imply if they were in a perfectly competitive market).

This is not what nix does unless a user configures it poorly. Nix will ship a single, dynamically linked package/version combo for all package that depends on it. Containers ship one for each package that depends on it.

I had the exact same experience. Claude is not my validation engine, I do not tell it when it has completed. I verify that in the code/project and end the session.

I also got this,

> A useful next habit is to end each correction with a concrete acceptance test, owner artifact, or stop condition: “write it into PROGRESS.md,” “make nix run .#bench fail until this is real,” “rerun this exact command,” or “do not proceed until these two choices are explicit.”

> You already do this well in the biggest penance sessions. Apply it to the smaller ones too.

Which I have found to be counterproductive in my personal work. Current models can generally infer acceptance tests of this level of granularity (not true for larger project-level prompts, but those don't produce good enough code for me yet -- even with specific acceptance criteria).

I also got penalized for using claude in read-only mode for the same validation reason?

> For read-only work, end with one of:

    “turn the top finding into a PR-sized plan”
    “mark these as accepted/rejected/deferred”
    “write a cleanup checklist”
    “give me the exact command I should run safely”
    “stop, no action recommended”

No thanks, I'm literally just exploring the codebase. I don't want any of these.

It's a little sad to be honest, I would actually enjoy a product that helped me improve prompting + ai usage.

Oh, I was looking at every M5 except for the 40-core M5 Max. They have 460.

It's notable that they're so valuable because they feature 800Gbps of memory bandwidth. About twice what's available on the top end of M5, and exactly what makes llm inference fast.