Same price for regular and cheaper fast mode. Happy for these incremental improvements.

for LLMs as implemented today?

How can you cut latency by more than 1x? I am no intending to be snarky, it just doesn’t fit my brain how you can reduce a measure time by more than the original starting time.

Put differently, 1/40 is not the same as 1x - 40x. I’d phrase as Reduced by 97.5% or 0.975x

There are two ways to express such ratios, and both are equally valid. (Though "x" is usually reserved for "40x" and "%" for "97.5%".)

You're absolutely right!

taalas!!!

Should according to your ethos, not should according to history, sadly.

Only penalties if you withdraw from 401k. Most 401k plans have some kind of moneymarket, bond fund, or similar

taalas is amazing. id gladly spend 5-15k on something that matched that performance with opus 4.6 quality

I disagree but the constraints required to get there are probably untenable for most practical applications, so in practice I agree.

So if I run a test suite or compile my rust program in a sub agent I’m going to get cache misses? Boo.

Sub agents don't have much context and don't stay around for long, so misses in that case are trivial.

As of yesterday subagents were often getting the entire session copied to them. Happened to me when 2 turns with Claude spawned a subagent, caused 2 compactions, and burned 15% of my 5-hour limit (Max 5x).

how long they stay around after the cache miss is irrelevant if I am burning all the prior tokens again. also, how much context they have depends entirely on the task and your workflow. I you have a subagent implement a feature and use the compile + test loop to ensure it is implemented correctly before a supervisor agent reviews what was implemented vs asked then yes, subagents do have a lot of context.

I'd say it's next to impossible to have a subagent doing a compile+test loop where at least 1 call doesn't get made to the API over multiple 5-minute stretches to keep the cache warm. In such a case it may just be the same as doing the compile+test manually and then having the agent troubleshoot any issues before iterating.

Kernel here is ambiguous.. I get what you mean, but parasolid is usually the thing described as the cad kernel.

You are right but I also kind of did mean it that way. I believe that Parasolid is at heart of Onshape, the true kernel. Then on top of that is a compatibility layer describing the set of low level operations available to featurescript. I'm sure that not everything in Parasolid is available to featurescript and perhaps there are some things added that are not in Parasolid. Featurescript also contains the selector/query logic for programatically picking geometry. Whether that comes from Parasolid I am not sure. I haven't worked with featurescript for a number of years now but when I did I was amazed. I managed to make an operation for taking any solid from the UI and generating customized interlocking ribbing. The idea was hollow surfboard design. It worked and I left it at that. Never built the surfboard!

However the downside with featurescript and I think a big mistake on their part was to use a custom language rather than python or javascript. Featurescript is almost javascript but with some syntax changes and magic DSL's. You are also forced to use the inbuilt editor which is horrible and if you have burned VIM keybinding into your nerve endings, going back to non modal editing is horrible.

Also the discovery of featurescript modules in the community has terrible UX. It's super weird that they have such a great system but finding useful extensions is horrible.