News for hackers. Stuff that matters.
Re:Claude Code Best Practices by simonw on Saturday April 19, 2025 @06:30PM (#43739993)
The "ultrathink" thing is pretty funny:

> We recommend using the word "think" to trigger extended thinking mode, which gives Claude additional computation time to evaluate alternatives more thoroughly. These specific phrases are mapped directly to increasing levels of thinking budget in the system: "think" < "think hard" < "think harder" < "ultrathink." Each level allocates progressively more thinking budget for Claude to use.

I had a poke around and it's not a feature of the Claude model, it's specific to Claude Code. There's a "megathink" option too - it uses code that looks like this:

  let B = W.message.content.toLowerCase();
  if (
    B.includes("think harder") ||
    B.includes("think intensely") ||
    B.includes("think longer") ||
    B.includes("think really hard") ||
    B.includes("think super hard") ||
    B.includes("think very hard") ||
    B.includes("ultrathink")
  )
    return (
      l1("tengu_thinking", { tokenCount: 31999, messageId: Z, provider: G }),
      31999
    );
  if (
    B.includes("think about it") ||
    B.includes("think a lot") ||
    B.includes("think deeply") ||
    B.includes("think hard") ||
    B.includes("think more") ||
    B.includes("megathink")
  )
    return (
      l1("tengu_thinking", { tokenCount: 1e4, messageId: Z, provider: G }), 1e4
    );
Notes on how I found that here: https://simonwillison.net/2025/Apr/19/claude-code-best-pract...
Re:Claude Code Best Practices by orojackson on Saturday April 19, 2025 @06:54PM (#43740108)
Not gonna lie: the "ultrathink" keyword that Sonnet 3.7 with thinking tokens watches for gives me "doubleplusgood" vibes in a hilarious but horrifying way.
Re:Claude Code Best Practices by 4b11b4 on Saturday April 19, 2025 @08:45PM (#43740664)
At this point should we get our first knob/slider on a language model... THINK

..as if we're operating this machine as analog synth

Re:Claude Code Best Practices by soulofmischief on Saturday April 19, 2025 @09:32PM (#43740878)
There are already many such adjustable parameters such as temperature and top_k
Re:Claude Code Best Practices by ljm on Sunday April 20, 2025 @08:01AM (#43743251)
Maybe a Turbo Think button that toggles between Ultrathink and Megathink.
Re:Claude Code Best Practices by antonvs on Sunday April 20, 2025 @09:18AM (#43743591)
If you use any of the more direct API sandbox/studio UIs, there are already various sliders, temperature (essentially randomness vs. predictability) being the most common.

The consumer-facing chatbot interfaces just hide all that because they're aiming for a non-technical audience.

Re:Claude Code Best Practices by ______ on Sunday April 20, 2025 @09:47AM (#43743737)
I use a cheap MIDI controller in this manner - there is even native browser support. Great to get immediate feedback on parameter tweaks
Re:Claude Code Best Practices by coffeebeqn on Sunday April 20, 2025 @10:02AM (#43743816)
A little bit of the old ultrathink with the boys
Re:Claude Code Best Practices by stavros on Sunday April 20, 2025 @10:04AM (#43743831)
Shot to everyone around a table, thinking furiously over their glasses of milk
Re:Claude Code Best Practices by pyfon on Saturday April 19, 2025 @09:23PM (#43740843)
Weird code to have in a modern AI system!

Also 14 string scans seems a little inefficient!

Re:Claude Code Best Practices by Aurornis on Saturday April 19, 2025 @10:46PM (#43741154)
14 checks through a string is entirely negligible relative to the amount of compute happening. Like a drop of water in the ocean.
Re:Claude Code Best Practices by bombela on Sunday April 20, 2025 @02:45PM (#43745653)
Everybody says this all the time. But it compounds. And then our computers struggle with what should be basic websites.
Re:Claude Code Best Practices by anotherpaulg on Saturday April 19, 2025 @09:53PM (#43740972)
In aider, instead of “ultrathink” you would say:

  /thinking-tokens 32k
Or, shorthand:

  /think 32k
Re:Claude Code Best Practices by westoncb on Saturday April 19, 2025 @10:24PM (#43741086)
That's awesome, and almost certainly an Unreal Tournament reference (when you chain enough kills in short time it moves through a progression that includes "megakill" and "ultrakill").
Re:Claude Code Best Practices by orojackson on Sunday April 20, 2025 @02:34AM (#43741934)
If they did, they left out the best one: "m-m-m-m-monsterkill"

Surely Anthropic could do a better job implementing dynamic thinking token budgets.

Re:Claude Code Best Practices by Quarrel on Sunday April 20, 2025 @03:20AM (#43742129)
Ultrakill is from Quake :)
Re:Claude Code Best Practices by Schaulustiger on Monday April 21, 2025 @04:46AM (#43749672)
It is not. Quake had "Excellent" for two kills in short succession, but nothing else if you chained kills after that.
Re:Claude Code Best Practices by Quarrel on Thursday April 24, 2025 @03:42AM (#43780170)
Meh, ok?

This fits my memory of it:

https://www.voicy.network/sounds/o_2r357a-kKEe5PaiWyYnA-quak...

Was it quake 2?

Re:Claude Code Best Practices by mr-karan on Sunday April 20, 2025 @12:16AM (#43741482)
What I don’t like about Claude Code is why can’t they give command line flags for this stuff? It’s better documented and people don’t have to discover this the hard way.

Similarly, I do miss an —add command line flag to manual specify the context (files) during the session. Right now I pretty much end up copy pasting the relative paths from VSCode and supply to Claude. Aider has much better semantics for such stuff.

Re:Claude Code Best Practices by gdudeman on Sunday April 20, 2025 @10:26AM (#43743985)
Maybe I’m not getting this, but you can tab to autocomplete file paths.

You can use English or —add if you want to tell Claude to reference them.

Re:Claude Code Best Practices by mr-karan on Thursday April 24, 2025 @09:39AM (#43782651)
Never knew about `tab` shortcut, thanks for letting me know!

BTW, as I was using it today, I was quite surprised to see `@` working now. Turns out in 0.27.5 they added this feature: https://github.com/anthropics/claude-code/blob/main/CHANGELO... :)

Re:Claude Code Best Practices by nulld3v on Sunday April 20, 2025 @07:53AM (#43743221)
Waiting until I can tell it to use "galaxy brain".
Re:Claude Code Best Practices by NiloCK on Sunday April 20, 2025 @11:14AM (#43744304)
Slightly shameless, but easier than typing a longer reply.

https://www.paritybits.me/think-toggles-are-dumb/

https://nilock.github.io/autothink/

LLMs with broad contextual capabilities shouldn't need to be guided in this manor. Claude can tell a trivial task from a complex one just as easily as I can, and should self-adjust, up to thresholds of compute spending, etc.

Re:Claude Code Best Practices by carlmr on Tuesday April 22, 2025 @10:50AM (#43762901)
>LLMs with broad contextual capabilities shouldn't need to be guided in this manor.

I mean finding your way around a manor can be hard, it's easier in an apartment.

Re:Claude Code Best Practices by adastra22 on Sunday April 20, 2025 @01:30PM (#43745145)
“think hard with a vengeance”