Jud Stephenson

[Jud Stephenson @JudStephenson]

Fable is quiet as a church mouse

My children will never know what its like to find the smoking gun!

[Jud Stephenson @JudStephenson]

/model claude-fable-5

[Jud Stephenson @JudStephenson]

@nvk openai also does little things (maybe by accident) like reset usage early

[Jud Stephenson @JudStephenson]

fwiw, i think this turned out to be gstack's skill routing wanting to /snapshot

[Jud Stephenson @JudStephenson]

anybody else notice claude (1m context) is VERY eager to 'pick this up in the next session' when ctx > 180k? I can only assume an agent hardcoded a number to sidestep a broken test at some point.

[Jud Stephenson @JudStephenson]

Opus 4.8 really wants to use GNU cli command flags on my Mac. Feels like the Sonnet 4 days in some ways tbh @bcherny

[Jud Stephenson @JudStephenson]

pair programming with claude is garage beers with your buddy. except you're the only one drinking and he never gets tired. tonight we're reviving a dead 48v golf cart battery.

[David Scott Harris @HVNYrefugee]

One of these is Hokkaido, Japan.
One of these is TN, USA

Which one is which?

[Jud Stephenson @JudStephenson]

@MacroTactical yea i'm not a hater, something is up.
https://t.co/OquVpDN0MA

[Jud Stephenson @JudStephenson]

ok real talk, claude code opus 4.6 is garbage now. routinely answers without actually reading files. can't interpret images. reasoning is max. @bcherny @AnthropicAI what is going on? defaulting to codex xhigh with much better results

[Jud Stephenson @JudStephenson]

ok real talk, claude code opus 4.6 is garbage now. routinely answers without actually reading files. can't interpret images. reasoning is max. @bcherny @AnthropicAI what is going on? defaulting to codex xhigh with much better results

[mickey friedman @mickeyxfriedman]

the current fear is is that AI homogenizes culture and turns humans into passive consumers

one counterpoint: in Go, human play showed very little improvement from 1950 to 2016 until alphago beat lee sedol - then human decision quality jumped. players started developing moves that were distinct both from previous human moves and from the novel moves introduced by machine intelligence

this seems more likely to me - fun times ahead

[Jud Stephenson @JudStephenson]

@VictorTaelin This 1000x
https://t.co/rTIyzjPNcB

[Jud Stephenson @JudStephenson]

watched one wrong comment propagate through five agents. an hour of confident wrongness until I manually reviewed. single line fix.

[Jud Stephenson @JudStephenson]

trying to avoid confirmation bias, but came here to complain about opus being very dumb lately (yes i always use max reasoning) and turns out everyone else feels the same way. was having claude do codex consults on almost every turn. then crazy idea -- what if I just used codex? turns out, codex is pretty awesome.

[Jud Stephenson @JudStephenson]

claude set my lineup today and we got the W

[Jud Stephenson @JudStephenson]

friday fun: `diamond-replay`

turn gamechanger baseball streams into analysis. tui for viewing. can also output json for things like generating youth specific game-health dashboards (e.g. https://t.co/USk3VQpglt)

https://t.co/kW7TpN6VV3

[Jud Stephenson @JudStephenson]

missing from gstack: /autobuild

an alias for: /loop 30m if we have nothing to do, build the next chunk. if we're stuck, run a /codex consult. when chunks complete run /simplify and /codex review. commit. repeat.

[Jud Stephenson @JudStephenson]

https://t.co/0mrM3IDtRL

[Jud Stephenson @JudStephenson]

mkdn 0.1.7 (+0.1.6).

- git-aware sidebar shows your current branch and dirty files now.
- footnotes click-to-scroll between reference and citation.
- new table layout engine.
- breadcrumb bar (cmd-j to jump to a section).

also fixed a link cursor bug that's been there since launch. turns out TextKit 2 can't do coordinate math.

`brew upgrade mkdn`

[Jud Stephenson @JudStephenson]

@Mayhem4Markets Qwen 3.5 is VERY forgiving with quantization

https://t.co/bJ9WM3PCwC

[Jud Stephenson @JudStephenson]

@TheAhmadOsman ive been optimizing 2B on transcription correction. there is so much knowledge embedded in the weights. pal-4 < 1gb, 65tok/s on m2, it is really incredible.

[Jud Stephenson @JudStephenson]

@0xSero Opus not afraid to get creative and bend the rules. Codex much more rigid. Both a strength and weakness