Calum Best🔸

CoS @GiveWell research, i like most things but especially noodle soup and climbing. 🔸means i signed the @GivingWhatWeCan pledge — you should too!

[Calum Best🔸 @calphabetsoup]

@Tim_Hua_ @livgorton noooooooooo

[Calum Best🔸 @calphabetsoup]

.@bcherny I'm noticing a bug in the latest version of Claude Code Desktop (updated a few mins ago) where Claude doesn't execute plans it makes. Here's how the bug goes:
1. Claude proposes plan in plan mode
2. I click "accept plan and allow edits."
3. Claude thinks for 1-2s (thinking traces show it's aware that my plan has been accepted) and then stops without executing the plan.
4. I have to reprompt Claude to pick up the plan and run with it.

This happened twice today. It doesn't happen in Claude Code in the terminal; I'm only noticing it today while experimenting with the new Desktop update.

Is that fixable? Sorry if this is a terrible way to submit a bug report, and thanks very much for all your work on Claude Code!

[Claude @claudeai]

We've redesigned Claude Code on desktop.

You can now run multiple Claude sessions side by side from one window, with a new sidebar to manage them all.

[Calum Best🔸 @calphabetsoup]

humble flex @AnthropicAI

[Calum Best🔸 @calphabetsoup]

Come work with meeeee to hire researchers at GiveWell!! We're growing and we need to expand the team - no recruiting experience required (though ofc helpful!), amazing team, fun work, good comp. I started my time at GiveWell on our recruiting team and had a blast.

[GiveWell @GiveWell]

We’re hiring Senior Recruiters because growing our team is one of the most important things we can do to increase our impact.

Remote (US), $149k (SF/NYC) or $136k (all other US)

👉 Senior Recruiter: https://t.co/g86YUlpPUR
👉 Senior Recruiter, Research: https://t.co/PKLjR22SNr

[𒐪 @SHL0MS]

the lengthy illegible footnote caveat is a highly advanced protest sign move

[Calum Best🔸 @calphabetsoup]

@rSanti97 @aarmlovi hell yeah, congrats!!

[Calum Best🔸 @calphabetsoup]

@VictorKnox99 Signal-boosting @austinc3301 's reply - everything stated there matches my experience with the same problem in GiveWell hiring. https://t.co/G0FTifiJda

[Agus 🔸 @austinc3301]

SPAR got 2.5k mentee applications per round, and a mentor has to review (and test and interview) on average ~75 apps (and up to several hundred), all in a few weeks. Giving personalized feedback well (so as to not create a reputational and legal risk) takes thoughtfulness, experience, and typically reviews by other people, which makes it unfeasible at our scale.

That said, one thing that is probably viable is feedback bands. Basically ask mentors to label rejections in buckets (e.g. lacking technical skill, bad comparative advantage, needs more experience, bad fit with the project, etc), and you could have pre-approved templates for that type of feedback.

It’s unclear how useful this is for people, but at least it gives more signal to candidates. Kairos uses this in Pathfinder, our other big program, and people have liked it, but we’ve yet to pilot it in SPAR, in part because we’re under capacity and it’s on the bottom of our backlog. But this makes me think maybe we should prioritize it for the next round of SPAR. I could see this being opt-in for mentors as well, and we could have some kind of simple algorithm that aggregates the bands from different mentor decisions.

That said even this carries some downside risks for us, particularly that it makes it much more likely for people to try to contend results by arguing why the rationale given does not apply to them. SPAR has a policy of not allowing this, but it can be hard to deal with hundreds of requests like this all in a week (we already get a bunch even without personalized emails). I think we can manage, but there’s tradeoffs everywhere.

[Calum Best🔸 @calphabetsoup]

my airbnb host has clearly been to @LighthavenPR

[Calum Best🔸 @calphabetsoup]

@NathanpmYoung She should do Hinge and Bumble but not Tinder. Prob better pool on Hinge but that’s based on my/my friends’ anecdata and potentially location dependent (I’m in SF).

She should also get lots of feedback on her profile from male friends, and look at other girls’ profiles!

[Calum Best🔸 @calphabetsoup]

@MaxGhenis “obviously”

[Calum Best🔸 @calphabetsoup]

@RosieCampbell Do you have a repo on Github? Would love to steal this!!

[Calum Best🔸 @calphabetsoup]

@MaxGhenis building a thing where I can drop in meeting transcripts, memos, Slacks, etc., and process them into a growing web of tacit knowledge that’s called on when necessary — but I want to integrate Obsidian better!! Do you use a plug-in or just give CC access to the folder?

[Calum Best🔸 @calphabetsoup]

when will we have @ChipotleTweets Wrapped?? i want to know what time i’m usually hungry, what percentile customer i am at my local shop, how much protein i ate, my longest daily streak, what my usual order says about my demographics/personality, etc.

[Calum Best🔸 @calphabetsoup]

@Tim_Hua_ 🙀🙀 I’ll make sure you get a replacement hat, DM me to coordinate on address/timing?

[Calum Best🔸 @calphabetsoup]

@NatPurser yeah it’s absolutely horrific — whenever i draft emails for people i instruct them to copy paste with Cmd-Shift-V, but then that strips out links and rich text, bleh

[Calum Best🔸 @calphabetsoup]

@Miles_Brundage also frustrating is that the Google Docs API doesn’t allow access to the locations of comments, so individual vibe-coders can’t even make a bootleg version 😭😭😭

[Calum Best🔸 @calphabetsoup]

@huel tysm for doing that king

[Calum Best🔸 @calphabetsoup]

watching people consume Huel on Thanksgiving feels extra sacrilegious somehow