Activity for cmgriffing.bsky.social

At Chain React last week, I went into ⁦@Powells⁩ for the first time and decided to snag these. They aren’t first prints but it’s nice to have some Vonnegut books to keep my Douglas Adams collection company.

I think the tweet was supposed to be "bad at the thing you're trying to use it for" or something like that. I didn't write the tweet, another person did, but I did write the article.

GitKraken@gitkraken.com

Hot take: the model everyone's memeing as "bad" might just be bad at the thing you're not using it for. Gemini ranks low on general intelligence. It ranks near the top on GitBench, our git task benchmark. Breakdown from our dev advocate @cmgriffing.bsky.social www.gitkraken.com/blog/gemini-...

AI will never be able to take the place of me housing an entire plate of mozzarella sticks

Preface: I’m an aging degen But I was reminded why sassys in Portland is the best. I lost myself in Korn’s Blind and more of the music more so than the show itself. Millenial paradise

SFO is so inclusive. They even have a restroom for “5 whiskeys deep” me.

Vibe coding your own SaaS replacement is only free if your time is worthless

Since I definitely don't do chicken as well as him, dillon_mulroy what is your preferred brining recipe?

Went and got measured/fitted. Does this count for “cocktail attire”? Also, tan or blue pants?

Nothing makes me feel more alive than an eye-twitch inducing CSS flaw

Dude... these clankers be clankin'. I thought I'd be done with stream hours ago, but their ability to string you along needs to be studied.

We were misled. Mr Ernst was not the bad guy.

I usually prefer it more rare but when cooking from frozen, I crank the temp a little.

How is it that the 6 months since the ralph loop feels like a longer amount of time than between now and last november?

I got Gemini 3.6 Flash and Flash Lite up onto GitBench and both are in the top 20 for success rate. When you do a 2d comparison, flash-lite is the #1 for API time and success.

Vibed a lot of coooode, some, some Pushed it up to prod, now it's down down Wish I had done a little review view Best to let Sol loop it now

A thing of beauty. I previously used Dredd.js for validating my OpenAPI specs. Schemathesis goes so much further!

Kimi is a very capable model for coding, but when it comes to git stuff it could be better. Still outscores Fable. This just emphasizes that we need to be spinning up subagents for more specific tasks rather than letting the more expensive models fail and recover.

Take it to the limit. Ride along the razor's edge We had a long running superpowers subagent skill started at ~73% of the session

Everyone is praising OpenAI dropping the session limit. tin-foil hat theory: They are doing that so they can reduce token budgets for their plans without it being as obvious.

This is happening to me again. Can't chat in the studio window. Windows rebooted and updated. Brave updated and ad blocking off. YouTube help?