everyday around lunchtime we all hop in the # food-ideas channel where Claude acts as a mediator to to help us figure out where to go for lunch

one tip for getting over heartbreak is to go outside and just look at the abundance of love in the world: a father pushing his son in a wheelchair, a couple holding their daughter’s hand, teenagers on a first date. impossible to feel sad while surrounded by a surplus of love

one tip for getting over heartbreak is to go outside and just look at the abundance of love in the world: a father pushing his son in a wheelchair, a couple holding their daughter’s hand, teenagers on a first date. impossible to feel sad while sounded by a surplus of love

Agency might require antagonism, some force to drive you forward. I don’t think it’s possible to exert agency without suffering. I really hope I’m wrong though

New Tidbit update is here. It's the best way to work with your team alongside Claude. - Mac app. - Invite your whole team. - Tool calling v1. - Channel-specific model behavior. We're stopped using Slack entirely. If interested, DM me.

ok but when will you add back text-davinci-003

here's my vibes-based gpt-5 review, mostly comparing it to opus 4. gpt-5 is a tool. it's utilitarian intelligence. it's cheap, fast, widely available. gpt-5 is tesla model 3 of intelligence. it *should* be the model most people use, and it probably will be. it gave good responses. it behaved predictably. it will be used by many companies and people. but... opus 4 is a much different model. it's slow, expensive, and in terms of pure utility it honestly doesn't provide much more to warrant it's higher price point. but man, opus's responses still have an indescribable feeling to them. it's the porsche 911 of LLMs. far too expensive, but otherworldly. maybe it's the better post-training. maybe it's just a bigger model. but i find myself wanting to keep using opus 4 even after trying to use gpt-5. in terms of real world utility, I GPT-5 feels like an improvement over the baseline. just not a pareto optimal improvement. it pushes the cost-adjusted intelligence frontier. but not the non-cost-adjusted frontier. that still belongs to Opus.

i love books they're just... waiting for you. they're not going anywhere. the perfect book for your current situation has been sitting on a shelf in an amazon warehouse for years. probably written probably before you were in second grade

i'm excited by Tidbit for one simple reason: most AI tools optimize for the median user. not Tidbit. we're deliberately building a deeply opinionated tool for outliers - the researchers, founders, scientists, and writers pushing their fields forward.

I'm looking for an engineer who believes AI interaction design is fundamentally broken. The right person: - Gets frustrated by today's AI interfaces - Prototypes interaction patterns that don't exist yet - Finds themselves tweaking CSS animation between model training runs If you've been waiting for the chance to build the AI interface you wish existed, let's talk.