Eddie

@Suhail probably on applying the ai thinking to individual edge cases

If you defer thinking to ai, how will you exercise your mind? when will you think?

just noticed claude code now has a todo list. it's doing great

is # of vr device owned actually growing exponentially? or is the data ignoring returns?

so when i search with llm, it's citing other llm-written blog posts, which I am imagining is using other llm-written posts. feels like it's getting harder to get to the truth/source

wonder if owning your own model becomes more important in the app layer with llms. controlling the thinking steps, e.g., using tools, seems to make the output better? it's possible to emulate it with agentic steps but the degree of control over the output would still be limited

@jaltma @garrytan What drives him continuously to tackle the tough challenges he’s been active on

What should I ask @garrytan on the podcast tomorrow? @garrytan

creating a markdown todo list of coding task has worked better than talking to 3.7 sonnet directly on @cursor_ai . latter causes too much hallucination and random walks. potentially because i write instructions better if its in document format vs chat

if gogh painted golden pavilion in kyoto

amazing how all these work

ghibli isn't the only style

wow, this is impressive. consumer grade voice companion

Introducing Claude 3.7 Sonnet. Our most intelligent model to date and the first generally available hybrid reasoning model in the world.

@vrcprl @lingodotdev @Cornell @kiwicopple @greybaker @jakemintz @kul @RichAberman @ooshma @garrytan @koomen @dflieb @Initialized Congrats!!

You can just... build things! I was born in a tiny European village that nobody had heard of. Today, we raised $4.2M for @lingodotdev to scale our AI Localization Engine! 🤯 Excited to see how far we've taken our @Cornell hackathon project! Greatful for all the support! 🥹

interacting with the best startup in any field as a customer feels different. it teaches a lot about dedication to quality and committing to customer

but you cant do it without being able to read code or know whats good/bad. i agree with the sentiment that its still better to learn to code

cursor is so well designed, i dont feel like im coding lol all i need to do is understand the code, tell ai to generate, and then check + refactor the output

spent some time last night hacking with realtime api and really struggled making it follow instruction + rag. is it a 4o issue or realtime api issue?

@igrowsaas openai sets tiered limits on API access. o1-preview is apparently only accessible to Tier5 now

@Edd_Shim What's tier5?

hmm... have tier5 but can't access o1-preview yet

@JPBrebner Thank you for sharing this! Is there a common cadence for each sprint that you see among the great founders? (e.g., 1 ~ 2 weeks/sprint) Or would it depend on the team size and the product? I always feel the need to move faster, but not sure how fast.

In -1 to 0, your intensity should be just as high as it is in 0 to 1, but you shouldn't have an artificial deadline on when you'll reach 0. But you SHOULD give yourself deadlines for each new sprint, each new hypothesis, each turn in the squiggle. Take your time, quickly.

Great founders approach the -1 to 0 stage with relentless intensity and urgency. It's not a break before the real work begins. It is the beginning of the work. We call this velocity. Don't slow down just because you don't know the destination. But don't pick your intended destination for the sake of shorter trip.