🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

vibe coding is all fun and games until you get gaslit by claude and chatgpt into the wrong path. sometimes you just gotta sacrifice several days and go straight to the source and read the damn library code

1/ Today, we are thrilled to announce our $20M Series A, led by a16z crypto, with participation from SV Angel, Credibly Neutral, and Blizzard the Avalanche Fund. This funding fuels the development of our Workflow Protocol, so developers never need to write a smart contract again.

🔊 Excited to announce: Codegen is now GA!! 🎉 2 years ago, we set out to build an agentic developer. With Claude 3.7, it's now possible like never before - and it's absolutely killer. Build with Codegen across Slack, Github and Linear. It's transformed how we work. 👇

something people don’t understand/appreciate is that great infra makes 80% of this possible. Codegen has spent the last two years perfecting this infra and has nicely packaged it in the codegen-sdk

…

they lowkey nerfed sonnet 4. this thing can’t one-shot entire apps anymore. think i’ll stick with 3.7 over 4 for Bfloat Go for now

something i’ve been thinking about the last couple of months is how code is the ultimate and universal digital creative tool. and what this means for a world where code can be generated on-demand by superhuman ai codegen systems. dms open if you want to jam and explore this idea

"RL from a single example works" "RL with random rewards works" "Base model pass@256 can match RL model pass@1" "RL updates a small % of params" Recent papers all point in the same direction: RL is mostly just eliciting latent behavior already learned in pretraining, not

taking redbull as the perfect case study for the future of Cluely, it becomes clear as to why they’ll win regardless

for professional programming, “pseudocoding” > “vibecoding”

I wouldn't have said this 6 months ago, but I now believe all serious agents will be RL'd on their specific task. The gains are too easy and too huge to ignore. Either OpenAI et. al. will provide APIs to do this on-platform, or open source will win.