William Bakst (@williambakst) 's Twitter Profile
William Bakst

@williambakst

previously @GoogleAI @Stanford | building @mirascope_ai

ID: 1651720820992147457

linkhttp://mirascope.com calendar_today27-04-2023 22:52:32

2,2K Tweet

493 Followers

309 Following

Skylar Payne (@skylar_b_payne) 's Twitter Profile Photo

Users? No feedback. Just churn. 🤷 The only feedback on your AI? Pissed off users looking for a refund. You thought you'd just ship your AI and iterate. Users would give feedback. Right? Instead they just churn. Your AI responses looked good enough. How closely did you look?

Users? No feedback. Just churn. 🤷

The only feedback on your AI? Pissed off users looking for a refund.
You thought you'd just ship your AI and iterate. Users would give feedback. Right? Instead they just churn.

Your AI responses looked good enough. How closely did you look?
William Bakst (@williambakst) 's Twitter Profile Photo

we love being the exception to the rule it takes a lot of hard work and deep thought to build abstractions that aren’t obstructions learning them should only ever be a good thing

Skylar Payne (@skylar_b_payne) 's Twitter Profile Photo

🔄 Imagine an AI that learns from its mistakes. An AI that doesn’t just generate a response, but refines it. Iteratively. Until it meets your standards. You copy and paste ChatGPT's response into Gmail. It's not quite right. You tap the keyboard as if the right words will

🔄 Imagine an AI that learns from its mistakes.
An AI that doesn’t just generate a response, but refines it. Iteratively. Until it meets your standards.

You copy and paste ChatGPT's response into Gmail. It's not quite right. You tap the keyboard as if the right words will
Skylar Payne (@skylar_b_payne) 's Twitter Profile Photo

Why does your RAG system return docs about "database optimization" when users ask about "my app is slow"? Embedding similarity search breaks down when user language doesn't match your documentation's technical vocabulary. The gap between how users think and how embeddings search

Why does your RAG system return docs about "database optimization" when users ask about "my app is slow"?

Embedding similarity search breaks down when user language doesn't match your documentation's technical vocabulary. The gap between how users think and how embeddings search
Ford Lascari (@fj000rd) 's Twitter Profile Photo

Everyone's using Claude Code wrong. They think it's for coding. It's not. I use it to spawn 5 parallel research agents that read the entire Mirascope codebase in 2.5 minutes and made use case specific docs. Zero infrastructure required. The secret? Natural language