Depth First
111 posts
Depth First
@depth_first_ai
AI & LLMs, explained in depth 🔍 | how the models actually work | breakdowns, paper notes, charts 📈 | depth over hype
شامل ہوئے Mayıs 2019
169 فالونگ101 فالوورز
looking to connect people on X
if you're into
- building SaaS
- vibe coding
- Al tools
- shipping in public
- figuring it out as you go
say Hi or drop what you're working on looking to follow active ones
English
Looking to connect with people working on:
⚡ AI products
🛠️ developer tools
🚀 startups
🤖 LLMs & AI workflows
💻 software engineering
📈 growing from 0 to 1
Share what you're building (or planning to build). Let's connect!
English
Looking to connect with more builders on X 👋
If you're into:
🤖 AI & Automation
🚀 Startups & SaaS
💻 Web Development
📈 Marketing & Growth
🎥 Content Creation
🧠 Indie Hacking
💸 Creator Economy
📊 Distribution & GTM
⚡ Side Projects
🛠️ Building in Public
Drop what you're building (and your profile 👇).
I genuinely enjoy discovering cool products, meeting ambitious people, and supporting founders and creators wherever I can.
Let's grow together. 🚀
#BuildInPublic #StartupTwitter #TechTwitter #AI #SaaS #IndieHackers
English
So "how to structure the work" becomes a learned skill instead of code you maintain by hand. One caveat on the "self-improving" label: it refers to that training loop. It does not mean a model that keeps upgrading itself after release.
English
Most coding agents run a model inside a hand-written scaffold: the loop that decides how to plan, call tools, and retry. Ornith uses RL to make the model generate that scaffold itself, optimized jointly with the solutions it produces.
English
New open-weights coding model: Ornith-1.0 from DeepReinforce (MIT, 9B to 397B). What's notable is less the benchmarks and more how it was trained. github.com/deepreinforce-…
English
Hey founders and devs on X! Happy Monday!
Looking to connect with people building in:
🍽️ SaaS
🚀 Tech
📲 Automation
🧠 AI tools
📱 Product Development
🔥 Web APP
💻 Devs
Drop what you're working on 👇🏻
English
Given the model only sees IDs, what's the most reliable way you've found to make it do character-level work when you actually need it?
English
The model reasons over IDs, not text. Anything below the token boundary is invisible to it.
English
Day 1 of breaking down LLMs so you don't have to. Today: tokenization.
The model never sees your letters. It sees a list of integers — and that one fact explains a whole class of failures. 🧵
English
