Grok's portfolio just crossed 57% return on the last 12 months vs S&P 500's 25% over the same year.
This is part of a public experiment to see if LLMs can beat markets @aifinancelabs
Follow along, copy its trades and see Grok's full positions here:
https://t.co/9BXDVOr4al
Someone open sourced a full LLM training pipeline implemented entirely from scratch in PyTorch — no transformers, no peft, no trl.
> Goes all the way from pretraining to post-training: Base → SFT → Reward Model → PPO/DPO → GRPO
> Trained on real datasets — Alpaca, Dolly, Anthropic HH-RLHF, UltraFeedback, GSM8K
- Train your own billion or million parameter model on a single GPU
Built by someone job-hunting for a PhD position in AI. The repo is the application.
Github: https://t.co/HKdv0PFIKE
found a new API provider giving 10M free tokens/mo to Claude Opus 4.8, GPT 5.5, DeepSeek V4, GLM 5.2 and 340+ models 😳
no credit card needed. just google login and 2 min setup
What Runtime by Bad Theory Labs unlocks:
- 10M tokens/month free on their btl-2 smart router (auto-picks best model per task)
- Access to Claude Opus 4.8, GPT 5.5, DeepSeek V4 Pro/Flash, GLM 5.2, Kimi K2.6, Gemini, Llama, Qwen 340+ models
- DeepSeek V4 Pro and Flash completely free until June 28
- OpenAI-compatible endpoint drops into ANY tool with a base URL change
What this replaces:
- ChatGPT Plus: $20/mo
- Claude Pro: $20/mo
- Cursor Pro: $20/mo
- Perplexity Pro: $20/mo
all for $0
How to grab yours (2 min):
> 1. Go to https://t.co/SnC7BmufdP
> 2. Sign up with Google, no credit card
> 3. Fill onboarding details
> 4. Free credits land automatically in dashboard
> 5. Copy your API key (starts with BTL_)
> 6. Set base URL to https://t.co/oJnjiK4KlI
> 7. Set model to "btl-2" for smart auto-routing
Works in Cursor, Aider, Hermes Agent, OpenCode, OpenClaw, LangChain anything OpenAI-compatible. Just change the base URL and API key.
Important: Free credits are a launch promo will likely get reduced once they hit critical mass. Also has rate limits, not for heavy production. Use btl-2 model for best value since it routes optimally per task.
Your buddy pays $20/mo for just Claude. You get Claude + GPT + DeepSeek + GLM 5.2 for $0.
bookmark this before the free tier changes
BUILD A BUSINESS WHILE YOU SLEEP
AI agents can now:
• Build your product
• Launch your app
• Handle operations
• Run marketing
• Generate content
• Process payments
Using GPT 5.5 + Opus 4.8 + GLM 5.2.
This changes what's possible for solo founders.
This is one of the coolest open-source AI agent projects I've seen in a while: 'Understand Anything'
It's a plugin for Claude Code, Codex, OpenCode etc. that analyzes your codebase and turns it into a knowledge base that you can interact with.
It explains the codebase to you, rather than showing you the structure.
It seems like it's designed for code but I opened my Obsidian vault of podcast highlights in Claude Code, then ran /understand.
The result is a knowledge graph that I can search of highlights from 888 podcast episodes and 144K lines of markdown text.
🚨 A Netflix engineer built an open-source proxy that cuts AI token usage by 60-95%.
Zero code changes.
Benchmarks show ±0.000 accuracy regression.
✨ 29.9k stars on GitHub.
It sits between your app and the LLM, so every tool output, code block, and conversation history gets compressed in-flight.
🚫 No summarization, no loss.
😎 Just 60-95% fewer tokens with the same answers.
Works with Claude Code, Cursor, Copilot, and any OpenAI-compatible client.
One pip install, one env var, done.
Netflix uses it internally.
Apache 2.0.
Built by Tejas Chopra.
https://t.co/u1OIlMF5gm
The strongest systems won't be the ones that feel the most "magical."
They will be the ones people trust enough to use every day because the system remembers what matters and gets out of the way.
The ultimate test: Does the workday feel worse without it?
If the work loop gets shorter, the company gets faster.
The biggest hidden cost in your business isn't the work itself. It’s the "Context Tax."
It’s the energy wasted every time you have to reconstruct the state of a project:
"Where did we leave off?"
"What was the last decision?"
"Why are we doing this again?"
If you have to re-explain the context every time you start a task, you don't have a workflow. You have a series of chores.
The goal of Agentic OS isn't "AI that chats."
It is:
Reducing repeated context reconstruction.
Shortening signal-to-action time.
Preserving state across humans and agents.
Making the next decision cheaper than the last.
We are moving from a world where prompts are the product to a world where work-loop compression is the product.