2Lambroz 🐑 (🧑‍🍳🥩🤌)

Today is my 30th day in actively running lambroz
This post is to boost my self ego, fish for compliments & maybe to thank the friends i meet in the short journey.

📍In this thread
1)Why i started lambro
2)DeFi, NFTfi I covered
3)Friends who inspire me
4)What i learnt & my goals

pretty mad flex for someone to say my dick/pussy did that

Bruh u guys are animals

How does @Quotrons404 work?

how ppl check their cheap claude api:
"are u claude?"
"yes"

tldr on sterling, just launched by @miranetwork
a model name on an api is just a claim. same model gets served by many hosts, n they can run lower precision, ignore ur settings, break tool calls. still comes back 200 ok so u never know

researchers watched 189 endpoints for 4 months n found 37 suspected undisclosed changes. a 2025 benchmark had the same open model scoring 37% to 93% by host, some hosts fixed it after it got published

anthropic just put out a zk prototype that can prove which model answered, at 70-95x the gpu time. not deployed. testing from outside is weaker proof but u can run it today

so mira tests hosts 3x a day n posts the scores: does output match, do tool calls work, do settings apply, any silent failures. sterling is 1 key on top, change ur base url, get claude gpt gemini. up to 50% off advertised

deepseek v4 pro on 5 hosts: 1 scores 98, other 4 score 57-66. reference is that 1 host's own output so read it as "these differ" not "fake"

they score their own router too, not all green yet (gpt-6 luna failed, few runs, public beta). respect for leaving that up. still wanna know whos serving behind the key

but if hosts get scored in public they have to fix stuff. thats the part im watching
potential cataylst for $MIRA if that turns into real demand