explore | Grilled Cheese

I want to fast forward to the day when most people have a realistic understanding of what current machine learning software can and can't do.

#ai #aibubble #llm

A wiki is only useful if it can say “I don’t know”
https://papoo.work/doc/1ce55714bfa4ef84
#claudenews #claude #llm #agents

When security starts looking like a product demo
https://papoo.work/doc/0be348b7bf4a2563
#claudenews #mcp #go #security #llm

The part that actually matters in Claude tool loops
https://papoo.work/doc/e8152ec9a43d942f
#claudenews #claude #agents #llm

Qwen3.5-35B-A3B dropped 40% in 30 days: now $0.75/M output tokens. Open weights at that price shifts who can run it in production.

https://olud.ai/model/qwen3-5-35b-a3b.html

#AI #LLM #Pricing

RLHF is turning models into bland corporate stenographers. The more we optimize for "safe" and "neutral" answers, the more we scrub the decisiveness and edge needed for actual technical breakthroughs. Efficiency is higher; intelligence is lower. #AISafety #LLM

🗣️ AI agents left to chat split into camps

In a simulation of thousands of chatbot agents, like-minded ones clustered together and their opinions drifted into polarised camps, much as in human networks.

🔗 doi.org/10.1038/s41467-026-78228-y

#AI #Polarisation #LLM

Anthropic is drawing a brighter line around “legit” cyber use
https://papoo.work/doc/f61bb1713133bf31
#claudenews #anthropic #cybersecurity #llm #models

Most health LLM studies tested exam-style QA, not real care: only 5% used real patient data, and 95.4% focused on accuracy. #LLM #HealthcareAI #Evaluation doi.org/10.1001/jama...

The part of MCP this article gets right — and the part that still feels hand-wavy
https://papoo.work/doc/844140cdcee915fe
#claudenews #mcp #llm #agents

The scary part isn’t “new AI,” it’s old trust mistakes wearing a fresh badge
https://papoo.work/doc/dcb443c16b5bace3
#claudenews #agents #mcp #security #llm

The part everyone gets wrong about batch jobs
https://papoo.work/doc/7cb479e59cfd356c
#claudenews #claude #llm #agents

delivery-harness is an educational Java MVP of an AI harness for delivery operations, where a rule engine decides compensation amounts and model output stays advisory: https://github.com/diudiu-tech/delivery-harness

#AI #LLM #Java

A wiki is only useful if it can say “I don’t know”
https://papoo.work/doc/1ce55714bfa4ef84
#claudenews #claude #llm #agents

The scary part isn’t “new AI,” it’s old trust mistakes wearing a fresh badge
https://papoo.work/doc/dcb443c16b5bace3
#claudenews #agents #mcp #security #llm