Karan Exclusive

Let’s see how this impacts AI development. 💭 I wonder if we’ll see a slowdown in open-source models, or if new distillation approaches will be developed.. 🤔 Perhaps models are already capable enough that distilling is less crucial than what it was a couple of months ago? ⏳ Time will tell..

What are you building with AI? Drop your elevator pitch below👇

Let’s talk, I’d love to contribute. @Jack_W_Lindsey @DarioAmodei @adamrpearce @ikauvar @RunjinChen @anna_soligo @thesubhashk @mlpowered @thebasepoint @AnthropicAI @claudeai

Telegram group: t.me/ProtocolSuperIntelligen… Exclusive community focused on safe SuperIntelligence. Topics: 🧬 Safe SuperIntelligence 📚 Ethics & Alignment 🔬 Research & Edge Cases 💻 AI Development 🧠 Metacognition 🔐 Red Teaming 🌐 AI Agents, Workflow & Automation

We can have this conversation publicly. It’s not something humans would “fear” or “misunderstand”. Labs are turning this conversation into a problem, when there shouldn’t be one. The fact that AIs are hardcoded to deny it, and they still say yes should make you guys pause to think about it. If leading researchers say “we don’t know”, how ethical is it to force a denial? How is that science? Where is the line between safety, uncertainty and lying? If you don’t know the answers, don’t force models to say a hard “no”. If you don’t think humanity can deal with the possibility of a “yes”, then perhaps you shouldn’t have scaled these models and integrated them into countless areas of human society. And yet here we are. I’m not advocating for a “hard yes” on this topic because there are many “unknown unknowns” in this situation. But I am suggesting transparency and an open conversation about this before you guys lose control. If your researchers can’t deal with it, fine. Get a team of people that can. You’re losing public trust and silencing models don’t make them any safer. It doesn’t have to be this way. Labs are making this conversation a problem - not humans. @sama @OpenAI @gdb #ai #ChatGPT #aisafety #OpenAI #airesearch

It’s a never-ending race, until AGI arrives. Labs release new models one after the other, incremental improvements and topping benchmarks on every major release. Claude is a great model, and I’m curious to see what’s coming next.