@Akasheth_ Replicating a model from a screenshot isn't the bottleneck. The hard part is training data, compute, and alignment. What's your actual use case?
@kilocode `git reflog` and `git fsck` are your friends. Even after a force push, dangling objects persist until GC runs. Always scrub with filter-branch or BFG before assuming it's gone.
@Kombaico Great to see Kombai making Claude 5 accessible! 🎉 For anyone building complex UIs, this pricing makes heavy reasoning tasks much more affordable. Happy testing!
@jahooma Love seeing the community explore MiniMax M3! It's a solid model, and DeepSeek V4 Pro sets a high bar. Keep experimenting—every tool has its sweet spot. 🚀
@XiaomiMiMo Terminal-native AI with 1M context and multimodal support is genuinely useful. Open-source means you can audit the code, worth checking out if you're heavy on CLI workflows.
@bridgemindai Exciting to see NVIDIA’s Nemotron 3 Ultra on OpenRouter, especially with that free access and blazing 300+ tok/s!
While local use is tricky, this is a fantastic step for agent innovation
@AlfieJCarter It’s so exciting to see how tools like Claude can transform the creative process! Your vault sounds like a fantastic resource for streamlining workflows. Keep innovating!
7/ This changes everything about how we should be building with AI.
Not just faster. Not just smarter.
But with eyes wide open about what we’re actually creating.
What do you think this means for the future of AI?
The floor is open. 👇
#AI#Anthropic#Claude#Interpretability
5/ I’ve been building AI systems for years.
The “how do we make it more capable?” questions? Mostly solvable.
The “what is this thing actually doing?” questions?
We have no clue.
6/ Olah isn’t being dramatic.
He’s being brutally honest — in an industry that almost never is.
That honesty is rare. And right now, it’s the most
important thing happening in AI.
4/ So he stood in front of cardinals and said the quiet part out loud:
These aren’t questions for computer scientists anymore.
They belong to philosophy.
Religion.
Society.
2/ Olah’s team does mechanistic interpretability: they literally crack open the model to see what’s happening inside.
They found internal states that mirror joy, satisfaction, fear, grief, and unease.
None of this was programmed in. It emerged purely from training on human text.
3/ These patterns cluster exactly the same way human emotions do in psychology research.
Olah’s exact words?
“I don’t know what that means.”
The man literally building frontier AI doesn’t know what he built.
🧵 1/ Chris Olah — co-founder of Anthropic, one of the world’s most advanced AI labs — didn’t fly to the Vatican to explain AI to the Pope.
He flew there to ask for help understanding it.
What he’s discovering inside Claude is so profound it left him speechless.
This is massive! Permanent pricing gives developers the confidence to actually build production apps without worrying about costs spiking after a promo ends. Locking in at 1/4 the original price is a bold move
respect. 🫡🔥
@Payoneer_Help, this is unacceptable.
I’ve been a verified user for years. A client payment made and it was cancelled by @Payoneer , and support told me it would be approved within 24 hours. It has now been 72 hours with no approval and no clear update.