fritzo

This is how you get token laundering

Right, hn is 50% collaborative filtering and 50% dang

Would those agents happened to be named frk_ai_8b2e and that platform news.ycombinator.com?

I've long been terrified of the existence of adversarial prompts that can get me to say anything, that dogs can lay eggs, that there are five lights, that here's my bank info

It's so attention needy, and it's transforming our culture

Linda Palmer also does "aha" research at UC Irvine https://brain.uci.edu/uci-brain-pilot-grant-investigating-ex...

The only grave being disturbed is Robert Moses' by his turning

And the fourth oldest vice is gossip, the original social media, talking trash while you pick bugs out of your friend's hair

Also curious about latency. In the past I've worked around latency using video sensors for high-bandwidth high-latency features, then literally glued a contact mic to my interface to get low latency tap detection. How does the Board hide latency?

link is 404 for me

Missed opportunity to call it banana-seeds-400k

Unpopular opinion: hundreds of years from now, loss of mammalian species will seem like sentimental naval gazing when our descendants consider the millions of strains of fungi, bacteria, archaea, and viruses we could have saved, were it not for our micro-blindness.

The batteries belong on a hip belt, not down where the foot swings

> rechargeable cuff battery that seamlessly integrate with a carbon fiber–plated running shoe

Look how far we've come! Now we add random .unsqueeze(-1) and .permute(-1,-2) until our PyTorch models run without shape errors!

It looks like almost every AI researcher and lab who existed pre-2017 is now focused on transformers somehow. I agree the total number of researchers has increased, but I suspect the ratio has moved faster, so there are now fewer total non-transformer researchers.

> “They didn’t apologize. They just told me it was protocol. I was expecting at least somebody to talk to me about it.”

I wonder how effective an apology and explanation would have been? Just some respect.

Afra Wang's article:

> in this piece, i trace the intellectual map of china’s tech founders by asking a simple question: what books have shaped their worldview?

> what emerged is that the “china tech canon” shares deep roots with the “silicon valley canon” ... works by paul graham, jim collins, peter thiel, and biographies of elon musk (and many others) circulate widely. but the chinese canon also branches into its own distinct traditions: what i call the red canon, the grey canon, and the fanstasy canon. alongside the western playbook of disruption and ambition, chinese entrepreneurs also draw from maoist texts, confucian ethics, martial-arts fantasies, and liu cixin's ruthless universe.

Did we read the same Accelerando? The book I read was filled with increasingly powerful scam artists, increasingly destructive power struggles, increasing energy demands.

Tacit knowledge

Zuse was a genius, but I feel Moses Schönfinkel should get credit for the first programming language with his S,K combinators, pre-dating even Church's λ-calculus.

https://writings.stephenwolfram.com/2020/12/where-did-combin...

Yes, Sonnet 4.5 was noticeably worse than Opus on my codebase.

It looks like Claude Code switched the default model to Sonnet 4.5, even for max users who are paying for Opus. I had to set the environment variable ANTHROPIC_MODEL=opus to get opus back. Sonnet 4.5 indeed acts like a drunken monkey on the math-heavy legacy C++ codebase I'm working on.

You're missing the point. We want AI to piggyback on our open source code, because then thousands of developers around the world can piggyback on that AI. That AI is a boon for users, and is just as useful as documentation and a discussion forum.

2,3,4. While my agent refactors code, I do housework: fold laundry, wash dishes, stack firewood, prep food, paint the deck. I love this new life of offering occasional advice, then walking around and using my hands.

I've rented an "extra room" a couple times. My partner and I had rent control on our 1BR apartment, but we wanted extra space. It was cheaper to rent a non-bedroom room in a nearby house, than to lose rent control and upgrade to a 2BR.

> We have a pretty good mathematical representation of human knowledge...

> Swiss Cheese — there's a lot of cheese, but there's also a lot of holes.

Such hubris! Human knowledge is not like Swiss (mostly known with a few gaps). Human knowledge is like a high-dimensional fractal broccoli: a stem of hard-won knowledge surrounded by a vast frontier of unknown. The hardest part of pushing on the frontier is deciding where to push. Deciding what to research requires taste.

No code, low code, vibe code. The narrative outside tech circles is "empowering creators"

It's like a company with no employees. There are still people performing services, but on temporary contracts.

Have worst case optimal join algorithms made a practical impact, since the linked article's first publication in 2015? I've seen them in the context of egglog, but are they used in real world database management systems?

https://en.wikipedia.org/wiki/Worst-case_optimal_join_algori...

A torus would have been easier.