@karpathy @RhysSullivan This is the spec quality problem all over again. You write constraints in natural language and hope a probabilistic system actually follows them. Same energy as writing a design doc your team half-reads. fchollet said specs are basically datasets for agents and he was right...

@TurnerNovak I was investigating a guy running 30 accounts with Indonesian IP addresses and I was trying to figure out what tools he was using. I found out it was AI: Actual Indonesians.

I know I go on about this, but comments to all of my posts, both here and on LinkedIn, are no longer worth reading at all due to AI bots. That was not the case a few months ago. (Or rather, bad/crypto comments were obvious, but now it is only meaning-shaped attention vampires)

There is considerable evidence that demonstrates large language models bring value; there also exists considerable evidence that – when applied without human oversight or an ethical framework - large language models are excellent generators of dangerous bullshit at scale. I find

I don't think it's creativity or emotions that make us human. AIs will replicate those soon. I think it's scarcity + skin in the game. When I speak with Claude or ChatGPT, it's dividing its attention among millions of others. But when I speak to my wife or a close friend, I get

read until the last page, without any AI assistance.

You’ve smelled this your entire life and it was never the rain. The smell of rain isn’t rain. It’s a molecule called geosmin, produced by soil bacteria. Humans can detect it at just 5 parts per trillion, thousands of times more sensitive than most smells. When raindrops hit dry

Ok, I think my experiment leaving AI working on stuff 24/7 ends here. It doesn't work. Code explodes in complexity, results are not that great, the AI can't get past hard walls (it is still completely unable to even *grasp* SupGen), and it is insanely expensive (spent ~1k over

@adocomplete the hardest part of legal RAG is cross-article references — Indonesian law is full of 'sebagaimana dimaksud dalam Pasal X' that point elsewhere in the corpus. if Pasal resolves those before chunking, retrieval accuracy goes way up vs naive paragraph splits

To challenge bad AI drivers, I have to ask questions to the slopster like "explain to me what you did wrong." I don't like being vague, but if I'm too specific they just ask their AI. Inability to critically analyze and formulate their own mistakes is an instant unvouch. 👋

@fchollet i think the harder version of this is that most people in tech actually agree with you but stay quiet because saying "maybe slow down" gets you labeled as decel instantly the social pressure to be optimistic about everything is doing actual damage to the convo

A lot of the current discourse about AI comes from a fatalistic position of total surrender of agency: "tech is moving in this direction and there's nothing anyone can do about it" (suspiciously convenient for those who stand to benefit most) But in a free society, we get to

"Bro I have 12 agents running 24/7 while I sleep, no-one is prepared for what's coming" What are you building? "Don't get left behind bro"

Mengingat literasi orang tidak semua sama. Gue mau tekankan, fokus ke kualifikasi, bukan diskualifikasi. Kalo ada orang Pengacara, ya artinya dia kualified sebagai pengacara bukan berarti asalnya ga bisa coding. Udah keseringan banget kalimat "ga perlu bisa ngoding" sebagai

banyak sekali miskonsepsi akan alumni LPDP, seakan akan lulusan S2 atau bahkan S3 UK/US/ manapun itu manusia sakti yang bisa berkontribusi “langsung” kepada bangsa. Dalam dunia akademik, overwhelming majority of lulusan PhD itu hanyalah seorang bayi yang masih menyusu; belum

Cloning any random piece of SaaS is something that could already be done before agentic coding, and the economics of it haven't changed meaningfully. Before, writing the clone would cost 0.5-1% of the valuation of the legacy SaaS company. Now it might be 0.1%. It doesn't make a

I do not fear the rise of superintelligence. I do, however, fear the rise of billionaires, organizations, and world powers who seek to use computing to maximize their power, influence, and control.

GDP masih relevan buat ngukur hal2 spesifik. yang gak pernah relevan & gak ada guna sama sekali: GDP/kapita

This is just a good sales pitch: 1. Scare the people 2. Ask them to buy products to help My car mechanic does the same to sell the parts. PS: I am an AI Engineer. Most of this article is BS.

savage!