Before you continue to YouTube

Should we design for iffy internet?

Around 97% of US households have internet access, but bandwidth and latency are likely worse than you expect.

Stop alle regole AI: Usa verso il divieto per 10 anni

Procuratori generali USA si oppongono alla legge federale che vieterebbe agli stati di regolamentare l'intelligenza artificiale per un decennio intero.

Fine-Tuning LLMs is a Huge Waste of Time

People think they can use Fine-Tune for Knowledge Injection. People are Wrong

Eight Reasons Why Quadratic Funding Is Not Optimal

Introduction Quadratic funding has received a lot of attention recently as a mechanism for funding public goods—especially in the cryptocurrency space. QF is appealing because it is theoretically optimal under certain assumptions1.\nThe problem is that these assumptions don’t ever hold in reality.\nThe theory behind QF is sound and elegant, and the authors of the original paper are clear about the assumptions. They don’t claim they are likely to hold in reality, and warn about the consequences when they don’t hold. Unfortunately, practitioners have sometimes been too enthusiastic, implementing QF in settings where theory actually predicts poor results.\n

Tokenization for language modeling: Byte Pair Encoding vs Unigram Language Modeling

Tokenizers used by the best-performing language models (Bert, GPT-2, etc.) poorly reflect the morphology of English text. I had hoped to use some quarantine time to design one that more closely aligns to relationships between wordforms. But Kaj Bostrom and Greg Durrett beat me to it and so this blog post materialized instead. I add some additional motivation, evaluate both methods against ‘gold standard’ tokenizations, and speculate about what might come next.

Human coders are still better than LLMs - <antirez>

Digital technology can be harmful to your health

Experts at a Zócalo/UCLA event point to lack of sleep, weight gain and other issues