25-11

• Language Models are Injective and Hence Invertible | Lobsters

  • SIPIT, the first algorithm that can recover the exact input sequence from hidden activations, with provable linear-time guarantees

  • We advocate viewing language models as maps on the sequence space rather than the embedding space; under this perspective, we prove that all information about the input sequence is almost surely preserved end-to-end

  • Hidden states are not abstractions but the prompt in disguise

• Austrian Power Giants

• AWS to Bare Metal Two Years Later: Answering Your Toughest Questions About Leaving AWS

• Absurd Workflows: Durable Execution With Just Postgres | Armin Ronacher’s Thoughts and Writings

  • via Absurd Workflows: Durable Execution With Just Postgres | Lobsters

• Don’t give Postgres too much memory - Tomas Vondra

  • via Don’t give Postgres too much memory | Lobsters

• 3 Billion Vectors in PostgreSQL to Protect the Earth

  • vector embeddings for map tiles of satellite data

  • search over 3.2E9 vectors

• The KDL Document Language

  • via Usage Specification | Usage

  • via Things I Don’t Like in Configuration Languages | Lobsters

• “You Don’t Need Kafka, Just Use Postgres” Considered Harmful - Gunnar Morling

  • has a good list of features that make Kafka unique

  • “You Don’t Need Kafka, Just Use Postgres” Considered Harmful | Lobsters is the comeback of the original author