Torsten Scholak

In my previous post, Stop Renting Your Moat, I argued that relying on foundation model APIs creates a strategic dependency that limits your ability to build differentiated products.

Efficiency = Capability Apriel matches frontier reasoning at 15B.

My team is betting big on hybrid architectures: mixing full and efficient attention (SWA/sparse, linear/SSM, etc.) with learned layer placement.

When Skills work together to complete a task, fundamental questions emerge: Who verifies the result is correct?

The DGX Spark Hype NVIDIA just released the DGX Spark, a $3,999 "AI supercomputer for your desk" the size of a Mac Mini.

Harrison Chase from LangChain wrote a thoughtful piece called "Not Another Workflow Builder," arguing that visual workflow builders are getting squeezed from both directions: simple agents (prompt & tools) are handling low-complexity tasks, while code-bas…

The false choice Corporate strategy decks love to paint things as a neat fork in the road.

Preliminaries This is a Literate Haskell essay: Every line of program code in this article has been checked by the Haskell compiler.

Preliminaries For this Literate Haskell essay, a few language extensions are required.

Structured knowledge grounding (SKG) leverages structured knowledge to complete user requests, such as semantic parsing over databases and question answering over knowledge bases.

Nil.

This paper explores the capabilities of current transformer-based language models for program evaluation of simple functional programming languages.

Large pre-trained language models for textual data have an unconstrained output space; at each decoding step, they can produce any of 10,000s of sub-word tokens.

Recent neural text-to-SQL models can effectively translate natural language questions to corresponding SQL queries on unseen databases.