How Modulo Lets LLMs Steer Their Own Activations
Modulo lets an LLM request a temporary activation shift. See how control vectors work and what models chose in the exploratory tests.
2 min read2 min listen
How agent harnesses, models and AI tooling actually behave once you build on them.
September 2026
Your next AI fix may be hiding in the original research. Revisit 6 landmark works for better prompts, useful retrieval, and the limits people leave out.
6 posts
September 2026
Jev promises fast, typed decisions. These explanations separate its interface, vendor benchmarks and confidence scores before you change an application.
3 posts
Modulo lets an LLM request a temporary activation shift. See how control vectors work and what models chose in the exploratory tests.
2 min read2 min listen
Rich Sutton argues that search and learning eventually beat hand-built knowledge in AI research. The distinction matters when deciding which rules to keep.
3 min read4 min listen
Jev is trained to return calibrated decisions. Learn how probabilities differ from confidence, and what a gateway still has to decide for itself.
4 min read5 min listen
DeepMind matched Gopher’s training budget with a smaller model and more tokens. The result explains why parameter count alone is a poor guide.
4 min read6 min listen
TypeSafe reports similar agreement at lower cost in one comparison, and better agreement from another model. Keep the two comparisons separate.
3 min read5 min listen
The original RAG model combined document retrieval with generation. Its factuality gains and index-swap experiment show what that split can and cannot do.
4 min read7 min listen
Jev returns choices and probabilities for software to act on. Its decision interface explains both the speed claims and the jobs it cannot take over.
4 min read7 min listen
Wei and colleagues found gains from worked examples on difficult tasks, but losses on smaller models. What their 2022 experiments actually establish.
4 min read8 min listen
The 2020 paper tested one model across many tasks without fine-tuning. Its gains and failures explain the promise and limits of in-context learning.
4 min read7 min listen
The 2017 Transformer traded sequential processing for attention between tokens. See how that speeds training and creates a long-context cost.
5 min read10 min listen