working notes. notes on my research, some with fuller experiments than others. not all blogs have an associated paper. a blast to the past twitter view
Speculative programmatic tool calling pre-launches tool calls from a partially generated REPL program, while the harness is still producing tokens, instead of waiting for the whole generation to finish.
Modern post-training brute-forces ever more environments and ever longer horizons, because frontier Transformers are still poor at compositional generalization. A good harness makes every model call locally in-distribution, so what it learns carries across lengths and domains.
We underestimate how good language models actually are, and they are inhibited by how we use them. On LongCoT, frontier models fall short of compositional reasoning tasks whose pieces they can already solve in isolation.
AI models are already good enough for the next leap in capabilities. The mismanaged geniuses hypothesis says frontier language models are severely underutilized, because each individual call is used poorly.
