Did Codex Reset
GitHub

EvoOntology paper reports self-evolving ontology layer lifts data-agent scores on DDR-Bench and BIRD

elvis

elvis (@omarsar0) highlighted a paper on self-evolving ontologies for agents. The post says data agents usually see tables, files, and databases through generic tools, while a hand-written semantic layer pasted into the prompt does not scale to many sources. EvoOntology instead builds an ontology with a dedicated agent and serves it as an MCP server with schema, content, and tool layers, which the data agent queries at runtime.

According to the post, the paper shows GPT-5.5 gains 26.7 points on DDR-Bench when the data agent can query an ontology of the data it works with. Across six backbones on DDR-Bench, accuracy rises 17.8 points on average, from 4.8 on Qwen3.5-Flash to 26.7 on GPT-5.5. On BIRD, execution accuracy rises 7.4 points. The ontology is edited in small typed steps, and each edit is kept only if a paired evaluation on the same backbone shows it helps; edits to the tool layer account for 57% of the gain from evolution.

Paper: https://arxiv.org/abs/2609.15779. Chat with Paper: https://academy.dair.ai/papers/evoontology-a-self-evolving-ontology-layer-for-data-agents-2609.15779.

#evoontology#data-agents#mcp#research#benchmarks

Best Taste-Bench model chooses the better direction 59.7% of the time

A paper from Microsoft and colleagues reports that the best model answers 59.7% of Taste-Bench questions correctly when choosing the better open direction in long engineering and research tasks. Forks decided by later evidence are much harder, a larger reasoning budget does not raise accuracy, and distilling a teacher’s outcome judgment improves held-out SWE-bench Pro success.