Google Research releases Agile Deliberation, a framework for building subjective image classifiers
Google Research on November 4, 2026 presented Agile Deliberation, an interactive framework for building image classifiers around subjective categories. It comes with open-source code and a notebook demo. In an 18-session user study, the researchers reported stronger classifier results and lower effort than manual prompt refinement.
Google Research on November 4, 2026 presented Agile Deliberation, an interactive framework for building image classifiers around subjective categories. The release includes open-source code and a notebook demo.
What was announced
Agile Deliberation targets categories whose boundaries depend on personal judgment, cultural context, or operating policy. The examples given are healthy food and inspiring artwork. The coverage of the announcement was published by Unite.AI.
In an 18-session user study, the researchers reported stronger classifier results and lower effort than manual prompt refinement. The figures behind those results were not provided in the summary, so no scores or effect sizes are reported here.
Why subjective categories are hard
Classifiers for categories such as "healthy food" or "inspiring artwork" have no single ground truth. The definition depends on who is judging, in what cultural setting, or under which policy. According to the researchers, the framework is built for these cases, and the approach is interactive: the title frames it as bringing human deliberation into the classifier-building process.
Availability
- Open-source code: released
- Notebook demo: released
- Model weights, pricing, context window and benchmark scores: not applicable or not disclosed in the announcement summary
What this means
The work is a method for building classifiers, not a new trained model. Its central claim, per the researchers, is that structured interaction can outperform manual prompt refinement for categories where the definition is itself contested. The result comes from an 18-session user study, a small sample, so teams working on content policy or personalized classification should treat the reported gains as an initial result until the full methodology and numbers are reviewed. The open-source code and notebook demo allow practitioners to test the framework on their own subjective categories.
Related Articles
Microsoft Research's TokenEconomizer plugin cuts coding agent costs 18.4% to 30.9% on SWE-Bench
TokenEconomizer, an open-source plugin from UC Berkeley, Microsoft Research and Microsoft Security, learns from an agent's past sessions in a repository to cut token costs. Across 277 SWE-Bench tasks and five agent and model combinations, mean cost per task fell 18.4% to 30.9%, according to the authors.
Amazon releases Mitra-v2, an open tabular foundation model that leads the single-model field on TabArena
Amazon has released Mitra-v2, a tabular foundation model for classification and regression that, according to Amazon, leads the single-model field on the TabArena benchmark of 51 datasets. Weights and code are available under Apache-2.0 on Hugging Face.
NVIDIA: Sparse attention indexer and Top-k take 47% of a DeepSeek V3.2 prefill layer at 128K tokens
NVIDIA published a co-design primer for model-level sparse attention. It reports that the indexer and Top-k take 47% of a DeepSeek V3.2 prefill layer at 128K tokens, and that reusing the prior decode step's selection speeds up Top-k by 2.1x to 2.4x.
Together AI says CUDA Graphs and PDL recover 98% of megakernel performance without custom kernels
Together AI reports that CUDA Graphs and PDL on top of default Torch Inductor kernels recover 98% of its earlier megakernel performance. With "multi-kernels" and agent loops, it claims up to 1.7x over its old megakernels, reaching over 1100 TPS for Llama 1B on H100. The work led to Together Compiler, a framework for continual AI inference optimization.
Comments
Loading...