Dev builds maap, a 4B model returning calibrated number distributions in one forward pass
fruiapps · reddit · 2026-10-20
An indie developer released maap, a 4B estimation model that reads a record as text and outputs a full distribution (median, quantiles, exceedance probability) for a numeric question in a single forward pass — targeting questions like repair costs or home prices where frontier LLMs answer confidently whether or not they know. On FermiEval, frontier models' nominal 99% intervals covered the truth only 65% of the time. The author openly lists weaknesses: 44 of 104 targets fail 80% interval coverage, tuned XGBoost/TabPFN beats it when labels exist, single training seed, don't use the mean on skewed targets. Demo on 5 held-out real records (house, used car, Medicare DRG, diamond, avocados) had all true values inside the 80% band, from $1.06 to $215,000. Weights not open yet; hosted API free for first 1,000 estimates.
More from Models
- OpenAI's Decisions API hits GA: near real-time routing up to 10x faster than GPT-6 Luna — OpenAIDevs · 2026-10-23
- DeepLearning.AI weekly: GPT-6.1 hits 97.9% byte parity on COBOL migration — DeepLearningAI · 2026-10-23
- Marker's lite mode scores 75.1 vs 42.3 for Azure Read on OmniParseBench — VikParuchuri · 2026-10-23
- Datalab launches lite OCR at $0.50 per 1,000 pages, 20-30x cheaper than Azure, Google and AWS layout APIs — VikParuchuri · 2026-10-23
- LiveNerf's 29-day data finds no nerfing of Claude Opus 5.5 — TheOnlyVibemaster · 2026-10-23
- PHP Foundation reveals Anthropic's Claude cyber model scanned PHP, validated with OpenAI's Daybreak Red — saibharadwaj · 2026-10-23