← All editions
The Human in the LoopNo. 001 · February 17, 2026

Welcome to the first edition of The Human in the Loop, a weekly newsletter exploring what's new in AI/ML with a lens on energy industry connections. Each week, you'll get one deep story, quick hits, social trends, and sometimes a hands-on experiment. Whether you're a data scientist, geologist, drilling engineer, or just curious about AI, this is for you. Let's get into it.

Read time: ~5 min

≠ Radar

What’s new and interesting in AI/ML this week.

MiniMax M2.5: Near State-of-the-Art Open Models at 1/20th Claude Opus Cost

Chinese startup MiniMax released open-source M2.5 models claiming top performance in reasoning and coding at low cost, sparking buzz on X and LinkedIn about accessible high-quality AI. Users highlight its speed and benchmarks like 80.2% on SWE-Bench.

GLM-5: Best Open-Source Model for Reasoning, Coding, and Agents

The 754B-parameter GLM-5 topped open models on benchmarks for complex tasks and dropped with MIT license on Hugging Face. Community celebrates its agentic prowess for engineering and long-horizon planning.

Qwen3.5-397B-A17B Multimodal Model Hits Hugging Face

Alibaba's Qwen3.5-397B-A17B excels in image-text processing, language understanding, and agentic tasks across text, images, audio, and video. It's drawing interest for real-time data analysis without heavy setup.

ByteDance Seedance 2.0 Generates Cinematic Videos in Seconds

ByteDance's Seedance 2.0 creates Hollywood-quality video clips instantly from text prompts, trending for its realism and speed in content creation. It's part of the wave of low-cost Chinese AI tools.

The Signal

AnomaMind: Agentic AI for Pinpoint Time Series Anomaly Detection

AnomaMind is an AI agent that detects anomalies in time series data using tool-augmented reasoning to localize issues precisely. It combines large language models with specialized tools to analyze sensor streams and explain findings. Unlike black-box methods, it reasons step-by-step for reliable decisions in complex setups.

This enables early detection in high-stakes monitoring without massive labeled data. It supports root cause analysis by querying external tools like calculators or databases during inference. The result is sharper, explainable alerts for systems generating continuous data.

THE O&G ANGLE

In oil and gas, AnomaMind applies directly to rig sensor data, pipelines, and refineries for detecting equipment failures or leaks early. It enables predictive maintenance on bearings or pumps by localizing anomalies in vibration or pressure streams. Safer operations follow from catching instabilities before production halts.

Read more →
≠ Signals

Quick hits worth your attention this week.

1Pseudo-differential-enhanced PINNs Boost Physics Simulations

These PINNs extend gradient enhancement in Fourier space for more accurate PDE solving on tough problems. In O&G, they sharpen reservoir simulations, gas lift modeling, and flow predictions without heavy data needs.

2AltTS Framework Improves Multivariate Time Series Forecasting

AltTS separates stable autoregressive patterns from intermittent cross-variable interactions for better long-horizon predictions. It aids O&G production forecasting using variables like well pressure, choke size, and gas volume for resource optimization.

THE WELLBORE

Agents Are the Next Layer on Time Series Tools

This week's papers show agents layering reasoning on raw time series analysis, turning data into decisions. In energy ops, that shifts monitoring from alerts to automated fixes. Expect these hybrids to hit production sensors soon, cutting response times from hours to minutes.

Data ≠ Decisions. Context changes everything. DrillSense is the intelligence layer for drilling operations, built for the people who make the calls.

Know someone who should be reading this? Send them the archive, or subscribe from drillsense.com.