Topic

AI Research

Reporting on the research behind the products: new training methods, benchmarks, retrieval and reasoning techniques, and the papers that move the field.

We summarise what a result claims, how it was measured and where its limits are, in language that does not need a machine learning background.

Showing 13–24 of 54 articles

RSS feed

Page 2 of 5

Machine Learning
MacBook as AI training cluster: why this idea matters
MacBook as AI training cluster sounds odd until you see how DiffusionBlocks changes on-device AI training and self-improving agents.
May 28, 20268 min read
Training Approaches
Train GPT on non language data: a practical field guide
Train GPT on non language data with fewer mistakes. Learn tokenization, scaling, vocabulary imbalance, evaluation, and model choices from 100M to 500M.
May 28, 202610 min read
AI Benchmarks
ChatGPT 5.5 vs Opus DeepSWE benchmark: what changed
ChatGPT 5.5 vs Opus DeepSWE benchmark results look strong, but here's what DeepSWE measures, misses, and means for coding teams.
May 28, 20267 min read
Machine Learning
Tencent Hy MT2 1.8B translation model: why it matters
Tencent Hy MT2 1.8B translation model may be the best small AI translation model for cost, speed, and reliable production deployment.
May 26, 20268 min read
AI Benchmarks
Redundancy in LLM reasoning: how much thinking is enough?
New research on redundancy in LLM reasoning asks how much thinking is enough, and how to cut latency, energy, and compute waste.
May 26, 20266 min read
AI Benchmarks
Agentic AI Energy Benchmark: Why Goal-Level Metrics Matter
Understand the agentic AI energy benchmark shift to goal-level accounting and learn how to measure energy per successful task in production.
May 25, 20268 min read
NLP Research
ImProver 2 neurosymbolic proof optimization explained
Learn what ImProver 2 neurosymbolic proof optimization adds to formal proof refactoring, self-improving language models, and AI theorem proving.
May 25, 20266 min read
NLP Research
MindLoom reasoning data synthesis explained
MindLoom reasoning data synthesis aims to build better frontier-level reasoning sets by composing thought modes for LLM training.
May 23, 20266 min read
Machine Learning
AI is not one thing: how AI actually works
AI is not one thing. Learn the difference between AI, machine learning, deep learning, search, and LLMs with a practical decision model.
May 22, 20268 min read
AI Benchmarks
OSCToM Theory of Mind Benchmark: What It Really Tests
OSCToM theory of mind benchmark explained: how RL-guided adversarial generation tests high-order social reasoning in LLMs and agents.
May 22, 20261 min read
Machine Learning
Common Uses of Python Programming Language 2026
Explore the common uses of Python programming language 2026, from AI and automation to cloud, data science, and business software.
May 21, 20267 min read
AI Benchmarks
Theory of mind in human AI interaction: what the data says
Theory of mind in human AI interaction looks promising, but new evidence points to where benchmark gains help users and where they don't.
May 18, 20268 min read

MacBook as AI training cluster: why this idea matters

Train GPT on non language data: a practical field guide

ChatGPT 5.5 vs Opus DeepSWE benchmark: what changed

Tencent Hy MT2 1.8B translation model: why it matters

Redundancy in LLM reasoning: how much thinking is enough?

Agentic AI Energy Benchmark: Why Goal-Level Metrics Matter

ImProver 2 neurosymbolic proof optimization explained

MindLoom reasoning data synthesis explained

AI is not one thing: how AI actually works

OSCToM Theory of Mind Benchmark: What It Really Tests

Common Uses of Python Programming Language 2026

Theory of mind in human AI interaction: what the data says