AIERAAIERA is a weekly blog for people who want more than headlines about AI. Each piece explains one idea in machine learning with real numbers, small code you can run, and diagrams you can explore.https://aiera.com/Attention, step by step: how a transformer decides what to look athttps://aiera.com/blog/attention-step-by-step/https://aiera.com/blog/attention-step-by-step/Queries, keys and values sound abstract until you follow a few tokens through the math by hand.Thu, 22 Oct 2026 00:00:00 GMTLLMsIntermediateGradient descent, felt rather than memorizedhttps://aiera.com/blog/gradient-descent-felt/https://aiera.com/blog/gradient-descent-felt/Build intuition for slopes, step sizes and why the loss sometimes explodes, before touching any framework.Thu, 15 Oct 2026 00:00:00 GMTMath foundationsBeginnerWhy your validation loss went up while training loss kept fallinghttps://aiera.com/blog/validation-loss-rising/https://aiera.com/blog/validation-loss-rising/Overfitting, explained with one small dataset and three fixes you can try today.Thu, 08 Oct 2026 00:00:00 GMTTrainingBeginnerWhat the first layer of a CNN learns to seehttps://aiera.com/blog/cnn-first-layer/https://aiera.com/blog/cnn-first-layer/Visualize trained filters and you find edge and colour detectors, much like the early stages of the visual cortex.Thu, 01 Oct 2026 00:00:00 GMTComputer visionBeginnerTokenizers are where your model's quirks starthttps://aiera.com/blog/tokenizers-quirks/https://aiera.com/blog/tokenizers-quirks/Byte-pair encoding explains why language models struggle to count letters and do arithmetic on long numbers.Thu, 24 Sep 2026 00:00:00 GMTLLMsIntermediateReward is not the goal: notes on reward hackinghttps://aiera.com/blog/reward-hacking/https://aiera.com/blog/reward-hacking/When an agent maximizes the number you gave it rather than the behaviour you meant.Thu, 17 Sep 2026 00:00:00 GMTReinforcement learningAdvancedDiffusion models, from pure noise to an imagehttps://aiera.com/blog/diffusion-noise-to-image/https://aiera.com/blog/diffusion-noise-to-image/Train a network to remove a little noise, then run it many times in a row.Thu, 10 Sep 2026 00:00:00 GMTComputer visionIntermediateShipping a model is the easy part: watching for drifthttps://aiera.com/blog/monitoring-drift/https://aiera.com/blog/monitoring-drift/How input data quietly changes after launch, and the checks that catch it before your users do.Thu, 03 Sep 2026 00:00:00 GMTMLOpsIntermediateEvaluating LLM outputs without fooling yourselfhttps://aiera.com/blog/evaluating-llm-outputs/https://aiera.com/blog/evaluating-llm-outputs/Contaminated benchmarks, position bias in model graders, and how to build an eval set you can trust.Thu, 27 Aug 2026 00:00:00 GMTLLMsAdvanced