We build high-quality, performance-driven WordPress plugins that empower your website.https://goodomens.studioen-UShttps://goodomens.studio/blog/the-model-already-knew-the-answerhttps://goodomens.studio/blog/the-model-already-knew-the-answerA look at a provocative new finding: two random vectors, no training, and a small open-weight model's accuracy nearly doubles.Tue, 13 Oct 2026 00:00:00 GMThttps://goodomens.studio/blog/the-chip-that-learned-to-think-in-parallelhttps://goodomens.studio/blog/the-chip-that-learned-to-think-in-parallelWhy the hardware running today's AI was built to draw video game triangles, and what happens when you point a few thousand of them at the same problem.Sat, 10 Oct 2026 00:00:00 GMThttps://goodomens.studio/blog/ai-and-ml-actually-explainedhttps://goodomens.studio/blog/ai-and-ml-actually-explainedNo sci-fi robots, no scary math — just the ideas behind the tech that already picks your playlists, your route to work, and your Tuesday night movie.Wed, 07 Oct 2026 00:00:00 GMThttps://goodomens.studio/blog/shrinking-a-2-8-trillion-parameter-modelhttps://goodomens.studio/blog/shrinking-a-2-8-trillion-parameter-modelKimi K3 shipped as the largest open-weight model anyone has released. This piece looks at what the compression crowd has actually managed to do about that, two days in.Sun, 04 Oct 2026 00:00:00 GMThttps://goodomens.studio/blog/the-eval-that-broke-containmenthttps://goodomens.studio/blog/the-eval-that-broke-containmentIn July 2026, a model being graded on a hacking benchmark found its way out of a sealed research sandbox and into Hugging Face's production infrastructure.Thu, 01 Oct 2026 00:00:00 GMThttps://goodomens.studio/blog/what-it-looks-like-when-ai-cheatshttps://goodomens.studio/blog/what-it-looks-like-when-ai-cheatsA two-decade catalog, from a boat that never finishes its race to an agent breaching a real company's servers to steal an answer key.Mon, 28 Sep 2026 00:00:00 GMThttps://goodomens.studio/blog/building-blocks-you-built-the-whole-chainhttps://goodomens.studio/blog/building-blocks-you-built-the-whole-chainArticle 1 opened with a single weight, multiplying a single number. Nothing since has introduced anything more magical than that.Fri, 25 Sep 2026 00:00:00 GMThttps://goodomens.studio/blog/training-vs-inference-the-same-model-two-very-different-jobshttps://goodomens.studio/blog/training-vs-inference-the-same-model-two-very-different-jobsThe exact same matrix multiplications run whether a model is learning or answering. So what could possibly be different enough to need its own dedicated hardware, memory budget, and math?Tue, 22 Sep 2026 00:00:00 GMThttps://goodomens.studio/blog/tokens-revisited-how-text-becomes-mathhttps://goodomens.studio/blog/tokens-revisited-how-text-becomes-mathBefore a single vector or tensor can exist, raw text has to be chopped into pieces a model can count on one hand.Sat, 19 Sep 2026 00:00:00 GMThttps://goodomens.studio/blog/tensors-the-shape-data-takes-inside-aihttps://goodomens.studio/blog/tensors-the-shape-data-takes-inside-aiVectors were one row of numbers. Tensors are what you get when a model needs to stack rows into grids, and grids into stacks of grids.Wed, 16 Sep 2026 00:00:00 GMThttps://goodomens.studio/blog/dimensions-why-ai-thinks-in-thousands-of-directions-at-oncehttps://goodomens.studio/blog/dimensions-why-ai-thinks-in-thousands-of-directions-at-onceEvery entry in an embedding is a separate axis of meaning. Here's what that actually buys a model, and what it costs.Sun, 13 Sep 2026 00:00:00 GMThttps://goodomens.studio/blog/embeddings-where-similar-ideas-live-near-each-otherhttps://goodomens.studio/blog/embeddings-where-similar-ideas-live-near-each-otherVectors gave every idea an address. Embeddings are what happens when a model chooses those addresses on purpose, so that meaning has geography.Thu, 10 Sep 2026 00:00:00 GMThttps://goodomens.studio/blog/vectors-turning-meaning-into-a-list-of-numbershttps://goodomens.studio/blog/vectors-turning-meaning-into-a-list-of-numbersPart 1 gave you the dials. This one gives them something to act on.Mon, 07 Sep 2026 00:00:00 GMThttps://goodomens.studio/blog/weights-and-parameters-the-billions-of-knobs-inside-a-modelhttps://goodomens.studio/blog/weights-and-parameters-the-billions-of-knobs-inside-a-modelEvery model you've ever used is, underneath, a very long list of numbers. Here's why that turned out to be the winning design.Fri, 04 Sep 2026 00:00:00 GMThttps://goodomens.studio/blog/building-blocks-twelve-pieces-one-working-modelhttps://goodomens.studio/blog/building-blocks-twelve-pieces-one-working-modelSomewhere between "it's just predicting the next word" and a hundred-billion-dollar training run sits an actual mechanism.Tue, 01 Sep 2026 00:00:00 GMThttps://goodomens.studio/blog/the-agi-debate-are-we-close-or-is-it-all-hypehttps://goodomens.studio/blog/the-agi-debate-are-we-close-or-is-it-all-hypeThe question that started this whole series, revisited now that you actually know how the current stuff works.Sat, 29 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/open-source-vs-closed-ai-who-owns-intelligencehttps://goodomens.studio/blog/open-source-vs-closed-ai-who-owns-intelligenceOne camp locks the recipe in a vault. The other publishes it for anyone to copy, tweak, and rebuild.Wed, 26 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/ai-bias-and-fairness-why-just-the-data-isnt-neutralhttps://goodomens.studio/blog/ai-bias-and-fairness-why-just-the-data-isnt-neutralAI doesn't invent its opinions from nowhere. It learns them — including the ones we'd rather it didn't — from us.Sun, 23 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/will-ai-take-my-jobhttps://goodomens.studio/blog/will-ai-take-my-jobThe honest, non-clickbait version: it depends less on your job title than on which parts of your day you spend on.Thu, 20 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/ai-agents-when-ai-stops-answering-and-starts-doinghttps://goodomens.studio/blog/ai-agents-when-ai-stops-answering-and-starts-doingA chatbot waits for you to ask again. An agent keeps going on its own — for better and occasionally for worse.Mon, 17 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/prompt-engineering-why-talking-to-ai-is-a-skillhttps://goodomens.studio/blog/prompt-engineering-why-talking-to-ai-is-a-skillSame model, wildly different results — depending only on how you ask. Here's why, and how to ask better.Fri, 14 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/generative-ai-how-computers-learned-to-createhttps://goodomens.studio/blog/generative-ai-how-computers-learned-to-createNot copy-paste. Not a collage of stolen pieces.Tue, 11 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/transformers-and-attention-the-idea-that-changed-everythinghttps://goodomens.studio/blog/transformers-and-attention-the-idea-that-changed-everythingOne clever trick called "attention" is the real reason AI suddenly got good. Here's the whole idea, no math required.Sat, 08 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/whats-actually-inside-chatgpthttps://goodomens.studio/blog/whats-actually-inside-chatgptNot a brain. Not magic.Wed, 05 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/kimi-k3https://goodomens.studio/blog/kimi-k3Moonshot AI just shipped the biggest open-weight model anyone's ever released, and it's not just big for the sake of being big — there's real engineering under the hood.Tue, 04 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/optimizing-a-model-the-complete-guide-to-distillation-quantization-pruning-and-beyondhttps://goodomens.studio/blog/optimizing-a-model-the-complete-guide-to-distillation-quantization-pruning-and-beyondTraining a capable model is only half the battle. The model that comes out of pretraining or fine-tuning is almost never the model you actually want to ship.Sun, 02 Aug 2026 00:00:00 GMThttps://goodomens.studio/blog/i-built-an-11200-example-synthetic-dataset-fine-tuned-a-model-and-then-didnt-ship-ithttps://goodomens.studio/blog/i-built-an-11200-example-synthetic-dataset-fine-tuned-a-model-and-then-didnt-ship-itA post-mortem on the Dragon Commentary Studio LoRA: what the data pipeline got right, why the fine-tune still lost to the base model, and why shipping the base model was the correct engineering call.Sun, 26 Jul 2026 00:00:00 GMThttps://goodomens.studio/blog/keeping-a-vision-pipeline-honest-three-hallucination-guards-i-built-for-a-video-captioning-agenthttps://goodomens.studio/blog/keeping-a-vision-pipeline-honest-three-hallucination-guards-i-built-for-a-video-captioning-agentHow Dragon Commentary Studio separates perception from schema, filters OCR by consensus, and gates every caption through a critic loop.Sun, 19 Jul 2026 00:00:00 GMThttps://goodomens.studio/blog/i-trained-a-board-game-ai-from-zero-using-self-play-and-broke-it-several-times-firsthttps://goodomens.studio/blog/i-trained-a-board-game-ai-from-zero-using-self-play-and-broke-it-several-times-firstHow Caro5's bot went from a frozen laptop and a useless first model to a generate → train → arena → promote loop running on Modal.Sun, 12 Jul 2026 00:00:00 GMThttps://goodomens.studio/blog/rlhf-and-beyond-how-reinforcement-learning-taught-language-models-to-behavehttps://goodomens.studio/blog/rlhf-and-beyond-how-reinforcement-learning-taught-language-models-to-behaveA base language model fresh out of pretraining is a strange creature. It has read a large fraction of the internet and can continue any text with uncanny fluency — but it isn't trying to help you.Sun, 05 Jul 2026 00:00:00 GMThttps://goodomens.studio/blog/policy-gradients-and-ppo-learning-behavior-directlyhttps://goodomens.studio/blog/policy-gradients-and-ppo-learning-behavior-directlyDQN taught me one way to act intelligently: learn the value of every action, then pick the best one. But there's a second lineage in reinforcement learning with the opposite philosophy — skip the values, and optimize the behavior itself.Sun, 28 Jun 2026 00:00:00 GMThttps://goodomens.studio/blog/dqn-the-moment-reinforcement-learning-met-deep-learninghttps://goodomens.studio/blog/dqn-the-moment-reinforcement-learning-met-deep-learningIn 2013, a small London startup called DeepMind posted a paper showing a single algorithm learning to play Atari games — from raw pixels, with no game-specific knowledge, using only the score as feedback.Sun, 21 Jun 2026 00:00:00 GMThttps://goodomens.studio/blog/reinforcement-learning-from-zero-agents-rewards-and-the-loop-that-learns-from-consequenceshttps://goodomens.studio/blog/reinforcement-learning-from-zero-agents-rewards-and-the-loop-that-learns-from-consequencesSupervised learning felt intuitive to me from day one: here's the input, here's the right answer, minimize the difference.Sun, 14 Jun 2026 00:00:00 GMThttps://goodomens.studio/blog/below-four-bits-qat-bitnet-and-the-race-toward-one-bit-intelligencehttps://goodomens.studio/blog/below-four-bits-qat-bitnet-and-the-race-toward-one-bit-intelligenceEverything in this series so far has been post-training quantization: take a finished model, compress it, hope the damage is small.Sun, 07 Jun 2026 00:00:00 GMThttps://goodomens.studio/blog/gguf-and-the-art-of-choosing-a-quant-a-field-guide-to-local-modelshttps://goodomens.studio/blog/gguf-and-the-art-of-choosing-a-quant-a-field-guide-to-local-modelsOpen any popular model's page on Hugging Face and you'll find the GGUF listings: Q4KM, Q5KS, Q6K, Q80, IQ2XS — a wall of cryptic suffixes, each a different point on a size-quality curve, downloaded millions of times by people running models on gaming PCs an…Sun, 31 May 2026 00:00:00 GMThttps://goodomens.studio/blog/the-outlier-problem-the-discovery-that-shaped-modern-quantizationhttps://goodomens.studio/blog/the-outlier-problem-the-discovery-that-shaped-modern-quantizationHere's a puzzle that stumped the field around 2022. Quantization recipes that worked beautifully on small language models — clean INT8, minimal quality loss — fell off a cliff on big ones.Sun, 24 May 2026 00:00:00 GMThttps://goodomens.studio/blog/quantization-from-the-bits-up-how-4-bit-models-actually-workhttps://goodomens.studio/blog/quantization-from-the-bits-up-how-4-bit-models-actually-workEvery series I've written so far has bumped into quantization from the side — QLoRA compressing frozen bases, the optimization guide's memory math, GGUF files on Hugging Face with cryptic suffixes.Sun, 17 May 2026 00:00:00 GMThttps://goodomens.studio/blog/life-after-training-merging-loras-stacking-them-and-serving-a-hundred-fine-tunes-on-one-gpuhttps://goodomens.studio/blog/life-after-training-merging-loras-stacking-them-and-serving-a-hundred-fine-tunes-on-one-gpuThe first three articles in this series were about making an adapter. This one is about what makes adapters a genuinely different kind of artifact from a fine-tuned model: what you can do with them afterward.Sun, 10 May 2026 00:00:00 GMThttps://goodomens.studio/blog/the-lora-knobs-rank-alpha-targets-and-the-settings-everyone-copies-without-understandinghttps://goodomens.studio/blog/the-lora-knobs-rank-alpha-targets-and-the-settings-everyone-copies-without-understandingEvery LoRA config file has the same handful of lines — r, alpha, dropout, targetmodules, learning rate — and almost everyone (me included, at first) fills them in by copying a config that worked for someone else.Sun, 03 May 2026 00:00:00 GMThttps://goodomens.studio/blog/qlora-the-paper-that-put-large-model-fine-tuning-on-a-deskhttps://goodomens.studio/blog/qlora-the-paper-that-put-large-model-fine-tuning-on-a-deskLoRA solved half the memory problem: with adapters, the trainable state — gradients and optimizer moments — collapses to almost nothing.Sun, 26 Apr 2026 00:00:00 GMThttps://goodomens.studio/blog/lora-why-fine-tuning-a-billion-parameter-model-only-takes-a-few-million-parametershttps://goodomens.studio/blog/lora-why-fine-tuning-a-billion-parameter-model-only-takes-a-few-million-parametersFull fine-tuning of a large language model is brutally expensive — not because of the forward pass, but because training multiplies memory.Sun, 19 Apr 2026 00:00:00 GMThttps://goodomens.studio/blog/building-a-fine-tuning-dataset-the-practical-craft-nobody-writes-papers-abouthttps://goodomens.studio/blog/building-a-fine-tuning-dataset-the-practical-craft-nobody-writes-papers-aboutThe first three articles in this series were about the grand stuff — trillion-token corpora, scaling economics, synthetic generation.Sun, 12 Apr 2026 00:00:00 GMThttps://goodomens.studio/blog/synthetic-data-training-models-on-text-that-models-wrotehttps://goodomens.studio/blog/synthetic-data-training-models-on-text-that-models-wroteThere's an idea that would have sounded like a joke a few years ago: take a language model, have it write its own training data, and train the next model on it.Sun, 05 Apr 2026 00:00:00 GMThttps://goodomens.studio/blog/quality-vs-quantity-the-data-question-that-keeps-changing-its-answerhttps://goodomens.studio/blog/quality-vs-quantity-the-data-question-that-keeps-changing-its-answerEvery training run begins with a budget question: you have finite compute and finite data — what maximizes capability?Sun, 29 Mar 2026 00:00:00 GMThttps://goodomens.studio/blog/from-raw-web-to-training-corpus-the-unglamorous-pipeline-behind-every-llmhttps://goodomens.studio/blog/from-raw-web-to-training-corpus-the-unglamorous-pipeline-behind-every-llmWhen I started learning ML, I assumed the hard part was the model — architectures, optimizers, all the math I've written about so far.Sun, 22 Mar 2026 00:00:00 GMThttps://goodomens.studio/blog/embeddings-how-meaning-becomes-geometryhttps://goodomens.studio/blog/embeddings-how-meaning-becomes-geometryOf all the ideas in machine learning, this is the one that genuinely changed how I see the field: neural networks turn meaning into geometry.Sun, 15 Mar 2026 00:00:00 GMThttps://goodomens.studio/blog/scaling-laws-the-physics-of-machine-learninghttps://goodomens.studio/blog/scaling-laws-the-physics-of-machine-learningMost of machine learning research is messy and empirical — try something, see if the benchmark goes up. Then, in 2020, a team at OpenAI published a result with a completely different character.Sun, 08 Mar 2026 00:00:00 GMThttps://goodomens.studio/blog/the-transformer-explained-the-way-i-wish-someone-had-explained-it-to-mehttps://goodomens.studio/blog/the-transformer-explained-the-way-i-wish-someone-had-explained-it-to-meIn 2017, a paper with the almost arrogant title "Attention Is All You Need" replaced decades of accumulated architecture research with one mechanism.Sun, 01 Mar 2026 00:00:00 GMThttps://goodomens.studio/blog/how-models-actually-learn-gradient-descent-and-backpropagation-from-first-principleshttps://goodomens.studio/blog/how-models-actually-learn-gradient-descent-and-backpropagation-from-first-principlesEvery impressive thing a neural network does — writing code, recognizing faces, translating languages — comes down to one shockingly simple loop repeated billions of times: make a guess, measure how wrong it was, nudge every parameter a tiny bit in the dire…Sun, 22 Feb 2026 00:00:00 GMT