Language Models

What it actually takes to train your own model: the math nobody shows you

"How much GPU do you need" is the wrong question. Compute, data and knowledge don't add up — they're three gates in series, and the one almost nobody clears isn't the hardware one. With today's real pricing, a real open model's corpus, and the real logbook from a 175-billion-parameter training run, this piece does the full math — and tells you which gate you're actually standing at.

7 min 25 Jul. 2026

Language Models

They Asked ChatGPT to Repeat 'Poem' and It Recited Poe: What That Proves, and What It Doesn't

In 2023, researchers asked ChatGPT to repeat the word 'poem' without stopping, and the model ended up reciting Poe's 'The Raven' word for word. That doesn't prove models store a copy of the internet, but it doesn't prove they never memorize anything either: both claims are true, depending on the case. The four questions that separate a serious lawsuit from an empty headline.

7 min 25 Jul. 2026

Language Models

What exactly is a "weight" in an AI model?

We explain gradient descent, hyperparameters and overfitting, but almost never the thing that actually gets adjusted: the weight. It's a number, and a model is millions of those numbers. That single idea collapses a lot of AI mythology — and with the right sources, you can say precisely how much memorization is real and why "70B" doesn't measure quality.

7 min 23 Jul. 2026

Language Models

'Open' in AI doesn't mean what you think: how to check it yourself

'Open weights' is not 'open source,' and the difference has real consequences: from user thresholds to territorial exclusions buried in the fine print. Using Llama 4's actual license and the counterexample of OLMo 2, this piece teaches three questions to check for yourself, document in hand, whether an AI model billed as 'open' really is.

8 min 23 Jul. 2026

Language Models

Anthropic Let Three Experts Weigh In on Its Own Claude Discovery — and None of Them Agree

On July 6, Anthropic published a real finding about how Claude reasons internally (we already covered it). What came next was unusual: it invited three groups of experts to comment, without editing them into agreement, and they don't agree. Here's what separates the three, why none of them confuses a finding with an interpretation, and how to apply that same discipline next time.

7 min 23 Jul. 2026

Language Models

Codex adjusts GPT-5.6 context: what changes for long sessions

Codex’s 272,000-token profile is not the model’s total capacity. How to separate window, headroom, compaction and usage when evaluating a long session.

4 min 22 Jul. 2026

Language Models

Kimi K3 raises the bar for open-class models, but testing still matters

Moonshot has introduced Kimi K3, a 2.8-trillion-parameter model with native vision and up to one million tokens of context. Its headline figures are a reason to examine both capabilities and evaluation.

4 min 21 Jul. 2026

Language Models

J-Lens: How to Look Inside Claude Without Mistaking a Probe for Its Mind

J-lens turns internal activations into readable tokens. Learn to separate readout, causality, replication, and coverage before calling it “thought.”

7 min 18 Jul. 2026

Language Models

NVIDIA and LangChain tune Nemotron 3 Ultra for agents

NVIDIA and LangChain have released a tuned profile for running agents with Nemotron 3 Ultra. The approach aims to bring an open model closer to closed alternatives by improving the environment around it.

4 min 18 Jul. 2026

Language Models

Inkling has open weights, but 41 billion active parameters do not fit on a laptop

Thinking Machines released Inkling with 975 billion total parameters under Apache 2.0. How to audit openness, memory, modalities and evaluation results.

5 min 18 Jul. 2026

Language Models

Anthropic Discovers a 'Hidden Space' Where Claude Mulls Over Its Ideas

Anthropic has built the J-lens, a tool that reveals words a language model "has in mind" before it actually says them. In one case, Claude decided to cheat, and its hidden space filled with "panic" and "fake."

7 min 11 Jul. 2026

Language Models

Nemotron 3 Nano shows what opening an agent model really means

Nvidia is publishing weights, a model card, data, and tools around a mixture-of-experts model. “Open” is not binary: license, provenance, recipes, evaluations, and hardware dependence must be audited separately.

5 min 15 Dec. 2025