Kronos vs Rival.tips

Kronos is the first open-source foundation model purpose-built for financial candlestick (K-line / OHLCV) data, accepted at AAAI 2026. Instead of treating price series like text or images, Kronos uses a custom two-stage architecture: a specialized tokenizer that converts continuous OHLCV data into discrete tokens, followed by an autoregressive Transformer trained on data from 45+ global exchanges. Four model sizes range from 4.1M to 499M parameters, all released under MIT license. The model learns the statistical structure of market microstructure directly from raw candlestick sequences, enabling zero-shot and few-shot forecasting across asset classes — equities, crypto, and commodities. It ships with a live BTC/USDT prediction demo, Qlib integration for A-Share markets, and a backtesting framework so researchers can evaluate strategies end-to-end. With 13.6k GitHub stars in a niche domain, the community reception has been unusually strong. Kronos matters because most "AI for trading" projects glue LLMs to news sentiment or financial reports — pattern-matching on text rather than market structure. Kronos is the rare project that treats price action itself as the primary modality, giving quants and ML researchers a base model they can fine-tune on proprietary data rather than starting from scratch on every new dataset.

Rival.tips is a research tool and interactive visualization that fingerprints the stylistic DNA of 178 AI language models — measuring vocabulary patterns, sentence structure preferences, hedging language frequency, formality registers, and punctuation habits — then clusters them into a navigable map showing which models write like which. The result is a kind of "accent atlas" for AI: you can see at a glance that GPT-4o and Claude Sonnet cluster together on formality but diverge sharply on hedging language, while Llama-3 and Mistral write more similarly to each other than either does to any OpenAI or Anthropic model. The tool works by running a standardized suite of 40 prompts across all 178 models, extracting 120 stylometric features per response, and reducing the high-dimensional space to an interactive 2D UMAP projection. The Show HN post hit 68 points with discussion focusing on the methodological choices and surprising cluster assignments — several models that market themselves as distinct turned out to be nearly indistinguishable stylistically. Practical applications include AI content detection research, model selection for brand voice matching, and detecting when a provider has silently updated their model (stylometric drift is often detectable before the provider announces it). The methodology and raw data are fully open.