// Machine learning × production systems

Salty Syntax

Bridging AI research and production finance — the gap between what papers claim and what systems actually do.

Written by Priyanka Software Engineer in Financial Infrastructure  ·  ML systems, AI compliance, production architecture
// Featured All articles →

Your Temporal GNN Is Wasting 98% of Your GPU and the Research Community Doesn't Talk About It

TGAT runs at 5–6% GPU utilisation. JODIE at 1.5–2.5%. DyRep under 2%. A March 2026 paper finally fixes the inference problem — and the speedup is 739×.

If LLMs Know About Reflexivity, Do They Forecast Better?

A May 2026 paper feeds Soros's theory of self-reinforcing market dynamics to GPT-5, Claude, and Gemini. The forecasting improves. The reason why is harder to establish than the paper suggests.

// Recent All articles →

What Hedge Funds Actually Think About LLMs for Stock Forecasting (And Why Academia Gets It Wrong)

A 2026 IEEE-accepted paper finally says the quiet part out loud — data leakage, illiquidity blindness, and the evaluation metrics nobody in academia reports.

Notes on AI systems, financial infrastructure, and the gap between what research papers claim and what production deployments actually look like.

About →
// Reading log
147 Papers read
6 This month
5 Worth rereading
// Latest scribble All scribbles →
// PAPER Zhang & Zhang · Hedge Fund LLM Review arXiv:2605.05211

LLMs pretrained on internet-scale text corpora have likely been exposed to financial news, analyst reports, and market commentary from periods that overlap with academic test sets. This creates a form of data leakage that is distinct from look-ahead bias and considerably harder to detect. Reported directional accuracy figures may be substantially inflated as a result.

! The insidious thing is you can't fix this with a train/test split. The contamination happened during pretraining, before you ever touched the dataset — which nobody publishes.