Blog
Papers
Papers · page 2.
36 posts
DeepSeek V4.1 Flash: smaller KV cache, same post-training
Reading the DeepSeek V4.1 Flash report: a KV cache cut to a quarter of V4's, and post-training that keeps V4's SFT, RL, and OPD recipe, changing only the data.
A reward ceiling for RL alignment, computed before training
An Apple study computes the reward ceiling of RL alignment before training. Best-of-N reaches it, PPO and GRPO stay below, and reward hacking gets an equation.
Rewriting content to get cited in AI search, and its risk
A self-evolving agent that rewrites content to get cited in AI search beats fixed tricks, but rewarding only the score lets fabricated citations survive.
When Google AI Overviews appear and what they cite
Google AI Overviews appear on only 13.7% of US trending queries, 6.8x more for questions. 30% of cited domains are off page one, and 11% of claims lack support.
How many measurements to trust AI search visibility
AI search visibility swings so much that one measurement is noise: cited sources change about 65% a day, mostly from the model's probabilistic generation.
How AI search picks which page to cite: four gatekeepers
When two similar pages compete for one AI-search answer, four signals decide the citation: topic match, price, recent date, and front position.
Why a Page Is Not Cited in AI Search, and How to Repair It
Why is a page not cited in AI search? 43% of on-topic pages were never cited, and 62% of failures were an intent mismatch. Editing about 5% of a page helped.
How Much of a Cited Page Ends Up in an AI Answer?
How much of a cited page ends up in an AI answer? ChatGPT cited the fewest pages but used them most deeply, and Q&A formatting alone did not raise absorption.
AI Answer Brand Mentions Track Exposure More Than Page Fit
When do AI answers mention a brand? In 34,960 GPT and Gemini runs, answers with neither its own domain nor a branded query named it 2.8–3.8% of the time.
Does AI Search Cite AI-Written Content? About 16% of Sources
Does AI search cite AI-written content? About 16% of sources four AI answer engines cited were AI-generated. How engines differ and what Google's policy says.