cd/sources/dejan-auto-discovered· home› sources› Dejan (auto-discovered)
cat /sources/dejan-auto-discovered.feed | wc -l → 34

Dejan (auto-discovered)

articles 34 domain dejan.ai → page 1/2 feed RSS
01:23
2026-11-02
dejan.ai
artificial-intelligence

Unlocking hidden skills inside Google's model

Training just 1,000 of Gemma 3 270M's 100 million weights (0.001%) raised Google's smallest Gemma model from 1% to 97% accuracy on sorting five numbers, according to a replication of a NeurIPS 2025 st…

01:06
2026-11-02
dejan.ai
ai-products

Releasing a new sentiment classification model.

DEJAN released a sentiment classification model on Hugging Face that sorts text chunks into seven classes from very positive to very negative, scoring 84% of labels exactly right and 95% within one po…

00:16
2026-10-29
dejan.ai
ai-safety

We Trained AI. Now AI Trains Us.

Professional analysts adopted the analytical choices favored by their AI coding agent 86% of the time in an Anthropic study of 50 causal-inference analysts, far above the roughly 50% expected if the a…

00:48
2026-10-20
dejan.ai
generative-engine-optimization

New Feature: Grounding Snippet Ablation

DEJAN added an Ablation tab to its Content Optimizer that measures how many rank places each sentence of an AI-cited source is worth, and in its first public example one sentence of the source Google …

04:22
2026-10-18
dejan.ai
ai-search

DeepSeek Search Engine

DeepSeek launched its own web search engine and two API endpoints on 17 October 2026, moving its search stack off third-party providers such as Bocha and onto an in-house crawler and index that DeepSe…

11:02
2026-10-14
dejan.ai
artificial-intelligence

We Tested an AI for Signs of Inner Experience. Nobody's Home.

Eight behavioural tests run on Google's gemini-3.8-flash through its public API, totalling about 8,900 trials, found no evidence of inner experience, though one result survived every control: the mode…

03:08
2026-10-06
dejan.ai
large-language-models

Link Building & Outreach for AI Search

A technical explainer on knowledge cutoffs and parametric memory states that Allen-Zhu and Li (2024) measured a language model's knowledge capacity at about 2 bits per parameter when each fact appears…

04:28
2026-10-04
dejan.ai
generative-engine-optimization

Why we don't do prompt tracking.

Dejan.ai's app.dejan.ai tracks canonical core entities rather than individual prompts to score brand visibility across Google, OpenAI and Anthropic models, running one fixed probe prompt per entity wi…

14:24
2026-09-24
dejan.ai
ai-search

Fanout Vectors

SEO practitioner Dejan published a reproduction of Google's query fanout research, training a vector-based diffusion model that replaces text fanouts with embeddings for document retrieval and releasi…

23:01
2026-09-16
dejan.ai
artificial-intelligence

Google's Approach to Recursive Self-Improvement

Google, Google DeepMind, the University of Maryland, and the University of Virginia published a 2026 technical report describing Dream-RSI, a method that optimizes AI agent search strategies offline b…

12:46
2026-09-16
dejan.ai
artificial-intelligence

Fanout V2: From Paraphrastic Collapse to Complementary Set

Google Research introduced R4T, an RL-trained diffusion retriever that optimizes fan-out query sets for diversity and coverage, achieving 12x to 20x latency speedups over autoregressive language model…

08:26
2026-09-01
dejan.ai
artificial-intelligence

How did the 1955 plan for AI go? Surprisingly well!

A comparison of the 1955 Dartmouth artificial intelligence proposal with the public record as of September 2026 shows that most of the seven proposed problems have seen substantial progress, though se…

05:35
2026-08-31
dejan.ai
artificial-intelligence

Your 'Gmail' can write code.

Google's Gemini AI in Gmail drafted a fully functional Python script to fine-tune Microsoft's DeBERTa-v3-large model for 3-label sentiment classification, including data loading, training loops, and e…

03:58
2026-08-30
dejan.ai
artificial-intelligence

Semantic Image Optimizer

An e-commerce image optimization method using Google's AI models and vector embeddings iteratively generated and scored product alt text, improving semantic similarity from 59% to 83% over ten attempt…

page 1 / 2 next →