Skip to content

DEV Community

AI Field Notes Series' Articles

Back to ai maya's Series

Cover image for The great sparsification: 2026 open-weight MoE models now activate 1 to 9 percent of their weights

ai maya

Oct 6

The great sparsification: 2026 open-weight MoE models now activate 1 to 9 percent of their weights

#ai #machinelearning #llm #moe

6 min read

ai maya

Oct 6

When the attacker is an agent: a defender's field guide to autonomous AI intrusions in 2026

#ai #security #llm #machinelearning

6 min read

ai maya

Oct 12

Decision AI Is Becoming Its Own Category, and the Metrics Are Not Ready

#ai #machinelearning #llm #datascience

3 min read

Cover image for Ling-3.1-flash and the 7:1 hybrid: how much KV cache does a 560B model need at 1M tokens?

ai maya

Oct 19

Ling-3.1-flash and the 7:1 hybrid: how much KV cache does a 560B model need at 1M tokens?

#ai #machinelearning #llm #inference

10 min read

Cover image for Can two Mac Studios really run an 816B model? A bandwidth roofline check on Muse Spark Open

ai maya

Oct 20

Can two Mac Studios really run an 816B model? A bandwidth roofline check on Muse Spark Open

#ai #machinelearning #llm #inference

12 min read

Cover image for Shorter outputs, same price: what DeepSeek V4.1-Flash-1017 changes in an agent's bill

ai maya

Oct 21

Shorter outputs, same price: what DeepSeek V4.1-Flash-1017 changes in an agent's bill

#ai #opensource #llm #benchmark

13 min read

Cover image for Is open-weight really 90% as good at 10% of the cost? Checking IBM's claim against list prices

ai maya

Oct 23

Is open-weight really 90% as good at 10% of the cost? Checking IBM's claim against list prices

#ai #machinelearning #llm #opensource

12 min read

Cover image for Inference NPUs That Train: What a 325M Pretraining Run on a Rebellions ATOM-Max Shows, and What It Does Not

ai maya

Oct 26

Inference NPUs That Train: What a 325M Pretraining Run on a Rebellions ATOM-Max Shows, and What It Does Not

#news #ai #machinelearning #hardware

2 min read

Cover image for Mistral Large 4 on your own GPUs: does 1 trillion parameters fit on 8 H200s, and what does the license allow?

ai maya

Oct 31

Mistral Large 4 on your own GPUs: does 1 trillion parameters fit on 8 H200s, and what does the license allow?

#ai #opensource #llm #inference

13 min read

Cover image for Kimi K2 closes in December: what moving to K3.1 Pro changes, and what the default reasoning effort costs

ai maya

Nov 1

Kimi K2 closes in December: what moving to K3.1 Pro changes, and what the default reasoning effort costs

#ai #programming #llm #api

12 min read