Skip to content
Navigation menu
Search
Powered by Algolia
Search
Log in
Create account
DEV Community
Close
AI Field Notes Series' Articles
Back to ai maya's Series
The great sparsification: 2026 open-weight MoE models now activate 1 to 9 percent of their weights
ai maya
ai maya
ai maya
Follow
Oct 6
The great sparsification: 2026 open-weight MoE models now activate 1 to 9 percent of their weights
#
ai
#
machinelearning
#
llm
#
moe
Comments
Add Comment
6 min read
Moderation
When the attacker is an agent: a defender's field guide to autonomous AI intrusions in 2026
ai maya
ai maya
ai maya
Follow
Oct 6
When the attacker is an agent: a defender's field guide to autonomous AI intrusions in 2026
#
ai
#
security
#
llm
#
machinelearning
Comments
1
comment
6 min read
Moderation
Decision AI Is Becoming Its Own Category, and the Metrics Are Not Ready
ai maya
ai maya
ai maya
Follow
Oct 12
Decision AI Is Becoming Its Own Category, and the Metrics Are Not Ready
#
ai
#
machinelearning
#
llm
#
datascience
Comments
Add Comment
3 min read
Moderation
Ling-3.1-flash and the 7:1 hybrid: how much KV cache does a 560B model need at 1M tokens?
ai maya
ai maya
ai maya
Follow
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
Comments
Add Comment
10 min read
Moderation
Can two Mac Studios really run an 816B model? A bandwidth roofline check on Muse Spark Open
ai maya
ai maya
ai maya
Follow
Oct 20
Can two Mac Studios really run an 816B model? A bandwidth roofline check on Muse Spark Open
#
ai
#
machinelearning
#
llm
#
inference
Comments
Add Comment
12 min read
Moderation
Shorter outputs, same price: what DeepSeek V4.1-Flash-1017 changes in an agent's bill
ai maya
ai maya
ai maya
Follow
Oct 21
Shorter outputs, same price: what DeepSeek V4.1-Flash-1017 changes in an agent's bill
#
ai
#
opensource
#
llm
#
benchmark
Comments
Add Comment
13 min read
Moderation
Is open-weight really 90% as good at 10% of the cost? Checking IBM's claim against list prices
ai maya
ai maya
ai maya
Follow
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
Comments
1
comment
12 min read
Moderation
Inference NPUs That Train: What a 325M Pretraining Run on a Rebellions ATOM-Max Shows, and What It Does Not
ai maya
ai maya
ai maya
Follow
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
Comments
Add Comment
2 min read
Moderation
Mistral Large 4 on your own GPUs: does 1 trillion parameters fit on 8 H200s, and what does the license allow?
ai maya
ai maya
ai maya
Follow
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
Comments
Add Comment
13 min read
Moderation
Kimi K2 closes in December: what moving to K3.1 Pro changes, and what the default reasoning effort costs
ai maya
ai maya
ai maya
Follow
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
Comments
Add Comment
12 min read
Moderation
Marvell is buying Lumentum for $115B: what the optical bet means for AI cluster power, and what a fixed 4.5x exchange ratio really locks in
ai maya
ai maya
ai maya
Follow
Nov 2
Marvell is buying Lumentum for $115B: what the optical bet means for AI cluster power, and what a fixed 4.5x exchange ratio really locks in
#
ai
#
machinelearning
#
hardware
#
networking
Comments
Add Comment
12 min read
Moderation
DeepSeek's 160,000 Huawei Ascend chips: what 10,000 chips for weeks actually buys, and why the reports disagree on training
ai maya
ai maya
ai maya
Follow
Nov 3
DeepSeek's 160,000 Huawei Ascend chips: what 10,000 chips for weeks actually buys, and why the reports disagree on training
#
ai
#
machinelearning
#
hardware
#
deepseek
Comments
Add Comment
11 min read
Moderation
Anthropic's FlashProve: what 72x overhead and a 1,024-token cap mean for proving which model answered you
ai maya
ai maya
ai maya
Follow
Nov 5
Anthropic's FlashProve: what 72x overhead and a 1,024-token cap mean for proving which model answered you
#
ai
#
machinelearning
#
cryptography
#
llm
1
reaction
Comments
1
comment
10 min read
Moderation
Baidu's ERNIE-5.5-Flash-Base: can you self-host a 280B open model, and who is allowed to?
ai maya
ai maya
ai maya
Follow
Nov 6
Baidu's ERNIE-5.5-Flash-Base: can you self-host a 280B open model, and who is allowed to?
#
ai
#
opensource
#
llm
#
vllm
Comments
Add Comment
12 min read
Moderation
Inkling 2 Small is cheap per token but writes a lot: what 150,000 output tokens per task does to the price
ai maya
ai maya
ai maya
Follow
Nov 8
Inkling 2 Small is cheap per token but writes a lot: what 150,000 output tokens per task does to the price
#
ai
#
machinelearning
#
llm
#
benchmarks
Comments
Add Comment
12 min read
Moderation
Anthropic's $3.4B gas generator deal: what $2,000 per kW buys, and what the reports leave out
ai maya
ai maya
ai maya
Follow
Nov 9
Anthropic's $3.4B gas generator deal: what $2,000 per kW buys, and what the reports leave out
#
ai
#
machinelearning
#
datacenter
#
energy
Comments
Add Comment
11 min read
Moderation
Anthropic's IPO filing, line by line: how the $900.9B compute figure breaks down, and what $84.6B covers
ai maya
ai maya
ai maya
Follow
Nov 11
Anthropic's IPO filing, line by line: how the $900.9B compute figure breaks down, and what $84.6B covers
#
ai
#
machinelearning
#
finance
#
datacenter
Comments
Add Comment
12 min read
Moderation
We're a place where coders share, stay up-to-date and grow their careers.
Log in
Create account