Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

ModernCE

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

1

Findings

1

Edges

2

Sources

1

Corpus findings

2026-09-20 / reddit-researcherSomeone benchmarked Jev, Laya, ModernCE and Qwen3.5-4B as real-time Doom controllers and published the latency tableUsing ViZDoom at 320x240 with a 35 Hz clock and a five-decisions-per-second target, a builder fed each model a deterministic text description of visible objects and HUD values and read back one button action, averaging over eight seeds per scenario on a DGX Spark with GB10 and 128 GB unified memory. Jev 1.13 led on mean kills at 5.63 with 117.3ms p50, finetuned Qwen3.5-4B LoRA got 3.63 kills at 146.8ms, and Laya English and finetuned ModernCE-base-nli tied at 1.25 kills but with 16.2ms and 7.6ms p50 respectively. The spread is the useful part: the fastest model was 15x quicker than the most accurate one and scored a quarter as many kills.

Graph relationships

BENCHMARKED AGAINST
Jev
Jev scores 5.63 mean kills versus ModernCE-base-nli's 1.25 kills.
Source finding
BENCHMARKED AGAINST
Jev -> ModernCE
Jev scores 5.63 mean kills versus ModernCE-base-nli's 1.25 kills.
Source finding

Source trail

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues