NEWSFERENCE
TUE · NOV 17 · 2026 · 19:16 AWST
LIVE
$ today --liveF1TodayF2YesterdayF3ArchiveF4About
NEXT SCAN 19:30

SUN · NOV 01 · 2026

59 clusters from 71 articles.

TOP CLUSTER · TIER 1
DeepSeek open-sources dsh-superagent router-worker coding agent

DeepSeek released dsh-superagent, an MIT-licensed prototype that routes GitHub and Feishu requests into parallel coding-agent workers. In DeepSeek's simulation, p95 task completion time fell from 29.5 minutes to 15.7 minutes.

ACROSS THE WIRE · TIER 3
20 ROWS
#
ENTITY
HEADLINE
SRC
X
TIME
12
NVDA
Viral post claims half of Nvidia GPUs sit idle in warehouses
Drama
1
—
16d
13
OPENAI
Parameter Labs exposes ChatGPT artifact account takeover flaw
Drama
1
—
16d
14
NVDA
Pigeon Labs publishes V100 SXM2 pinout and NVLink hardware reference
1
—
17d
15
HERETIC
Heretic 2.0 expands open model modification beyond censorship removal
1
—
16d
16
GROOT
Groot releases open 19B to 280B mixture-of-experts models
1
—
17d
17
QWEN
Developer compiles an NPU into Qwen weights to run a nested LLM
1
—
17d
18
NVDA
NVIDIA releases Nemotron 3.5 Bolt E4B for on-device agents
1
—
17d
19
DATASSETT
Datasette LLM 0.1a9 adds default model discovery
1
—
16d
20
CROWDGPT
CrowdGPT v2.6 expands collaborative open LLM training project
1
—
17d
21
LLAMACPP
llama.cpp context checkpoints explain sliding-window attention memory use
1
—
17d
22
OPENAI
OpenAI reuses Daybreak Blue data despite opt-out settings
Policy
1
—
17d
23
XAI
Judge advances NAACP lawsuit over xAI data center pollution
Policy
1
—
17d
24
ADOBE
Adobe researchers build EditAgent for professional image editing
Research
1
—
16d
25
OPENAI
GPT-5.6 Sol beats Factorio on peaceful mode
Research
1
—
17d
26
FINEDNA
Hugging Face releases FineDNA dataset for DNA-specialized LLMs
Research
1
—
16d
27
CHINAAI
Study finds Chinese AI models display varied political alignments
Research
1
—
16d
28
QWEN
Qwen3.8-Flash-Next struggles with general research tasks
Research
1
—
17d
29
GOOGL
Google team presents StreamVibe video agent
Research
1
—
16d
30
QWEN
Qwen3.8 quantization tests reveal higher loop rates with INT4
Research
1
—
17d
31
MEMDOOR
Memdoor uses five Qwen3.8-27B models to beat Fable 5
Research
1
—
16d