VQV Signal

OPEN SOURCE WATCH TECHNICAL

Atomic Chat posts uncensored Inkling 2 Mini builds, claiming refusals fell to 0% in its test

According to a post on X by @atomic_chat_hq, Atomic Chat released an "uncensored" Inkling 2 Mini as a LoRA adapter plus three GGUF files ranging from 8.67 to 16.53 GB. The post claims refusals fell from 100% to 0% in its test while the model kept 98% of its MMLU accuracy. A separate post on X by @shinboson shared 5.4 GB and 8.9 GB quantized builds, which the poster says keep 91.2% and 96.8% of original performance. That second post is mainly a video.

Source: Scobleizer News Pulse (X) · x.com Found via Robert Scoble’s Scobleizer News Pulse · a post on X, not independently verified Published 2026-11-21T17:45:00+00:00 Detected 2026-11-21T22:03:14+00:00
View original source

According to a post on X by @atomic_chat_hq, Atomic Chat released an "uncensored" Inkling 2 Mini as a LoRA adapter plus three GGUF files ranging from 8.67 to 16.53 GB. The post claims refusals fell from 100% to 0% in its test while the model kept 98% of its MMLU accuracy. A separate post on X by @shinboson shared 5.4 GB and 8.9 GB quantized builds, which the poster says keep 91.2% and 96.8% of original performance. That second post is mainly a video.

AI-assisted summary based on the listed source.

Atomic Chat released an "uncensored" Inkling 2 Mini as a LoRA adapter plus three GGUFs (8.67 to 16.53 GB); it claims refusals fell from 100% to 0% in its test while keeping 98% of MMLU accuracy. @shinboson posted 5.4 GB and 8.9 GB quantizations he says keep 91.2% and 96.8% of original performance (mainly a video)....

The releases show how quickly safety tuning on open-weight models can be removed, which matters for teams that rely on built-in guardrails. Developers and businesses evaluating open models may need to test refusal behavior themselves rather than assume it carries over. These are claims from social posts, so independent checks of refusal rates and benchmark results are the next thing to watch.

Signal Strength 88% Technical label WATCH Public Interest 38 Category OPEN SOURCE Reader Depth TECHNICAL

Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.

Public Interest components
Recognizable Entity Score 0 Practical Impact Score 36 Novelty Interest Score 94 Consequence Score 34 Curiosity Score 16 Shareability Score 51

VQV surfaced this signal because it is recent, relevant to LLM Inference, connected to Scobleizer News Pulse (X).