VQV Signal

RESEARCH WATCH TECHNICAL

DNeo claims up to 30% faster LLM inference; preprint finds simple model averaging beats replay

According to a post on X by @Scobleizer, DNeo, a new diffusion-based drafter, claims up to 30% faster LLM inference. The same post points to a preprint co-authored by Jonathan Frankle, which finds that simple model averaging (SFA) beats more complex methods such as replay for continual learning. The snippet gives no further detail on the methods or test setup, so both points are reported as stated in the post and its linked sources.

Source: Scobleizer News Pulse (X) · x.com Found via Robert Scoble’s Scobleizer News Pulse · a post on X, not independently verified Published 2026-11-19T17:30:00+00:00 Detected 2026-11-20T01:05:34+00:00
View original source

According to a post on X by @Scobleizer, DNeo, a new diffusion-based drafter, claims up to 30% faster LLM inference. The same post points to a preprint co-authored by Jonathan Frankle, which finds that simple model averaging (SFA) beats more complex methods such as replay for continual learning. The snippet gives no further detail on the methods or test setup, so both points are reported as stated in the post and its linked sources.

AI-assisted summary based on the listed source.

DNeo, a new diffusion-based drafter, claims up to 30% faster LLM inference. Separately, a preprint co-authored by Jonathan Frankle finds that simple model averaging (SFA) beats more complex methods such as replay for continual learning. Why it matters: Cheaper inference and easier model updating both lower the...

Teams that run LLMs in production would be affected if DNeo's speedup holds up, since faster inference can lower serving costs. Developers who update models over time could also benefit if a simpler method outperforms replay for continual learning. Watch for the full preprint and for independent testing of both claims.

Signal Strength 84% Technical label WATCH Public Interest 25 Category RESEARCH 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 0 Novelty Interest Score 94 Consequence Score 10 Curiosity Score 0 Shareability Score 45

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