Live wire
The faceless channel that finally works, because it posts news Breaking-news triggers fire the second a story drops in your niche Auto-posts to TikTok, YouTube Shorts and Instagram Reels Animated charts, timelines and graphics. Pro-grade, never generic slop

AI

A startup claims it broke through a bottleneck that’s holding back LLMs

Featured from AI & Machine Learning Desk

Miami-based AI startup Subquadratic has shared the results of an independent evaluation by Appen to validate claims about its new LLM, SubQ.

A startup claims it broke through a bottleneck that’s holding back LLMs
Almira's Lens

Miami-based AI startup Subquadratic has shared the results of an independent evaluation by Appen to validate claims about its new LLM, SubQ. The model uses dynamic sparse attention instead of traditional dense attention to bypass the quadratic scaling bottleneck. Appen's tests found SubQ was 56 times faster than FlashAttention, scored 89.7% on LiveCodeBench, and achieved a 98% retrieval score on needle-in-a-haystack tests with context windows up to 12 million tokens. However, the model is not yet widely available, and Subquadratic bootstrapped SubQ by reusing weights from the Chinese open-source model Qwen.

By the numbers

12x
Text processing capacity compared to most other models
56x
Speed increase over FlashAttention in Appen's baseline test
89.7%
SubQ score on the LiveCodeBench coding test
$2,600
Cost to run Anthropic's Opus 4.6 through RULER 128
98%
SubQ needle-in-a-haystack retrieval score at massive scale

Source: mit_tech_review

Want a channel like this?

Pick your niche. We file the news, make the video, and post it. On autopilot.

Start your channel