Suno users say v6 is a downgrade from v4.5

Nature Medicine study validated the platform in 13 patients (11 scalp, 2 intracranial EEG); the AI reports sleep, medication, cognition and seizure-probability biomarkers via smartphone app. In a pseudoprospective simulation, seizure detectors improved precision from patient feedback alone, without expert intervention.

ByteDance researchers posted a paper (arXiv 2610.12102) on a new few-step distillation method for diffusion generation, shared to r/StableDiffusion. The Reddit poster notes it is "another" few-step distillation method from the company.

Cerebras breaks down the two sides of AI performance: latency is response time from prompt to result, including Time to First Token and inter-token latency.

A Reddit user asked Claude to spot red flags in publicly available 2024 CMS data; most flagged entities turned out to be companies already known to authorities.

Anthropic has added limit resets to Claude, letting users reclaim usage after hitting caps; a Reddit user reported using their free reset for the first time. Anthropic's support docs describe the feature as a 'limit reset.'

A LocalLLaMA user ran benchmark tests on multiple models in Opencode after being impressed by Exo's results, reporting that many models performed poorly, especially mid-level ones.

Paper (arXiv 2609.40305) runs a shared group of transformer blocks several times inside each denoising step instead of scaling up a text-to-image DiT. Naive looping didn't reliably help; the authors fix it with deep supervision on every loop, and loop depth can be changed at sampling time.