HomeTagsVoice AI

Voice AI - Page 21

Stories from Voice AI come first, newest first, then stories from connected topics (ElevenLabs, AI Models, AI, Benchmarks, Hugging Face, open weights), each labelled with where it comes from. This view is not indexed by search engines.

Latest

Sakana AI’s MASS Lets a 27B Model Train Itself Using a Team of Its Own Subagents; the Results Are Self-Reported

Sakana AI's MASS lets one 27B model run, judge and learn from a team of its own subagents. Gains are Sakana's own, in a preprint not yet peer-reviewed.

Sperid Labs Releases Iris-3B, an Open Image Model With No VAE; Its Own Paper Finds No Clear Gain for Depth or Restoration

Sperid Labs released Iris-3B, a 3B image model with no VAE. Its paper finds no clear gain for depth or restoration. Code and weights are open.

Odyssey Opens Its Odyssey-3 World Model to the Public; Its Headline Physics Score Is a Best-of-Eight, and Every Benchmark Is Odyssey’s Own

Odyssey opened a public preview of Odyssey-3. Its 66.1 physics score is best-of-eight; the four-run average is 63.37. All figures are Odyssey's own.

Nvidia Reportedly in Talks to Acquire or Deepen Its Investment in Reflection AI

The Financial Times reports Nvidia is in talks to acquire or deepen its investment in open-weight start-up Reflection AI. Neither company has confirmed it. No deal value.

Polars 2.0 Is Out: Streaming Is the Default and Row Order Is No Longer Guaranteed

Polars 2.0, released 6 October, makes the streaming engine the default for lazy queries and no longer guarantees row order in joins and group-bys. Its benchmarks are its own.

Timer-M1 Preprint Says It Ranks First on Two Forecasting Benchmarks, With Its Authors’ Own Caveats

A preprint says Timer-M1, a forecasting model, ranks first on FEV and TIME. Its authors call the TIME margin over TimesFM-3 narrow. Not peer reviewed.

DuckDB 2.0 Ships Agent Mode That Cut AI Agents’ Token Reading 59%

DuckDB says the command-line tool in its unreleased v2.0 will print leaner results for AI coding agents. Its own test cut tokens read by 59%, but not total cost or runtime.

ReSI Preprint Says Automated Retraining Cut Attack Success to 31%

A preprint says its ReSI method cut mean X-Teaming attack success across four models from 86.01% to 31.45%. The authors' own tests; not peer reviewed.

Arena Raises $200M at $3.1bn Valuation to Rank AI Models

Arena, the public AI model leaderboard, raised $200m at a $3.1bn valuation and says it has passed $100m annualised revenue, up from $30m in January.

Underdog’s Saluki 27B Shrinks Qwen3.8-27B to 7.89 GB, at a Maths Cost

Underdog's Saluki 27B is a 7.89 GB, 2-bit version of Qwen3.8-27B. Underdog says it passes more tool-calling tasks than the full model on its own 120-task test but scores 15 to 18% lower on competition maths. What quantisation means, and what the numbers do not prove.

Latest on TSN

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.