NHL Turns to Romance Audiobooks as AI Video Makers Chase the Same Attention

The National Hockey League started publishing short romance-styled audio stories about its own players this week. Some of the writers it approached have publicly refused to take part. WIRED reported the first batch, part of a campaign called "Hot for Hockey," on Friday and said the league contacted popular hockey romance authors to write and narrate 60 to 90 second pieces for audiobook platforms.

The stories are not romances. They read as back-cover blurbs designed to sell tickets.

WIRED quoted the complete text of one, titled The Summer I Turned Gritty with Trevor Zegras: "Trevor Zegras came to Philly looking for a fresh start. But after battling with Pittsburgh in the playoffs last year, he found out how gritty things could get. One heated summer later, Philly's pretty boy is looking a whole lot grittier." The first batch also features the Carolina Hurricanes' Sebastian Aho and the Vegas Golden Knights' Jack Eichel, according to WIRED, and was written by Pucking Wrong author C.R. Jane.

Not everyone in the genre is amused. Emily Rath, a best-selling hockey romance novelist, said on Threads that the league had contacted her and that she had "never turned down an offer so fast," WIRED reported. Rath said the campaign's slide deck contained jokes she considered mockery of the genre and what she alleged were AI cover renders. Chelsea Curto, author of the D.C. Stars series, also flagged the AI graphics, writing that it was probably for the best she declined because her books "can be very critical of the league."

An NHL spokesperson, Jon Weinstein, told WIRED the campaign is a "playful nod to something that's clearly going on in the cultural zeitgeist." He said the league was "very proud to partner with real hockey authors, illustrators and voice actors in the space."

The visual side keeps moving

The NHL story is a useful reminder that AI-generated or AI-assisted content is now a routine part of how large organisations market themselves, and that audiences notice. On the model side, the past week produced a steady stream of releases aimed at video and image work.

Ideogram said on 1 October that its new Ideogram 4.5 model can edit part of an image without altering the rest, a problem the company says leading models such as GPT-Image 2.5 and Nano Banana still show artifacts on after repeated edits. The Decoder reported the model ships in four quality tiers priced from 0.8 to 22 cents per image at native 2K resolution, is available now on Ideogram's platform and API, and has partners including Picsart, Runway, Pika and Leonardo AI. An open-weight release is promised soon.

Tavus, meanwhile, introduced Griffin, which it calls the first "Human Interaction Model," a system that processes speech, facial expressions, tone, gestures and pauses while both receiving and generating video. According to The Decoder, 48 percent of participants in a Tavus study believed Griffin was a real person after a one-minute video call, against a previous high of two percent. In what Tavus describes as an independent Nvidia test measuring how human an AI feels in direct audio-video conversation, Griffin scored 3.83 points against 3.92 for actual humans and 2.80 for the previous best AI model.

A preview called Griffin-Lite is open to "select testers as a research preview," with a fuller version promised once "safety concerns are addressed," The Decoder reported. Tavus lists tutoring, practising difficult conversations and camera-based tech support as possible uses.

Voice and decision models arrive in the same window

Microsoft AI pushed out new speech models on 2 October. According to The Decoder, MAI-Transcribe-2-Streaming transcribes 60 languages and returns its first partial results in just over 100 milliseconds, which Microsoft says lets voice agents respond while a person is still mid-sentence. An hour of audio costs $0.54 at the introductory price through the end of the year. The two new text-to-speech models, MAI-Voice-2.1 and MAI-Voice-2.1-Flash, cover 23 languages in the same voice, with the Flash variant at 150 milliseconds of latency and $15 per million characters instead of $22. Both can clone a voice from a few seconds of reference audio. In one test cited by The Decoder, about half of 4,000 participants thought the voices belonged to a real person.

Decision models are having their own moment. Amazon Web Services released Strands Decider 2B, an open source model that sorts between pre-decided options and returns a confidence measure, TechCrunch reported on 1 October. It was built by distinguished engineer Marc Brooker after he saw TypeSafe's Jev, and briefly topped the Jevbench ranking for models of its size. "What originally piqued my interest in this class of models was that they make a perfect decider for a workflow step," Brooker told TechCrunch.

TypeSafe's own Jev, released on 1 October according to InfoQ, returns typed probabilistic decisions rather than text, at $0.042 per million input tokens with output free, a 32,000-token context window and quoted end-to-end latency of 70ms to 500ms. Vercel engineer Pranit Sharma found a safety classifier ran five to 18 times faster than the LLM it replaced. An analysis of 12,759 launch tweets by OpenChamber put user-reported speedups at a median of 7x against a 193.6x headline, cost savings at a median of 30x and latency at a median of 76ms.

Cloudflare answered on 1 October with Clef and Clef-flash, open-weight decision models that can also handle images and video and support a 64k context window, The Register reported. Cloudflare says Clef costs $0.24 per million tokens, nearly six times Jev's price. The company self-reported its benchmark scores and they have not yet been reproduced on the official Decision Index, The Register noted.

Google's Gemini 4 Argon, announced 30 September, is being restricted to a vetted group of cybersecurity partners under the company's Fairwind Program. "Safely releasing frontier capabilities at this level requires a phased approach," wrote Koray Kavukcuoglu, Google's chief AI architect, in a blogpost cited by The Guardian. TechCrunch reported that Google claims Argon beat OpenAI's GPT-6 Astra and Anthropic's Fable and Opus models on several benchmarks, citing the startup Vals.

On the copyright front, Anthropic has urged the Australian government to consider "conditional approval" for training on Australian copyrighted works under an opt-out model, according to a submission reported by The Guardian. The ABC and SBS pushed back, with the ABC warning of "cannibalisation of the Australian news industry" and calling for AI companies to face the same copyright, defamation and privacy rules as media outlets.

OpenAI said it disrupted an "adversarial distillation" campaign that began in early July and peaked with 16,000 requests from more than 4,000 users over two days, attributing a core cluster to individuals associated with China's Moonshot AI, CNBC reported. The Register noted OpenAI did not say which of its models were targeted and that Moonshot did not respond to a request for comment.