Skip to content

AI News This Week: Google Ships a Fast, Cheap AI Judge for Video (Nov 13)

Chris
Chris

AI News This Week · Nov 13

On Wednesday, Google released the Gemini Decisions API, a model that answers yes-or-no, multiple-choice and scoring questions about text, images, audio and video in a fraction of a second, for ten cents per million input tokens. Fast, cheap judgment calls on video are the building block behind every “is this a highlight?” decision, and this week also brought a firmer line on AI athlete lookalikes, AI-made highlights at the Youth Olympics and more proof that live vertical works.

The week in numbers

$0.10

per 1M input tokens for Google's new Decisions API (output is free)

1,000

questions Google says one Decisions request can carry

16 / 25

Youth Olympic sports getting AI-generated highlights from OBS

~507

hours of host coverage OBS planned for Dakar 2026

01Google ships a fast AI judge

Google launched the Gemini Decisions API on November 11. Instead of writing an answer, the model scores a fixed set of options and returns a probability and a confidence score from 0 to 1. It takes text, images, audio, video and PDFs, up to 256K tokens of input, and supports three question types: yes/no, multiple choice (2 to 255 options) and a graded rubric score. One request can carry up to 1,000 separate questions.

Google prices it at $0.10 per million input tokens with no charge for output, and says shared input is billed once no matter how many questions you ask. It reports end-to-end latency of 150 to 250 milliseconds and calls it nearly five times cheaper than Gemini 3.5 Flash-Lite on its own benchmark (all self-reported). It is live in the Gemini API and Google AI Studio, with Google Cloud support “coming soon.”

What it is not

■  A video generator

■  A chatbot writing long answers

■  Proven on sports footage (Google's examples are routing, safety and design)

What it is built for

■  Fast closed-set calls: yes/no, pick one, score it

■  A confidence number to send unsure cases to a person

■  Many questions over the same input, billed once

Why it matters for live sports video

A lot of clipping work is a chain of small decisions: is this a scoring play, which team, is it worth posting, is anyone in frame who opted out. A model that answers those as calibrated yes/no questions, with a confidence number attached, fits that chain well. Treat the speed and price as Google's claims and test on your own footage, but the direction is clear: the judgment layer of video AI is getting faster and cheaper, and confidence scores make human review easier to target.

02AI labels don't cover lookalikes

Google clarified on November 11 that ads using AI to impersonate public figures break its Unacceptable Business Practices policy even when they disclose that AI was used. Google's policy names athletes, entertainers, influencers and officials as public figures, and says implying an endorsement needs documented consent. Violations count as egregious: accounts are suspended on detection, without warning. Google says the note explains the policy rather than changing enforcement.

March 2024

Google starts enforcing its public figure impersonation ban.

July 9, 2026

Google adds a “How this ad was made” panel showing whether an ad used AI.

Aug 2, 2026

EU AI Act transparency obligations for AI content begin.

Nov 11, 2026

Google: an AI label does not make impersonation acceptable.

Source: PPC Land, Nov 12, and Google Ads policy help.

Why it matters for live sports video

This lands a week after LeBron James went after a betting app over a lookalike promo. The message from the biggest ad platform is that labeling is not a license. If your clips or promos feature athletes, especially around sponsors and betting, you need to know who is on screen and what rights cover the use. Accurate athlete tags and consent records are what let you answer that quickly.

Quick check for athlete clips in ads

Know which athletes appear in every clip you promote

Keep written consent for anything that implies an endorsement

Disclose AI edits, and don't treat the label as permission

03Youth Olympics run on AI highlights

The Dakar 2026 Youth Olympic Games close today, November 13. For these Games, Olympic Broadcasting Services said it would use AI to generate highlights for 16 of the 25 sports so key moments reach broadcast, digital and social platforms quickly. It also uses AI-assisted audio mixing under human supervision, an Automatic Media Description tool built with Alibaba that writes descriptions, keywords, transcripts and summaries for search, and a trial of AI-run continuous channels. Smartphones were set to shoot vertical content at skateboarding and taekwondo.

Dakar 2026 sports coverage

Competition sports

 
25

Covered live

 
22

AI-generated highlights

 
16

Source: OBS via Olympics.com, Oct 30.

OBS describes the AI work as still developing. Its senior audio manager told SVG Europe there is “still work to be done,” and SVG Europe framed Dakar as a test of whether AI workflows can help Olympic broadcasting.

Why it matters for live sports video

The world's biggest multi-sport producer is using AI for highlights and searchable descriptions at an event where most sports never get full broadcast attention. That is the same problem every governing body and event operator has: more sports and athletes than editors. Highlights plus descriptions you can search are becoming the standard, not an extra.

04Live vertical beats the crop

Gravity Media built a live 9:16 feed for Meta Connect 2026, reported on November 10. It turned the 16:9 program into a 1080x1920 portrait output with its own vertical graphics, running in real time next to the main show. Meta asked for something more than a reformatted broadcast. Gravity says the vertical feed beat landscape coverage on live views, engagement and impressions (self-reported, with no figures given).

Why it matters for live sports video

Meta Connect is a tech event, not a game, but the lesson carries over. Audiences that watch on phones respond to video built for the phone. For live sports, that means vertical is part of the plan from the start, with graphics and framing designed for it, not a crop applied afterward.

The Slipstream take

Signal this week What it rewards
Cheap, fast AI judgment on video Clips found and scored in seconds
AI labels aren't permission Knowing exactly who is in each clip
AI highlights at the Youth Olympics Coverage for every sport, not just the headliners
Live vertical beats the crop Publishing in the format fans watch

The pieces of an automated highlight workflow keep getting cheaper and more common, from quick AI judgments on video to AI highlights at the Olympics. What separates good output from noise is the same as ever: the right moments, tagged to the right athletes, published fast in the formats fans watch. That's what Slipstream does. It turns live sports streams into finished, searchable, publishable highlight clips.

Book a demo

Share this post