Slipstream Blog

AI News This Week: Instagram Tests 90-Minute Videos (Oct 30)

rayodesoltx@gmail.com (Chris) — Sat, 31 Oct 2026 16:16:49 GMT

AI News This Week · Oct 30, 2026

Instagram is testing 90-minute videos and a tool that turns long videos into Reels that link back to the full thing. That puts the platform built on short clips into the long-form business, and it makes the clip the front door to everything else you publish.

Elsewhere this week: Google's earnings showed where YouTube's ad money now comes from, Europe's AI labelling rules got a sports-specific reading, Mistral readied open weights for a giant multimodal model, and the FIA picked an AI partner. Here is what happened and what it means if you run live sports video.

Instagram goes long-form

On Monday, TechCrunch reported that Instagram is testing a new upload type called "Features" that allows videos up to 90 minutes. The test started in Brazil and will expand to a small U.S. test. Instagram is also testing "Clips as Reels," which turns long videos into Reels that link back to the full video, and viewers can save the longer video to watch on the Instagram TV app. Instagram is also working on better Live discovery and a profile hub for episodic Reels.

Money is the open question. Instagram head Adam Mosseri said the current performance payout program costs Instagram far more than it earns, and a Meta spokesperson said it is too early to share a long-term monetization model. For comparison, the same report notes YouTube Partner Program creators get 55% of ad revenue from their videos.

Why it matters for live sports video: This is the format sports publishers already think in: a full game or session, plus the moments that point back to it. If clips become the trailer for a full replay on the same platform, the job shifts from posting highlights to linking every highlight to its source game. That only works if your clips are cut fast, labelled with the right teams and athletes, and easy to find later.

YouTube's money is in the living room

Alphabet reported third-quarter results on Wednesday. YouTube ad revenue was $11.21 billion, up 9.2% from $10.26 billion a year earlier. On the earnings call, Google said the living room made up more than one-third of YouTube's ad revenue in the quarter, that more than 20 million videos are uploaded every day, and that Shorts now has a conversational editing assistant and "series" that group Shorts into seasons and episodes. YouTube Music and Premium passed 150 million subscribers, including trials. Per TheDesk, YouTube topped 15% of all U.S. TV viewing time in Nielsen's August measurement for the first time.

Sundar Pichai also said the Gemini app has 1 billion users, and described Gemini 4 Carbon as Google's strongest model, rolling out first to a limited group of cyber defenders.

Why it matters for live sports video: Sports video on YouTube is not just a phone product anymore. It is TV. Clips that hold up on a big screen, with clean framing and correct names, are worth more than ones cut only for a vertical feed. And with 20 million uploads a day, the clips that get found are the ones with good metadata.

Europe's AI labels reach sport

A Lewis Silkin briefing published today lays out what the EU AI Act's deepfake disclosure rules, in force since August 2, mean for sport and media. Businesses that use AI to create content that could pass as real must label it. Under the approved voluntary Code of Practice, most deepfakes need a label at the start, repeated at intervals and after breaks such as ad breaks, placed by default away from other graphics, such as the top-right corner.

The briefing says sports content will rarely qualify for the lighter rules for creative or fictional works. Rights holders and federations count as deployers, the rules can reach non-EU businesses whose content foreseeably reaches EU audiences, and fines can reach €15 million or 3% of global turnover. Virtual presenters, synthetic set extensions and AI-assisted graphics may all count.

Why it matters for live sports video: Cutting a real moment from a real game is different from generating one. But anything synthetic in the frame, from a virtual host to an AI-built graphic, now needs a plan for labels that survive distribution. Put it in partner contracts: who checks each asset, which label, and who keeps it downstream.

Mistral's open multimodal giant

Mistral Large 4 is a roughly 1-trillion-parameter model with 52 billion active parameters, natively multimodal with image input. Mistral says the weights drop by the end of this month, so the release may land any day. The preview API is priced at $0.70 per million input tokens and $2.10 per million output tokens, and the model was trained on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's European data centers. The announcement does not mention video input. Mistral's benchmark scores are self-reported.

Why it matters for live sports video: Strong open-weight multimodal models can run on private cloud or on-premises, which matters to leagues and broadcasters with rights or data rules that keep footage in-house. Image understanding is not video understanding yet, but the gap keeps closing.

The FIA picks an AI partner

The FIA named AWS its Official AI and Cloud Computing Partner on Thursday, covering motor sport from grassroots to Formula 1. The announced uses are mostly behind the scenes: aerodynamic simulations for future regulations, AI tools for the FIA's technical teams, and an expanded digital licensing platform. No financial terms were given, and the announcement does not mention video or fan content.

Why it matters for live sports video: Governing bodies are signing official AI partners, starting with operations rather than content. Expect the next wave of these deals to reach media rights, archives and fan video, where federations already sit on years of footage.

The Slipstream take

The week's thread: platforms want both the full game and the clip that points to it, the biggest audience is watching on TVs, and regulators want to know what was made by AI. All three reward the same thing: real moments, cut fast, labelled correctly and easy to find. That is the job Slipstream does, turning live sports streams into finished, searchable, publishable highlight clips.

[YouTube episode embed goes here]

Book a demo

Go deeper

Why the First Hour After a Moment Matters Most

rayodesoltx@gmail.com (Chris) — Thu, 29 Oct 2026 14:00:01 GMT

Speed on social

A last-second winner is the most valuable piece of content your organization will produce all week. It is also the most perishable. The fans who care most are on their phones while it happens, and the platforms they use reward posts that pick up engagement quickly. Post the clip three hours later and you are no longer part of the conversation. You are a recap of it.

This post looks at why that first window matters so much, what the platforms themselves say about how they rank content, and what a social lead can realistically get out the door in the first hour after a big moment.

Fans are already scrolling

The old model assumed fans watched the game, then went looking for highlights afterwards. Younger fans increasingly do both at once, or skip the full game entirely.

In a survey of 2,500 Americans run in May and June 2026, Bango found that 58% of 18 to 24-year-olds and 59% of 25 to 34-year-olds often follow games through clips, updates or social media instead of watching the full match. Across all respondents the figure was 46%. Around half of the younger groups also said highlights are replacing live games for them.

Often follow games through clips, updates or social instead of the full match

Ages 18-24 · 58%

Ages 25-34 · 59%

All US respondents · 46%

Source: Bango, Subscription Snapshot: Sports SVOD (2,500 US consumers, May-June 2026)

Other 2026 research points the same way. Altman Solon's Global Sports Survey of 6,000 fans across the US and five European markets found that only 39% of 18 to 24-year-olds typically watch entire live events, against 61% of fans aged 65 and over, as The Desk reported. And Hub Entertainment Research found that 45% of US fans under 35 get most or all of their non-live sports content from social platforms.

The Hub study also captures what happens during the game. Among fans aged 13 to 34 who use social media while a live game is on, 62% say that activity relates to the game itself. For fans aged 55 to 74, it is 33%. That is your audience, phone in hand, looking for the moment they just saw or just heard about.

The conversation about a moment peaks while the game is still live. The clip that arrives during that peak gets shared into it. The clip that arrives afterwards has to start a new one.

What the platforms say

No platform publishes a rule that says "post within an hour." It is worth being honest about that. But the official explanations they do publish make the case for speed fairly clearly.

Instagram. In its 2023 ranking explainer, Instagram lists popularity signals for Feed posts that include "how quickly people are liking, commenting, sharing and saving a post," alongside when it was posted. For Reels, it says the key predictions are whether someone will reshare a reel, watch it all the way through, or like it. Early, fast engagement is a signal in its own right, and reshares matter.

YouTube. YouTube's guide to how recommendations work centres on what each viewer has watched, liked and told YouTube they enjoyed, including satisfaction surveys. Its Trending Charts are more explicit about momentum: they weigh how fast a video's views are growing and the age of the video, and they refresh roughly every 30 minutes.

TikTok. TikTok said in 2020 that follower count is not a direct factor in its For You recommendations, which lean on how people interact with videos plus details like captions, sounds and hashtags (TikTok Newsroom). For a smaller club or event, that matters: a clip that lands while people are actively searching and sharing can travel well past your follower base.

Put those together and the logic is simple. Platforms reward content people engage with, and in some places reward how fast that engagement arrives. People engage most with a sports moment while it is happening. So the clip that is ready when the interest is highest has the best shot at the signals that drive reach.

It also helps to know where young fans are. Pew Research Center found in 2025 that eight in ten US adults aged 18 to 29 use Instagram, and roughly half of that age group goes on TikTok at least once a day. Those daily habits are the windows your clips are competing for.

A first-hour playbook

Most social teams at clubs, leagues and events are small. The goal is not to post everything instantly. It is to make sure the moments that matter are out while people still care, then build from there.

1Decide what counts before kickoff Agree in advance which moments get posted immediately: scores, big saves, records, upsets, a standout ride or run. Pre-write caption templates, hashtags and athlete handles so nobody is typing from scratch under pressure.

2Ship the moment, not the montage One clean clip of one play, with who, what and the score in the caption, beats a polished package that arrives after the final whistle. The montage can come later.

3Build it to be shared Instagram says reshares and full watch-throughs feed its Reels predictions. Keep clips tight, open on the action, and tag the athlete, team and opponent so the post spreads through their followers too.

4Get clips to athletes fast Athletes posting their own moment while it is fresh multiplies your reach. Make it easy for them to get the file in minutes, not days.

5Stay in the comments Replying, pinning a good comment and reposting fan reactions in the first hour keeps the post active while engagement is building.

6Recut once the dust settles After the event, turn the same clips into a recap, a best-of or an athlete feature. Fast clips earn the attention in the moment. Well-organised clips keep paying off next week.

One more thing to check before game day: make sure you know which footage you have the rights to post and where. Speed does not help if a clip gets taken down.

The hard part of this playbook is not knowing what to do. It is having the clip ready in time. For most small teams, the bottleneck is someone scrubbing through a stream, trimming, exporting and uploading while the next moment is already happening. We wrote about that gap in The Camera Problem Is Solved. The Content Problem Is the Opportunity.

Where Slipstream fits

Slipstream takes the live video your organization already produces and turns it into timely, finished, searchable, publishable highlights, delivered in under a minute. It works on top of the cameras and streams you already have: it ingests the feed, understands what happened, segments and clips the moments, enriches and transforms them, organizes them and gets them ready to publish.

That means your social lead spends the first hour choosing and posting moments, not hunting for them. And through our platform partner LiveHeats, athletes get a locker of their own clips, which makes step four a lot easier.

Book a demo

Go deeper

Seattle AI Week 2026: The Best Clips From Each Day

rayodesoltx@gmail.com (Chris) — Mon, 26 Oct 2026 19:43:39 GMT

Seattle AI Week 2026

Slipstream is clipping the Trnout sessions at Thinkspace during Seattle AI Week, October 26 to 30.

Each session runs through Slipstream while it's live, and the clips post to YouTube the same day in horizontal and vertical. Monday was AI Frontier Day. Tuesday was AI GTM Day, which ended with a VC panel. This post has three clips from each day, newest day first, and gets a new section after each day of the event.

Every clip from the week is in the Seattle AI Week 2026 playlist.

Day 2: AI GTM Day

Tuesday, October 27. Aniket Naravanekar of Trnout moderated the VC panel with Martin Tobias, General Partner at Incisive Ventures, and Kat Winston of Equity Angels. All three clips are Tobias.

The founder at the poker table

Tobias made a seed investment after watching how a founder played poker. He says he learns more about how someone handles risk at a card table than in a pitch meeting.

DocuSign said no to $20M a year

FedEx offered DocuSign $20 million a year to white-label its software. DocuSign turned it down so customers would keep seeing its own name.

Test your sales channels before you raise

Naravanekar opens with a startup that reached millions of customers through referrals, influencers and brand placement before it raised a dollar. Tobias wants founders to run small, cheap tests on their distribution channels first, so an investor's money goes into a channel that already works.

Day 1: AI Frontier Day

Monday, October 26. GeekWire chairman Jonathan Sposato interviewed Alvin Wang Graylin, author of Our Next Reality, former China president of HTC and a Digital Fellow at Stanford HAI.

China already regulates AI

A common case against US AI regulation is that China won't regulate either. Graylin says China's AI safety rules are among the strictest anywhere, and executives can be held personally liable when their models cause harm.

Use AI backwards

Graylin writes his own first draft, then asks AI to tear it apart. Starting with the AI's answer, he says, means you stop working the problem yourself.

Enterprise AI stalls on people

Graylin puts about 90% of failed enterprise AI rollouts down to culture. Staff worry about their jobs, nobody is rewarded for experimenting, and AI gets bolted onto old workflows instead of changing them.

Day 3, Agentic AI + Infrastructure Day, goes up here after today's sessions. The Seattle AI Week calendar has the rest of the schedule.

Want clips from your own event the same day? Book a demo.

Book a demo

AI Model Rankings: Kimi K3.1 Pro Debuts at P4 (Oct 26)

rayodesoltx@gmail.com (Chris) — Mon, 26 Oct 2026 16:08:37 GMT

Weekly AI rankings · Wk 44

A brand-new model just landed one point off the podium. Moonshot's Kimi K3.1 Pro entered the Artificial Analysis Intelligence Index at P4 with a score of 55, three points behind the leader. Claude Opus 5.5 still holds P1 at 58.

The short answer

No. 1 this week: Claude Opus 5.5 (Anthropic), 58. Unchanged for a third straight week.
Biggest move: Kimi K3.1 Pro (Moonshot) debuts at No. 4 with 55, 11 points above Kimi K3.
Left the top 10: Meta's Muse Spark 1.3. Artificial Analysis no longer lists it, so this is a delisting, not a score drop.
Compared: Oct 19 and Oct 26, 2026.

The top 10 this week

Each model appears once, at its best score across reasoning-effort settings, so the top 10 is ten different models.

Top 10 by model · Artificial Analysis Intelligence Index · Oct 26, 2026

Rank	Model	Lab	Score	vs Oct 19
P1	Claude Opus 5.5	Anthropic	58	=
P2	Claude Sonnet 5.5	Anthropic	56	=
P3	Muse Wonder Full	Meta	56	=
P4	Kimi K3.1 Pro	Moonshot	55	New
P5	Claude Fable 5.1	Anthropic	53	Down 1
P6	GPT-6 Astra	OpenAI	53	Down 1
P7	Gemini 4 Argon	Google	53	Down 1
P8	Muse Wonder	Meta	52	Down 1
P9	GPT-6.1 Sol	OpenAI	52	Down 1
P10	Grok 4.7	SpaceXAI	46	=

Muse Spark 1.3 (48), last week's No. 9, is no longer listed on the leaderboard. Grok 4.7 holds No. 10 on page order; Xiaomi's MiMo-V2.6-Pro also scores 46.

What moved

What is Kimi K3.1 Pro?

Kimi K3.1 Pro is a new reasoning model from Moonshot AI. Artificial Analysis added it on Oct 23 with an Intelligence Index score of 55, which puts it fourth among all models, behind Claude Opus 5.5 (58), Claude Sonnet 5.5 (56) and Meta's Muse Wonder Full (56). That is 11 points above its predecessor, Kimi K3, at 44.

According to Artificial Analysis's launch write-up, the biggest gains are in agentic work. K3.1 Pro posts the highest score of any model on Terminal-Bench 4.0, at 67%, up from 13% for Kimi K3. It also ranks third on GDPval-AA, behind only Claude Opus 5.5 and Muse Wonder Full.

Is Kimi K3.1 Pro open weights?

Not yet. Moonshot has not published the weights, and Artificial Analysis says it expects them soon after launch. Once they are out, K3.1 Pro becomes the highest-scoring open-weights model, and the gap between the best open-weights model and the best proprietary one narrows from 12 points to 3.

Five models slide one place

Claude Fable 5.1, GPT-6 Astra, Gemini 4 Argon, Muse Wonder and GPT-6.1 Sol each dropped one place. None of their scores changed. They moved only because Kimi K3.1 Pro entered above them.

Why did Muse Spark 1.3 disappear?

It is no longer listed on the leaderboard. The same happened this week to Kimi K3 and to the earlier DeepSeek V4.1 Flash, which have newer versions in the table. A model leaving the list is not the same as a model getting worse, so we mark these as delisted rather than as drops.

Further down the table

Two more models entered the top 50: InclusionAI's Ling 3.1 Flash 1019 at 44 (added Oct 22) and MiniMax-M3.1-Flash at 43 (added Oct 19). The changelog lists no score updates for models already on the board.

What it means for sports video

Near-top scores are getting cheaper. On the leaderboard, Kimi K3.1 Pro costs about $2.11 per Intelligence Index task, against $5.98 for Claude Opus 5.5 at max effort and $5.46 for Claude Sonnet 5.5 at max. Moonshot's list price is $2 per million input tokens and $10 per million output tokens, 33% below Kimi K3. For teams processing a high volume of clips, cost per task matters as much as the top score.

Speed is the trade-off. Its median output speed on the leaderboard is 51 tokens per second, about half of Claude Opus 5.5 at max (98). Live workflows that need answers in seconds should test both numbers, not just the score.

Agentic gains count most for automation. The jump on Terminal-Bench and agentic workflow tests is the part most relevant to automating multi-step jobs, such as tagging, packaging and publishing clips. K3.1 Pro takes text and image input. The Intelligence Index does not measure video understanding, so test any model on your own footage before switching.

How to read these rankings

Frequently asked questions

What is the best AI model right now?

As of Oct 26, 2026, Claude Opus 5.5 from Anthropic ranks first on the Artificial Analysis Intelligence Index, with a score of 58.

Who makes Kimi K3.1 Pro?

Moonshot AI, the company behind the Kimi models. K3.1 Pro has a 1 million token context window and accepts text and image input.

Why does the official leaderboard list one model several times?

Artificial Analysis scores each reasoning-effort setting separately, such as max, xhigh and high. We list each model once, at its best setting. That is why Kimi K3.1 Pro is No. 4 here and the fifth row on the official page.

How often are these rankings updated?

Artificial Analysis adds models as it evaluates them, often within days of release. We publish this comparison every Monday, against the week before.

Source: Artificial Analysis Intelligence Index, best score per model, compared Oct 19 and Oct 26, 2026. Cost per task and output speed from the same leaderboard on Oct 26.

Slipstream turns live sports video into finished, searchable, publishable highlights. If you want to see how that works for your events, book a time with us.

Book a demo

Go deeper

AI Model Rankings: Meta Debuts at P4 (Oct 19)

rayodesoltx@gmail.com (Chris) — Mon, 26 Oct 2026 15:12:58 GMT

Weekly AI rankings · Wk 43

Meta's Muse Wonder Full (max) entered the Artificial Analysis Intelligence Index at P4 this week with a score of 56, tied with P2 and two points behind the leader. Claude Opus 5.5 (max) held P1 at 58, so the top spot did not change, but the chasing pack got a lot more crowded.

Every week we track the top 50 rows of the Artificial Analysis leaderboard and compare them with the week before. This edition compares Oct 9 with Oct 19, 2026. Here is the top 10 and what changed.

The top 10 this week

Top 10 · Artificial Analysis Intelligence Index · Oct 19, 2026

Rank	Model	Creator	Score	vs Oct 9
P1	Claude Opus 5.5 (max with fallback)	Anthropic	58	=
P2	Claude Sonnet 5.5 (max with fallback)	Anthropic	56	=
P3	Claude Opus 5.5 (xhigh with fallback)	Anthropic	56	=
P4	Muse Wonder Full (max)	Meta	56	New
P5	Claude Opus 5.5 (high with fallback)	Anthropic	54	Down 1
P6	Claude Fable 5.1 (max with fallback)	Anthropic	53	Down 1
P7	Claude Fable 5.1 (xhigh with fallback)	Anthropic	53	Down 1
P8	GPT-6 Astra (max)	OpenAI	53	Down 1
P9	Gemini 4 Argon (high)	Google	53	Down 1
P10	Muse Wonder (max)	Meta	52	New

Effort levels count as separate rows on the leaderboard, as they do on the page. That is why one model, such as Claude Opus 5.5, can appear several times at different settings.

What moved

Meta jumps into the top 4

The Artificial Analysis changelog shows two Meta models added on Oct 12: Muse Wonder Full (max) at 56 and Muse Wonder (max) at 52. The first landed at P4, level on score with Claude Sonnet 5.5 (max) and Claude Opus 5.5 (xhigh). The second landed at P10.

Before this, Meta's best row in the top 50 was Muse Spark 1.3 (max) at 48, which now sits at P21. So Meta's top score rose by 8 points in one release, from 48 to 56. That moves Meta from the middle of the table to the front group in a single week.

P1 holds, with a two-point gap

Claude Opus 5.5 (max with fallback) stays at P1 with 58. The gap to P2 is 2 points, the same as last week. Anthropic still holds the top three rows, and Claude Opus 5.5 also appears at P5 on its high setting.

The leader did not lose any ground. The story at the top is not a pass. It is a new name arriving just two points behind.

Everyone else slides, but no scores fell

Five rows each dropped one place: Claude Opus 5.5 (high), both Claude Fable 5.1 rows, OpenAI's GPT-6 Astra (max) and Google's Gemini 4 Argon (high). None of their scores changed. They moved down only because Meta's model entered above them.

That matters when you read a ranking. A lower position does not always mean a model got worse. The changelog lists no score updates or methodology changes to the Intelligence Index since version 4.3.2 on Sep 19, so every move this week comes from new models being added. In other words, this was a week of arrivals, not upgrades.

The top 10 by lab

Count the top 10 by creator and the shift is clear. On Oct 9, Anthropic had seven rows, OpenAI two and Google one. On Oct 19, Anthropic has six, Meta two, OpenAI one and Google one. OpenAI's GPT-6 Astra (xhigh) and Anthropic's Claude Sonnet 5.5 (xhigh) were the two rows pushed out of the top 10, to P11 and P12, again with unchanged scores of 52.

Across the full top 50, Meta now has five rows, up from two. Anthropic has 16 and OpenAI 11. Anthropic and OpenAI still fill more than half the table, largely because each of their models appears at several effort settings.

Further down: a newer DeepSeek and four drop-outs

On Oct 18 the changelog added DeepSeek V4.1 Flash 1017 (max) at 45. It enters at P30, six points above the earlier DeepSeek V4.1 Flash (max) row, which scores 39 at P48. Meta also added Muse Spark Open (xhigh) at 39 on Oct 15, which enters at P49.

With four new rows in the top 50, four rows scoring 38 fell out: GPT-6 Luna (max), GPT-5.6 Terra (xhigh), MiMo-V2.6-Flash and Claude Haiku 5.5 (high). Their scores did not change either.

What it means for sports video

Four labs now sit within five points of P1. Anthropic leads at 58, Meta is at 56, and OpenAI and Google are both at 53. For teams turning live sports video into clips, that means more than one strong option at the top end of general intelligence, and less reason to lock into a single provider. When the top of the table is this close, a small change in price, speed or availability can matter more than a two-point score difference.

The table also shows how much the effort setting changes the score. Claude Opus 5.5 scores 58 at max, 54 at high and 51 at medium. Clipping is high-volume work, so the question is rarely which model scores highest. It is which setting is good enough for the job at a speed and cost that works across thousands of clips. The Intelligence Index tells you about general reasoning. It does not tell you how well a model understands video, so test against your own footage before switching.

How to read these rankings

All scores are Artificial Analysis's own Intelligence Index results, shown as whole numbers as they appear on the leaderboard. The index measures general intelligence across a mix of evaluations and is not a video-specific benchmark.

Source: Artificial Analysis Intelligence Index, top 50 rows, compared Oct 9 and Oct 19, 2026. The Oct 9 view was rebuilt from the Oct 16 leaderboard and the changelog, since the leaderboard does not publish history.

Slipstream turns live sports video into finished, searchable, publishable highlights. If you want to see how that works for your events, book a time with us.

Book a demo

Go deeper

Live From Laguna: Coast Summit, Clip by Clip

rayodesoltx@gmail.com (Chris) — Thu, 22 Oct 2026 16:34:26 GMT

Live updates, Thursday, October 22. Newest at the top.

Slipstream clipped the live stream from Coast Summit, part of the Coast Film & Music Festival in Laguna Beach. As the stream ran, Slipstream picked out the moments worth sharing and turned them into finished clips.

Here's everything we clipped on Thursday.

Live updates

2:32 PM PT · Sledding down an iceberg

Andy Mann's case for falling in love with nature: footage of him sliding off a giant Greenland iceberg on an inflatable pizza slice.

Clip: Sledding Down an Iceberg on an Inflatable Pizza Slice

Andy Mann · Coast Summit, Coast Film & Music Festival, Laguna Beach

Clipped from the live stream by Slipstream · Posted by Stronghold Labs on YouTube

Slipstream Blog

AI News This Week: Instagram Tests 90-Minute Videos (Oct 30)

Instagram goes long-form

YouTube's money is in the living room

Europe's AI labels reach sport

Mistral's open multimodal giant

The FIA picks an AI partner

The Slipstream take

Go deeper

Why the First Hour After a Moment Matters Most

Fans are already scrolling

What the platforms say

A first-hour playbook

Where Slipstream fits

Go deeper

Seattle AI Week 2026: The Best Clips From Each Day

Day 2: AI GTM Day

The founder at the poker table

DocuSign said no to $20M a year

Test your sales channels before you raise

Day 1: AI Frontier Day

China already regulates AI

Use AI backwards

Enterprise AI stalls on people

AI Model Rankings: Kimi K3.1 Pro Debuts at P4 (Oct 26)

The short answer

The top 10 this week

What moved

What is Kimi K3.1 Pro?

Is Kimi K3.1 Pro open weights?

Five models slide one place

Why did Muse Spark 1.3 disappear?

Further down the table

What it means for sports video

How to read these rankings

Frequently asked questions

What is the best AI model right now?

Who makes Kimi K3.1 Pro?

Why does the official leaderboard list one model several times?

How often are these rankings updated?

Go deeper

AI Model Rankings: Meta Debuts at P4 (Oct 19)

The top 10 this week

What moved

Meta jumps into the top 4

P1 holds, with a two-point gap

Everyone else slides, but no scores fell

The top 10 by lab

Further down: a newer DeepSeek and four drop-outs

What it means for sports video

How to read these rankings

Go deeper

Live From Laguna: Coast Summit, Clip by Clip

Live updates

2:32 PM PT · Sledding down an iceberg

2:30 PM PT · Walk toward the light

2:26 PM PT · Keeping up with Sylvia Earle

2:22 PM PT · Tagging a sixgill shark

2:20 PM PT · A tiger shark takes the camera

2:18 PM PT · First dive, with a Navy SEAL

2:17 PM PT · 50 bull sharks in Fiji

2:10 PM PT · A polar bear at the porthole

2:09 PM PT · Nothing at the top

2:02 PM PT · The $400 photo

1:59 PM PT · Andy Mann on visual storytelling

1:55 PM PT · Building awe into daily life

1:53 PM PT · Alexander Hillary on starting small

1:49 PM PT · Awe as a moral emotion

1:48 PM PT · Measuring kindness

1:46 PM PT · One experience, a whole community

1:40 PM PT · Peter Hillary on his father

1:10 PM PT · Breathwork with Ryah Arthur

11:50 AM PT · Music panel on Phil Lesh

11:47 AM PT · A sushi fun fact

10:22 AM PT · Biomimicry, continued

10:21 AM PT · Biomimicry and nature's timeline

10:07 AM PT · We are nature

9:36 AM PT · Whale songs in space

9:12 AM PT · Opening ceremony on the beach

How this works