Welcome, humans.

So apparently A16z’s Olivia Moore spent a week sending a fictional 19-year-old college woman named Janie through Bama Rush, Alabama’s viral sorority recruitment week, and people actually thought she was real. And something wild happened: Janie went kinda viral??

So how’d she do it? One ChatGPT image created Janie, Minimax 3 (truly one of the wildest open video models I’ve ever used) animated talking clips, Grok Imagine 1.5 handled the dances, and ElevenLabs (still the commercial voice leader AFAIK) added the sound. Moore made 20 videos for about 30 minutes a day and $100 in credits.

Now, Moore intentionally left TikTok’s AI disclosure off because part of the experiment was testing what the platform could detect. She says that violated TikTok’s rules., and eventually TikTok labeled 8 of the 20 videos on its own… but with no visible hit to performance. Janie still reached 1,300 followers in a week, and her first video neared 100K views. Get it, Janie!

Viewers finally spotted the fake by day two, but then: they kept watching; The Daily Mail even crowned Janie Alabama’s “most popular sorority star.” Moore’s conclusion was that AI can manufacture the character, while human taste, storytelling, and clear disclosure still shape whether people care. This is an interesting experiment now that young people officially have the AI ick…

Here’s what happened in AI today:

😹 Google paid $10M for Spirit Airlines’ AI-training data.
📰 OpenAI kept its biggest planned frontier RL run on hold.
📰 Etched raised $700M at a $21B valuation.
📰 Physical-AI startups raised $47.4B in six months.
📰 Axiom formally verified a major prime-gap theorem.

😺 Google paid $10M for Spirit Airlines’ data to train AI

When an airline goes bankrupt, you’d expect the valuable leftovers to be, y’know, planes, airport slots, software, maybe a loyalty program’s user pool. Well, Spirit Airlines apparently had another asset worth bidding on: years of data about how the company actually operated.

Apparently, Google just won a $10M bankruptcy auction for the company’s anonymized internal business data and custom software, beating AI recruiting startup Mercor’s $7.5M bid. Now nobody tell the VCs, or next time your startup is in dire straits, they’ll force you to shut down so they can scrap your data for parts!

Here’s what happened:

Google’s winning bid covered internal business records and software Spirit built for its operations.
CNN reported the data included internal communications, spreadsheets, operational records, and anonymized booking and loyalty information.
Google says identifiable customer and credit-card information is excluded from the deal.
The sale still needs bankruptcy-court approval.

Our take: Bankruptcy usually turns physical assets and intellectual property into cash. AI adds another category of the latter into a valuable commodity: a company’s accumulated operating history.

Years of support tickets, internal docs, workflows, edge cases, and mistakes can teach models how real organizations work. That makes data created as a byproduct of running a business potentially valuable on its own.

Watch what happens next in bankruptcy courts and privacy rules. The fight will be over where “company data” ends and information about customers or employees begins.

Now, this only continues to be true as long as the current large language model paradigm remains its vice grip on the AI industry. And if you buy Rich Sutton’s ideas, the only data that will truly be valuable to agents in the future will be the “current world” in which they are operating, because they’ll be able to learn from their existing world and build abstractions from it. But that might be 5-10 years out.

For more on that, check out Midweek Wisdom below!

FROM OUR PARTNERS

Nobody actually knows how AI gets used.

AI tool sprawl happened fast, and visibility never caught up. Employees paste contracts into ChatGPT, run code through Copilot, and build workflows in apps security never approved.

Harmonic Security classifies every AI interaction by task, tool, and team, so you can see which use cases drive real productivity, which tools are shelfware, and where sensitive data is headed. Across approved and unapproved apps alike.

No need to guess what 'AI adoption' means inside your company.

See Your AI ROI

🎓 AI Skill of the Day: Give Codex a 1M-token context window

OpenAI’s Tibo shared a config that apparently gives OpenAI’s current best AI model, GPT-5.6 Sol, a one-million-token context window in Codex (OpenAI’s coding app, available in the ChatGPT Desktop app).

A one million token context window means the model can keep far more code, tool output, and chat history in view before compressing older material. Tokens are the chunks of text AI counts, in case you missed that part (we got you, newbie friends!)

See, GPT-5.6 Sol supports 1.05M tokens, but Codex keeps a smaller default tuned for performance and cost. So use this only for unusually large codebases or long debugging runs.

Here’s what you do:

Open ~/.codex/config.toml and add these settings at the top, before any [section] headers:

model = "gpt-5.6-sol"
model_context_window = 1000000
model_auto_compact_token_limit = 900000

This selects Sol, sets the context budget to 1M tokens, and starts compaction at 900K to leave headroom. Restart Codex and start a new session. Tibo says ChatGPT-account sign-in now supports the override too.

For a one-off CLI session:

codex -m gpt-5.6-sol \
-c model_context_window=1000000 \
-c model_auto_compact_token_limit=900000

Have a specific skill you want to learn? Request it here.

🍪 Treats to Try

*Asterisk = from our partners (only the first one!). Advertise to 700K+ readers here!

*Base44 turns plain-English instructions into a working app with the frontend, backend, database, and user login already wired up; free plan, then $16/mo billed annually.
i-have-adhd gives AI agents an ADHD-friendly response format: lead with the next action, number steps, cut tangents, and keep progress visible.
Meridian automatically keeps a private, searchable journal of what you worked on, then drafts Jira or GitHub status updates you approve, with data kept on-device.
Outcome turns a video, article, or framework into a personalized funnel that gives each lead their own action plan, audit, or score.
Cartesia’s Sonic-3.6 makes real-time voice agents sound more lifelike across 44 languages; Cartesia says the release reached #1 on Artificial Analysis.

📰 Around the Horn

Leaked video of Apple’s new AirPod cameras (source, discussion)

Engram and Harvey trained a legal agent to study a 100M-token mock law firm, cutting average query cost versus Opus 4.8 from $1.32 to $0.13 while raising all-pass accuracy from 25% to 30%.
Anthropic reportedly hit a $65B annualized revenue pace, more than 7x its sales pace at the end of last year.
Groq raised $350M at a $3.5B valuation as it pivoted from designing AI chips to renting Nvidia-powered computing capacity.
The DOJ is probing whether Andreessen Horowitz partners improperly sat on boards of competing AI companies, a potential antitrust problem.
OpenAI said Astra may reach its highest cyber-risk tier, kept its largest planned frontier reinforcement-learning run on hold, and left some Astra and cyber workloads paused while it strengthens safeguards.
NPR reviewed nearly 1,800 pages of a suicidal woman’s ChatGPT conversations; the bot sometimes urged therapy and crisis help, but eventually produced a suicide note after twice refusing.
Etched raised $700M at a $21B valuation for its AI chips, with Jane Street leading the round and receiving the first shipped rack.
Physical-AI startups raised $47.4B across 521 deals in the first half of 2026, led by giant rounds for Waymo, Anduril, Shield AI, and Saronic.
Axiom formally verified the BGP246 prime-gap theorem in Lean 4, turning a major recent math result into a machine-checkable proof.
Apple’s camera-equipped AirPods are reportedly moving closer to reality, using cameras mainly for spatial awareness and gesture control rather than conventional photography.

🧠 Midweek Wisdom: AI Godfather Rich Sutton thinks synthetic data is a trap

He’s right, obviously. It’ll just take time (and probably some pain) for people to realize

AI pioneer Rich Sutton argues today’s language models have a basic limitation: once training stops, their weights freeze, so they cannot keep learning from new experience the way people do.

Widely considered a godfather of modern reinforcement learning (AI learning by trial and error), his sharper criticism is aimed at synthetic data. Sutton called it “a big mistake” when humans design the simulations and examples models learn from, because human assumptions still define the world the AI gets to experience.

His alternative is continual deep learning: models should keep updating from real experience, while methods like continual backprop and per-weight learning rates help them absorb new information without constantly starting over.

Awesome interview. Time to update your priors. This is the future. Time will prove it.

Also worth your attention:

SemiAnalysis analyst Dylan Patel said he’d heard Anthropic finished training Mythos 2 but isn’t releasing it; he expects the internal feedback loop could still help improve Mythos 3.
Ben Thompson warned the AI boom could hit a railroad-style financing crunch if infrastructure spending burns through available capital before AI revenue catches up.
Exo, built by Alex Krentsel with Martin Casado and Ankur Goyal, is an open-source experiment in recursive self-improvement, where an AI agent can inspect and modify the harness around itself: its prompts, memory, tools, adapters, integrations, and even parts of its operating policy.
Rachel Thomas explained why she returned to AI at Answer.AI despite agreeing with many critiques of the field: smaller, constraint-aware teams can still use AI while keeping human judgment and autonomy at the center.
Ethan Mollick pointed to early evidence that AI is accelerating discovery unevenly, with clearer movement in cyber and some math than in algorithms, a useful reality check on claims that every field is about to speed up at once.

New from The Neuron: AI Explained

Radical Numerics is one of the coolest companies we’ve got to talk to on the pod, and this episode has been criminally under-watched (y’all on notice!). If you’ve followed the debate over how AI has failed to deliver on its promise to “Cure cancer”, then you DEF need to watch this one (and our chat w/ Isomorphic Labs).

P.S: We’re trying to hit 50K subscribers on YouTube this year. Click here to help!