FORMA's
Five Things.

Live And In Person

AI Is Moving At Lightning Speed.
Keeping Up Is Our Job.

One day at the AC Hotel South Bay in El Segundo, built to hand you the latest tools and AI skills you can deploy in your everyday work the moment you walk out. Catered lunch, networking, free premium AI tools, and third party accreditation issued on completion. You leave with ready-built AI workflows, not notes.

Liquid Death Takes A Stand
Against Data Centers.

Liquid Death doesn't bottle water. It bottles marketing expertise. Seven years of going exactly where every other brand won't, from a death metal album built out of its own hate mail to a valuation north of a billion dollars earned almost entirely on nerve.

This month they went after AI.

On August 18, Liquid Death and Garage Beer released "We Want Your Pee," a 90-second music video starring Jason Kelce. It opens with him filling a mason jar at his toilet while explaining that AI data centers burn through millions of gallons of water for cooling. His proposed solution is human urine. He walks into a meadow, starts singing, and a crowd joins him with jars in hand, marching toward a post office. A disclaimer near the end asks you not to actually mail anything.

They also sold a limited edition glass jar called the Data Center Coolant Collector, with instructions to fill it and ship it to the facility of your choosing. It sold out.

The reason this works is that the sentiment was already there. Gallup found this year that 71% of Americans oppose an AI data center being built in their area, with 48% strongly opposed. More people object to a data center nearby than a nuclear power plant, which sits at 53%. Strong support is 7%.

Notice too that there is no AI anywhere in this creative. No generated footage, no synthetic voice, no prompt-built script. They used AI as a subject instead of a tool.

The backlash is the thing to watch here. Complaints about data centers have lived in city council meetings and utility bills for a couple of years now, mostly out of earshot. This week they showed up in a beer ad with a former NFL center in it. That is what it looks like when a local gripe turns into a national mood.

Read more and watch the spot →

The Music Industry Has Had It.
And Now They Are Doing Something About It.

Starting with the chart dated August 31, Australia's recording industry body will bar wholly AI-generated tracks from its official charts. Songs using AI in a supporting role stay eligible. ARIA is the first national chart body anywhere to put real rules on paper.

The song that forced it is the good part. A Queensland producer's cover of "Like a Prayer" became the most-played song on Australian radio and reached number four on two ARIA charts in July. Critics suspected AI. He said he used it as a tool, then quietly updated his Spotify credits to disclose that the vocals and drums were AI-generated. The disclosure arrived after the chart run, not before.

Then on August 28, Sony Music Publishing and Warner Chappell sued Anthropic in California federal court over lyrics used to train Claude, naming the company's chief executive and cofounder as individual defendants. The complaint calls it "one of the largest and most blatant ongoing thefts of intellectual property in history." They're seeking up to $150,000 per work across thousands of songs, including "Eye of the Tiger." Universal, Concord and ABKCO already sued in 2023, so all three major publishers are now in court. Anthropic says it disagrees and will defend itself.

Two institutions on opposite ends of the world, completely unrelated to each other, moving in the same week. What connects them is which end of the business each one is protecting. ARIA is defending the output, deciding what gets to count as a human record. The publishers are defending the input, deciding what can legally go into the model in the first place.

Look at where ARIA drew the line. A track qualifies if humans wrote it and performed the lead vocal and the primary instruments. Somebody finally had to put on paper what human-made means, and that is the sentence they landed on. Everything not in it is still up for grabs, and the courts will define how this plays out.

Read more →

A Billion People Use ChatGPT.
About 50 Million Pay For It.

ChatGPT crossed a billion monthly users in 2026, faster than any app in history. Faster than TikTok, YouTube, or Instagram. Roughly 50 million of those people pay for it.

The race for downloads is over. We all did it. What starts now is the race for customers, and specifically for paying ones.

Sensor Tower estimates ChatGPT converts about 8% of its users into subscribers. Claude converts 13%, the best rate of any major assistant, pulling in around $2.76 per US mobile user against ChatGPT's $1.74. A platform with a fraction of the audience earns more per head than the one that created the category.

Card data from PNC points the same direction. Only 2.2% of households paid for any generative AI this year, and the ones who did were spending about $31 a month, up from a year earlier. Very few payers, each writing a bigger check.

The money is moving even as growth slows. In-app spending on AI apps is on pace to pass $4.2 billion in the first half of 2026, up from $1.83 billion a year ago. That's a 130% increase. Google launched a $4.99 plan. OpenAI launched an $8 one. Nobody builds a cheap tier when the expensive tier is working.

The last two decades of the cell phone wars tell you where this goes. A crowded field narrows to a few players, and then those players spend years fighting over the same monthly line item on your statement. AI subscriptions are about to become one of the biggest recurring consumer categories out there.

So if you've been using all of this for free, you're not being cheap. You're the majority, and you're the part of the business model nobody has solved yet.

Read more →

AI's Most Famous Optimist
Just Wrote A Very Different Memo.

On August 26, Bill Gates published an essay arguing that even under the best circumstances, the transition into the AI era will be one of the most turbulent periods in human history. He wrote that we aren't preparing for it, that he sees no evidence leaders or communities are confronting the challenges adequately, and that there's no plan for easing the entry.

He told Axios that on the current course and speed, there's a very high chance of a net negative outcome. The good things are arriving slowly. Apart from robots, he says the bad things are already here.

Then he said the part nobody in his position says out loud. Speaking to GeekWire, Gates described an industry where anyone willing to admit AI might have negative effects gets told they're hurting the PR while everyone tries to raise trillions. He added that he knows they're all worried, essentially every AI leader he knows.

And don't think the people running these companies missed it. This industry was built by a generation that grew up on Gates and Jobs, that studied how those two turned software into infrastructure. When the man who wrote that playbook tells the industry to get its act together, it doesn't land like criticism from the outside. It lands like a note from the person who invented the job.

The reversal is real but it's been building. His January letter was already hedged, what he called optimism with footnotes. This is where the hedging tipped over. He told GeekWire that the further into the memo he got, the more convinced he became.

A man who spent a decade funding AI to make vaccines wrote himself into a worse mood.

Read more →

Your AI Reads Like You. Skim.
Top, Bottom, And Not Much In Between.

Paste a long document into any AI tool and ask a question about it. Where the answer sits inside that document changes whether you get it.

Stanford researchers tested this by hiding the answer at different depths in a long input and measuring accuracy at each position. The results plotted into a U. Strong at the opening, strong at the close, sagging badly through the middle. Call it the U-lost in the middle, and it held up across every model they tried.

The number that should change how you work: when the answer was buried in the middle, one model scored worse than when it was handed nothing at all. The right information, in the wrong position, was worse than no information.

This isn't forgetting. Everything you pasted is still sitting in the context window. The opening gets disproportionate attention no matter what it says, the ending wins because these systems are built to predict what comes next, and the middle is stuck competing with everything on both sides of it.

There's a reason baked into training too. Instructions in training data almost always sit at the top. Human writing front-loads the point and back-loads the conclusion. After enough examples, the model learned where we hide the important parts.

So a bigger context window is not the same thing as a context window that gets used. The capacity went up. The attention did not spread out to match.

FORMA's Take

You know the important part is in there. The AI knows it's in there too. What it doesn't know is that it matters to you.

Position is the only signal it has for importance, and you buried yours on page four. Say the critical thing at the top, then say it again at the bottom.

FORMA Tools

Not sure which AI model is actually right for your work?

Six questions. A personalized model recommendation. No benchmark tables, no guesswork. Just a clear answer based on how you actually work.

FORMA Programs

You have learned to talk to AI.
Now learn to build with it.

The next Build With AI session is forming now. Tell us what you want to get your hands on and we will build the session around it.