Howdy. It’s Barsee again.

Happy Tuesday, AI family, and welcome back to AI Valley.

Today’s climb through the Valley reveals:

• OpenAI figured out why ChatGPT hallucinates

• OpenAI backs AI-generated animated movie

• Anthropic to pay $1.5B in author copyright settlement

• Google court filing admits “rapid decline” of the open web

• Plus trending AI tools, posts, and resources

Let’s dive into the Valley of AI…

WISPR FLOW

Your keyboard called, it’s taking a long weekend ⌨️✈️

We’ve been banging away on keyboards for 150 years.

Until now, voice dictation hasn’t been reliable enough to change that.

Wispr Flow finally delivers the no-edit confidence we’ve all been waiting for:

• 4× quicker than typing. Dictate emails, docs, and DMs in real time and save precious hours every week.

• AI auto-edits on the fly. Flow cleans filler words, fixes grammar, and formats perfectly as you speak.

• Works inside every app with no setup. Fly through Slack notifications, give more context to ChatGPT, or brain dump into Notion.

• Use it at your desk or on the go. Available on Mac, Windows, and iPhone. 

“This is the best AI product I’ve used since ChatGPT.” — Rahul Vohra, CEO, Superhuman

Give your hands a break ➜ start flowing for free today. 

_{*This is sponsored}

THROUGH THE VALLEY

1/ OpenAI figured out why ChatGPT hallucinates

OpenAI published a new paper explaining why language models sometimes make up false but confident answers, and how to fix it. They argue hallucinations don’t mainly come from data or model size but from the way benchmarks are set up to reward guessing instead of admitting uncertainty.

Most tests work like multiple-choice exams: correct answers get points, while saying “I don’t know” gets nothing. This pushes models to guess, since even a wrong but believable guess can improve scores. Over time, this teaches models to sound confident even when unsure.

OpenAI’s fix is to change evaluations: punish wrong answers more than “I don’t know,” and give partial credit when a model admits it doesn’t know. This would make humility the best strategy and help reduce hallucinations.

Why does it matter?

Hallucinations are still a major barrier to using AI in important areas. By changing incentives during training and evaluation, OpenAI’s method could make models more reliable, prioritizing accuracy and honesty over confident but wrong answers.

2/ OpenAI backs AI-generated animated movie

OpenAI is helping produce Critterz, an AI-assisted animated film, testing if generative tools can speed up production and cut costs. The movie aims to finish in nine months on a budget under $30M (far cheaper than the usual three years and much bigger budgets for animated films).

The film will use GPT-5 and image-generation models to turn sketches into full animation, while human actors provide the voices. OpenAI is aiming for a 2026 Cannes premiere.

Why does it matter?

AI has been quietly used in Hollywood before, but Critterz is the first big project openly branded as AI-driven. If it succeeds with budget and deadlines, studios may rethink animation pipelines, but audience reaction will be the true test.

3/ OpenAI projects $115B cash burn through 2029

According to The Information, OpenAI now expects to spend $115B by 2029 (about $80B more than earlier forecasts that assumed break-even by then). Costs will top $8B this year, rise to $17B in 2026, and hit $47B by 2028. Most of this is from computing costs for training and inference, plus nearly $100B planned for data centers and custom chips by 2030 to rely less on cloud providers.

On the revenue side, OpenAI projects $13B this year and $200B by 2030, up 15% from earlier estimates. ChatGPT is the main driver, expected to bring in $10B this year and nearly $90B by 2030. OpenAI also expects $110B from free users between 2026 and 2030, through ads and commissions, assuming two billion weekly active users.

4/ Anthropic to pay $1.5B in author copyright settlement

Anthropic agreed to a settlement worth at least $1.5B with authors who said their books were used without permission. The deal includes payments of about $3,000 per book plus interest, and requires deleting datasets built from that material, according to court filings.

Why does it matter?

If approved, it would be the largest reported copyright payout so far. A judge earlier said training might be “fair use,” but left the final copyright questions for trial. This settlement sets a precedent for future copyright fights, showing how expensive unlicensed data could be for AI companies.

5/ Google court filing admits “rapid decline” of the open web

In a recent court filing, Google admitted that “the open web is already in rapid decline,” even though publicly it claims search is strong and sending traffic to publishers. The filing was part of an antitrust case on Google’s ad tech business, where the DOJ has suggested breaking it up. Google argued that a breakup would make things worse, speeding up the decline of open-web display ads and hurting publishers.

The filing points to bigger changes in ad tech: AI reshaping the market, connected TV and retail media rising, and competitors moving money away from open-web ads. Google later clarified that it was only referring to open-web display ads. Leaders like Sundar Pichai and Liz Reid still defend Google Search, saying it continues to drive traffic and clicks for publishers despite AI search features.

Why does it matter?

This shows the gap between Google’s public image and its court defense. In reality, many publishers already report less traffic from AI-powered search. The outcome of the antitrust case could affect not just Google’s ad business but also the survival of the open web for independent publishers.

TRENDING TOOLS

• MovieFlo - Built by Lucasfilm & ILM vets to turn ideas into cinematic videos with an intuitive workflow *

• AlterEgo - A near-telepathic wearable that lets you communicate silently, almost at the speed of thought

• 100 Vibe Coding - Learn by doing: go from zero to your first real project in 100 interactive challenges

• Solid - AI that builds full, production-grade web apps, not disposable prototypes

• Trace - A lightning-fast AI calendar built for people who hate traditional planning

• Claude - Now capable of finding nearby spots, checking your calendar, and scheduling events on your phone

_{(*) signifies sponsored tool}

THINK PIECES / BRAIN BOOST

• Study explores potential impact of an AI bubble

• Why language models hallucinate

• GEO: Generative Engine Optimization by Arxiv

• GPT-5 Thinking in ChatGPT (aka Research Goblin) is really good at search

• How to build with Nano Banana: Complete developer tutorial

THE VALLEY GEMS

What’s trending on social today:

THAT’S ALL FOR TODAY

Thank you for reading today’s edition. That’s all for today’s issue.

💡 Help me get better and suggest new ideas at [email protected] or @heyBarsee

👍️ New reader? Subscribe here

Thanks for being here.

REACH 100K+ READERS

Acquire new customers and drive revenue by partnering with us

Sponsor AI Valley and reach over 100,000+ entrepreneurs, founders, software engineers, investors, etc.

If you’re interested in sponsoring us, email [email protected] with the subject “AI Valley Ads”.