October 26

Your Daily Note,

One economist questioned whether AI is delivering measurable financial returns, while OpenAI revealed how it built an advertising system around conversations and Block introduced a voice model designed to sound remarkably human.

In today’s Noted:

AI may be changing how we work, but where are the profits?

We noted:

Apollo Global Management chief economist Torsten Slok says AI has yet to deliver measurable financial gains across large US companies outside the technology sector. His analysis found that revenue per employee and profit margins remained broadly flat across the S&P 493, the S&P 500 excluding seven major technology companies.

Businesses have spent heavily on AI tools, models and computing resources, but Slok argues that this spending has not yet translated into stronger financial results for much of corporate America. Meanwhile, major technology companies have seen revenue per employee and profit margins rise.

Why it matters

This is an economic interpretation of available business data, not proof that AI has produced no benefits anywhere. Individual companies may be saving time or improving specific workflows without those gains appearing in company-wide financial figures yet.

Businesses may face greater pressure to demonstrate what their AI spending actually achieves. Tools that save measurable time, reduce operating costs or improve conversion rates will be easier to justify than tools adopted simply because competitors are using them.

Block’s new voice model sounds almost human

We noted:

Block, the financial technology company behind Square and Cash App, introduced Dessa, a full-duplex speech model designed to listen and speak in real time, with natural-sounding pauses, laughter, crying and changes in vocal emotion. Block says the model was trained entirely on human speech and can run locally on consumer-grade hardware.

Traditional voice assistants often convert speech into text, generate a response, then convert that response back into speech. That process can lose tone, timing and emotional cues. Dessa is designed to process speech more directly, allowing it to listen and respond while preserving more of the qualities of natural conversation.

Why it matters

If this approach becomes practical for everyday products, voice assistants could become more useful for customer support, ordering, accessibility and hands-free workflows. Running locally could also reduce dependence on cloud infrastructure, although actual performance will depend on the hardware and application.

The more natural a voice assistant sounds, the easier it may be to use. But realism also creates risks: people may mistake an AI voice for a real person, making disclosure, consent and safeguards increasingly important.

OpenAI reveals how it built ChatGPT’s advertising system

We noted:

OpenAI published a technical explanation of the advertising infrastructure behind ChatGPT Ads. The company says the system reached more than $1 billion in annualised revenue run rate in under 200 days after launch, and more than 100,000 advertisers have run campaigns on the platform.

Traditional advertising systems often match ads to a search query or webpage. Conversations are more complicated because a user’s goal can change as they ask follow-up questions.

Why it matters

As conversational AI becomes a place where people research products and make purchasing decisions, advertising may become more contextual than traditional search ads. Businesses may need to think about how their products are represented in conversational discovery, not just how they rank in search results.

Grok releases seven minutes of its AI-generated Odyssey

We noted:

Grok released the first seven minutes of its AI-generated version of Homer’s Odyssey, calling it Episode 1 of a planned series. The project was created with Wonder Studios and director Dev Walker, following Elon Musk’s earlier promise of a full-length, historically accurate adaptation.

The release offers a glimpse of how generative video is being used to produce longer-form storytelling, not just short clips. Some scenes are visually convincing, although visual realism is not the same as historical accuracy, consistent character development or a coherent film.

Why it matters

AI-generated video may lower the cost of prototyping scenes, testing visual styles and creating storyboards. The competition is moving beyond generating impressive individual shots, and the bigger challenge is maintaining consistency and creative quality across an entire production.

For a small business, AI video can be useful for testing a campaign concept before investing in a full shoot. Start with a short scene and validate the idea before trying to automate an entire production.

OpenAI explains how it is scaling storage for more than one billion users

We noted:

OpenAI published the second part of its engineering series explaining how it scaled its internal storage platform, Habitat, to support more than one billion users across 40 geographic regions. The system manages over 500 petabytes of data, according to the company.

Habitat provides the fast, reliable access to data that OpenAI’s products need. Its engineers describe challenges including uneven database workloads, storage partitions losing throughput as they grow, regional outages and delays caused by data being far from the computing resources that need it.

Why it matters

Most users will not interact with Habitat directly. The potential benefit is more reliable and responsive access to data as OpenAI adds users, features and new products.

AI infrastructure is not just about training increasingly powerful models. Serving those models to huge numbers of people also requires databases, storage, networking and careful engineering. Reliability and latency can shape the user experience as much as model intelligence does.

Products worth knowing

 

Everything else in AI

NVIDIA’s micro duck becomes an unexpected hit, with more than 20,000 preorders

Jensen Huang showcased small robot ducks at NVIDIA GTC Berlin, and the demonstration has turned into a commercial surprise. Hugging Face CEO Clem Delangue said preorders for Pollen Robotics’ open-source Microduck had surpassed 20,000 units after the video went viral.

The 25-centimetre robot can walk, sit, fall over and perform tricks. Its movements were trained in simulation using NVIDIA’s technology, and owners are meant to be able to teach it new behaviours and share them with the community. The first deliveries are expected before Christmas 2026.

It’s a reminder that physical AI doesn’t always need to look like a factory robot. Sometimes, an approachable little duck can introduce more people to robotics.

 

LitPivot helps researchers rethink ideas as they read papers

Allen Institute for AI and researchers at the University of Pennsylvania introduced LitPivot, a system that connects research proposals with relevant academic literature. As a researcher edits an idea, LitPivot retrieves related papers, offers critiques supported by excerpts and suggests revisions that the researcher can accept, edit or reject.

In a study involving 17 researchers, participants rated ideas developed with LitPivot as better grounded in existing literature than ideas developed using a text editor and an AI chat tool. The study is small, but the concept addresses a real challenge: making sure an apparently new idea has been considered in the context of what researchers already know.

 

Six big technology promises face a reality check

MIT Technology Review’s The Download examined six ambitious technology promises, including exercise pills and nuclear-powered spacecraft, and asked how close they are to becoming reality. The issue also explores blind spots in biotech through the lens of the documentary You Can See Everything.

The broader lesson is relevant to AI, too: a compelling demonstration or bold prediction is not the same as a technology that works reliably outside controlled conditions. Progress is better judged by evidence, limitations and what remains unresolved.

See you tomorrow
same time, different note.

~ the Stackari team