- Charbel X
- Posts
- You are Unknowingly Manufacturing Data, UX Design to Generate Monetary Value, Integration problems and more
You are Unknowingly Manufacturing Data, UX Design to Generate Monetary Value, Integration problems and more
If you've ever played Pokémon-Go, you've produced valuable data for a company.
Charbel Zeaiter
November 22, 2024
Good morning!
It’s Friday. How? As always, so much going on in design, tech and product:
UX design isn't just about crafting intuitive interfaces—it's about driving measurable business outcomes. At a design event this week, leaders noted a shift in hiring preferences, favouring UX designers with strong business acumen and the ability to demonstrate game-changing metrics like increased retention, higher conversions, or operational efficiency.
Meanwhile, Niantic is showcasing the untapped potential of geospatial data by transforming Pokémon GO player insights into a cutting-edge AI navigation system, bridging gaming and AI for real-world applications such as autonomous navigation and urban planning. However, not everyone is thrilled with the implications of leveraging user data in this way.
Let’s dive in!
Yours in wonder,
Charbel
Founder of Velvet Onion, Faster Zebra and more to come …
Today’s Highlights
AI: DeepSeek’s R1-Lite AI: A More Human-Like Intelligence
Design: How UX Design Brings (Monetary) Value to a Business: A Case Study
Science & Tech: Pokémon Go is a huge AI-data warehouse
Founding: Inference time scaling > Pre-training scaling of AI models
Product: Integration problems create annoyed users
Today’s AI image: Subconscious Creation of Data to be Used by Firms
Quote for the day: Humility
AI
DeepSeek’s R1-Lite AI: A More Human-Like Intelligence
DeepSeek, a rising AI powerhouse in China, has debuted R1-Lite-Preview, a reasoning-focused model that pars OpenAI's o1 in benchmarks while offering a refreshingly open look at its decision-making process.
What makes it tick:
While OpenAI's o1 condenses its reasoning into neat summaries, R1-Lite-Preview takes the scenic route, revealing its full thought process live and unfiltered.
Early tests show it competes head-to-head with o1 in tasks like AIME and MATH, with performance actually improving as the complexity grows.
The model is accessible via DeepSeek Chat, where free users can enjoy unlimited basic interactions. Premium-level reasoning, however, is capped at 50 messages a day.
The pièce de résistance? DeepSeek plans to open-source the full R1 model, throwing open the doors to its intellectual treasure trove.
Behind the curtain
DeepSeek’s infrastructure boasts an eye-popping arsenal of 50,000 H100 chips, rivaling the computational might of top Western AI labs.
Why is this a big deal?
Just two months after OpenAI’s o1 reshaped the landscape of AI reasoning, DeepSeek’s rapid counterpunch highlights the speed of progress in the field. Although the firm remains relatively low-profile in Western circles, its open-sourcing plans could turbocharge global AI innovation—and send a strong signal to U.S.-based, closed-door AI players.
Also in AI
Design
How UX Design Brings (Monetary) Value to a Business: A Case Study
A Pay Now feature was developed by the UX team of an invoice-tracking app for small business owners. Initially, this feature struggled to catch on, with only 5% adoption due to poor discoverability and a less-than-ideal workflow that made payments cumbersome for users.
The Fix
After listening to users and gathering feedback, the UX team revamped the feature. The result? Adoption soared to 65%, and payments via the feature surged, resulting in a massive $33 million revenue increase.
Business owners (the users) raved about the improvement in cash flow and faster payments, cementing the feature as a game-changer.
Metrics That Matter(ed)
Instead of conventional UX metrics like task time or satisfaction scores, the UX team focused on outcome-driven metrics:
Total invoices sent
Invoices with Pay Now button
Dollar value of invoices
Payments via Pay Now
Revenue from payments
Payment processing time
User experience improvements (qualitative metrics)
Why It Worked
By focusing on metrics tied directly to business outcomes and user impact (faster payments, improved cash flow), the UX team delivered a narrative that not only captivated executives but demonstrated how UX is a strategic partner in business growth.
The Lesson
Forget the fluff metrics. Focus on how your design sculpts smiles on users’ faces.
Also in Design
Science & Tech
Pokémon Go is a huge AI-data warehouse
Millions of Pokémon Go players walk down streets, alleys and roads to capture virtual Pokémons when every bit of their perception is being recorded and used as data.
Niantic, the company behind Pokémon Go, is tapping into player data to build an AI-powered navigation system.
This new system processes physical spaces using images captured through the game and its Scaniverse app, creating a “Large Geospatial Model” (LGM) akin to large language models (LLMs) but for the real world.
How It Works
The model uses data from over 10 million scanned locations, with players contributing about 1 million new scans weekly. These scans, from a pedestrian’s perspective, capture areas that street-view cameras can’t reach. The result? A digital map that helps pinpoint your location with incredible accuracy.
Potential Uses
Beyond games, this tech could enhance augmented reality (AR) products, robotics, autonomous systems, spatial planning, and even logistics. It's poised to make the physical world more navigable and interactive.
Are players all happy about it?
Many Pokémon Go players may not have realised their scans were feeding an AI system. While it’s likely covered by the game’s terms of service, some players aren’t too happy about being part of this massive data collection effort, as discussed in Reddit threads.
Also in Science & Tech
Founders
Inference time scaling > Pre-training scaling of AI models
Pre-training scaling: LLMs are scaled by scaling on their pre-training sessions. In this phase, AI models learn to identify, structure and supply information from large datasets. But this way, the stakes of a data shortage are pretty high.
Inference time scaling: Recently adopted in OpenAI’s o1 model, this method is all about scaling LLMs by increasing the interpretation time of the model. This enables it to procure the right knowledge using the right tools, reflect better on output and hence generate accurate responses.
Why is the latter better?
Simply put, the inference time scaling law suggests that rather than constantly feeding AI models bulks of data pre-usage (i.e pre-training) and reaching the point of data-extinction, founders can allow these models to “think” and process more, that is, increase the inference time.
Also in Founding
Product
Integration problems create annoyed users
If you sell softwares or any other tech product, it is not necessary that all of your customers are going to be tech savvy.
So, these unaware individuals would curse your product if it doesn’t go well with the rest of their tech pile. Here’s 3 ways you can deal with it-
↳ Work on integration tools and processes: Create APIs, plugins, and connectors that make it easy for users to link your product with their existing systems. A smooth integration process reduces frustration and keeps users happy.
↳ Get into win-win partnerships with other tech companies: Collaborate with other tech companies to ensure your product plays well with others. Joint solutions can make integration seamless and provide more value to customers.
↳ If you want a lesser-friction solution, provide roadmaps simplified to the understanding level of a non-tech-person.
Today’s AI Image
Location Data to be Used by Firms
Source - DALL-E
Quote of the Day
Humility
"The only true wisdom is in knowing you know nothing."
Socrates
What we’re working on
 Velvet Onion & Friends We’re in the process of rebranding Velvet Onion & Friends. Why? It’s an important stage in our evolution, and deepens the link between agency, product & education. |  We’re at the final stages of planning for our pilot program. Working name is “99 Problems But A Pitch Ain’t One;” cute for internal projects, not sure it’s the name. Coming soon! |
