PLUS: Anthropic's warning for coders and a new psychology-inspired neural network

It's a new day AI Rockstars!

Perplexity is moving beyond search with Comet, its new AI-native browser built on Chromium. The browser integrates its answer engine directly into the user experience, aiming to change how we interact with the web.

It's a move toward a browser that acts more like an interactive assistant than a simple content viewer, with early agentic features that can perform tasks across sites. The question is, how far are we from the browser becoming a true autonomous partner in our daily workflow?

In today’s Lean AI Native recap:

• Perplexity's launch of its AI-native browser, Comet
• A new AI architecture inspired by human psychology
• How asymmetric similarity can improve model performance
• Why model-picker dropdowns hurt the user experience

Perplexity's Comet Browser

The Report: AI search engine Perplexity has launched Comet, an AI-native browser built on Chromium that integrates its answer engine into the core browsing experience to change how you interact with the internet.

Broaden your horizons:

• Comet transforms the browser into an interactive assistant, letting you get AI-powered summaries directly from the address bar, summarize articles or YouTube videos, and ask follow-up questions with the full context of the page you're on.
• It features early agentic capabilities that can perform tasks for you across different websites, like analyzing a list on one page and automatically adding those items to a shopping cart on another.
• The browser still has limitations, as its AI agent can be inconsistent, and its search function sometimes struggles to surface the most recent results, prioritizing explanatory content instead.

If you remember one thing: Comet offers a compelling glimpse into the future of web browsing, where AI is deeply integrated rather than a separate tool. This shift points toward a new paradigm where your browser actively assists and automates tasks for you.

Psychology-Inspired AI

The Report: Researchers, including AI leaders Christopher Manning and Dan Jurafsky, have introduced a new AI architecture inspired by human psychology. Tversky Neural Networks replace standard similarity calculations with a more human-like, asymmetric model that boosts performance and interpretability.

Broaden your horizons:

• The new model addresses a core flaw in deep learning: Human similarity is asymmetric (a son resembles a father more than the reverse), while standard AI models use symmetric math that can't capture this nuance.
• Instead of just using vector math, the Tversky framework allows models to weigh the importance of common features against distinctive features, much like people do when comparing two things.
• Early results are promising, as replacing linear layers with Tversky layers improved image recognition accuracy and led to a language model with reduced perplexity and 35% fewer parameters.

If you remember one thing: This approach moves beyond black-box models by building AI with an inherently understandable reasoning process based on shared and unique features. It shows how concepts from cognitive psychology can lead to more effective and transparent AI systems.

Curing the Model Picker Disease

The Report: A new analysis argues that the model-picker dropdown in most AI apps is a strategic mistake that complicates the user experience and creates unsustainable pricing models.

Broaden your horizons:

• The argument uses Spotify as an analogy: users care about listening to artists, not the record labels behind them. Likewise, AI app users want a task completed, not to choose the underlying model that does the work.
• Removing the model picker allows companies to turn a pricing problem into a product challenge, using routers to send queries to the most cost-effective model (from powerful to smaller, open-source options) for any given task.
• This approach enables companies to move beyond being simple API wrappers and build a more defensible product by focusing on the overall user experience, smart routing, and unique non-AI features.

If you remember one thing: Abstracting away the underlying model allows founders to focus on solving the user's actual problem. This shift makes the product experience simpler and the business model far more sustainable.

Curing the Model Picker Disease

The Report: A new analysis argues that the model-picker dropdown in most AI apps is a strategic mistake that complicates the user experience and creates unsustainable pricing models.

Broaden your horizons:

• The argument uses Spotify as an analogy: users care about listening to artists, not the record labels behind them. Likewise, AI app users want a task completed, not to choose the underlying model that does the work.
• Removing the model picker allows companies to turn a pricing problem into a product challenge, using routers to send queries to the most cost-effective model (from powerful to smaller, open-source options) for any given task.
• This approach enables companies to move beyond being simple API wrappers and build a more defensible product by focusing on the overall user experience, smart routing, and unique non-AI features.

If you remember one thing: Abstracting away the underlying model allows founders to focus on solving the user's actual problem. This shift makes the product experience simpler and the business model far more sustainable.

Curing the Model Picker Disease

The Report: A new analysis argues that the model-picker dropdown in most AI apps is a strategic mistake that complicates the user experience and creates unsustainable pricing models.

Broaden your horizons:

• The argument uses Spotify as an analogy: users care about listening to artists, not the record labels behind them. Likewise, AI app users want a task completed, not to choose the underlying model that does the work.
• Removing the model picker allows companies to turn a pricing problem into a product challenge, using routers to send queries to the most cost-effective model (from powerful to smaller, open-source options) for any given task.
• This approach enables companies to move beyond being simple API wrappers and build a more defensible product by focusing on the overall user experience, smart routing, and unique non-AI features.

If you remember one thing: Abstracting away the underlying model allows founders to focus on solving the user's actual problem. This shift makes the product experience simpler and the business model far more sustainable.

Curing the Model Picker Disease

The Report: A new analysis argues that the model-picker dropdown in most AI apps is a strategic mistake that complicates the user experience and creates unsustainable pricing models.

Broaden your horizons:

• The argument uses Spotify as an analogy: users care about listening to artists, not the record labels behind them. Likewise, AI app users want a task completed, not to choose the underlying model that does the work.
• Removing the model picker allows companies to turn a pricing problem into a product challenge, using routers to send queries to the most cost-effective model (from powerful to smaller, open-source options) for any given task.
• This approach enables companies to move beyond being simple API wrappers and build a more defensible product by focusing on the overall user experience, smart routing, and unique non-AI features.

If you remember one thing: Abstracting away the underlying model allows founders to focus on solving the user's actual problem. This shift makes the product experience simpler and the business model far more sustainable.

Curing the Model Picker Disease

The Report: A new analysis argues that the model-picker dropdown in most AI apps is a strategic mistake that complicates the user experience and creates unsustainable pricing models.

Broaden your horizons:

• The argument uses Spotify as an analogy: users care about listening to artists, not the record labels behind them. Likewise, AI app users want a task completed, not to choose the underlying model that does the work.
• Removing the model picker allows companies to turn a pricing problem into a product challenge, using routers to send queries to the most cost-effective model (from powerful to smaller, open-source options) for any given task.
• This approach enables companies to move beyond being simple API wrappers and build a more defensible product by focusing on the overall user experience, smart routing, and unique non-AI features.

If you remember one thing: Abstracting away the underlying model allows founders to focus on solving the user's actual problem. This shift makes the product experience simpler and the business model far more sustainable.

The Shortlist

Amodei predicts that AI will write 90% of software code within 3-6 months, pushing developers to focus on high-level design and system oversight rather than line-by-line coding.

Otter.ai faces a class-action lawsuit alleging its Notetaker feature secretly records virtual meetings without consent from all participants, a violation of California's wiretapping laws.

OpenAI’s upgraded advanced voice mode for ChatGPT-5 offers more natural, human-like conversations, though hands-on testing reveals it's better for brainstorming than for sourcing detailed, factual information.

Graduates pivot toward skilled trades as concerns grow over AI automating white-collar office work, highlighting a trend where manual-labor jobs are seen as offering greater long-term job security.