Gemini 2.5 Pro: An AI That Knows Its Limits

2025-03-26
Gemini 2.5 Pro: An AI That Knows Its Limits

The author attempted to get Gemini 2.5 Pro to recreate the famous 90s synthesizer, ReBirth RB-338. Surprisingly, instead of attempting the impossible, Gemini 2.5 Pro assessed the task's difficulty and explained its infeasibility, demonstrating powerful reasoning capabilities. The author negotiated a simpler, yet functional synthesizer. This showcases AI's progress towards understanding its limitations and making rational judgments.

Read more
AI

Two AI Programming Paradigms: Dialog vs. Autonomous

2025-01-27
Two AI Programming Paradigms: Dialog vs. Autonomous

Eleanor Berger shares her experience with two primary approaches to AI-assisted programming: dialog-based and fully autonomous. Dialog-based programming involves interacting with AI assistants for advice, code completion, and suggestions. Fully autonomous programming relies entirely on AI to generate code, with the programmer only reviewing the output. Berger finds both methods valuable but difficult to mix effectively, as understanding code written by others (or AI) is challenging and error-prone. She explores strategies for combining the approaches, such as per-project or per-module separation. She predicts a future shift towards fully autonomous programming due to improving AI capabilities and the inherent inefficiency of manual coding in comparison.

Read more
Development

Cheap AI Enables 'Stupid' Ideas: The Birth of Gongzilla

2025-01-25
Cheap AI Enables 'Stupid' Ideas: The Birth of Gongzilla

The author used ChatGPT's o1 and v0 functionalities to create a small game called Gongzilla in under an hour, without writing a single line of code, through multiple iterations. While the game itself isn't perfect, it showcases the possibilities of rapid prototyping and creative realization in the age of cheap AI. This post explores the ease of AI-assisted creation and the value of exploring 'stupid' ideas at low cost—even if those ideas ultimately aren't perfect, the fun of learning and creating is invaluable.

Read more

The AI Productivity Paradox: Why Aren't We Seeing Economic Growth?

2025-01-04
The AI Productivity Paradox: Why Aren't We Seeing Economic Growth?

Despite rapid advancements in AI, economic productivity hasn't seen a corresponding surge, contrasting with the impact of previous technological leaps like the internet. The article argues that productivity in the knowledge economy is difficult to measure, with humans tending towards 'satisficing' rather than maximizing output. AI and other technologies are primarily used as human augmentation tools, not productivity multipliers. The 'human-in-the-loop' model currently limits AI autonomy, but the future will see AI gain more independence and contextual understanding, potentially breaking the current productivity growth bottleneck.

Read more