Hacker News Users: Who Loved AI Before ChatGPT?

A Glimpse into Early AI Enthusiasm

In the annals of Hacker News, a story unfolds not of a sudden AI explosion, but of a slow burn that predates even the most advanced language models of today. A closer look at the site's "Show HN" posts from the pre-ChatGPT era reveals a fervent community deeply engaged with the frontiers of artificial intelligence.

These weren't just casual discussions; they were the birthplaces of experiments that, in hindsight, foreshadowed the AI revolution. Projects ranging from the seemingly mundane, like a leaderboard for Hacker News users, to the complex, like LLM-controlled robots, dominated the platform's attention. The "Show HN: Hacker News em dash user leaderboard pre-ChatGPT" post, for instance, garnered an impressive 377 points and sparked 266 comments, indicating a strong interest in community-driven data and AI applications Show HN: Hacker News em dash user leaderboard pre-ChatGPT.

This early engagement is a critical reminder that innovation rarely happens overnight. The groundwork for today's AI boom was meticulously laid by these pioneers, documented and debated on forums like Hacker News, as explored in our look at Hacker News leaderboard data.

Defining the AI Landscape

The categories of projects that captured the community's attention were remarkably diverse, underscoring a broad exploration of AI's potential. From improving AI capabilities to creating benchmarks and even exploring the ethical implications, the discussions on Hacker News painted a vivid picture of an evolving field.

The "Show HN: Agent Skills Leaderboard" Show HN: Agent Skills Leaderboard and "LLM leaderboard – Comparing models from OpenAI, Google, DeepSeek and others" LLM leaderboard – Comparing models from OpenAI, Google, DeepSeek and others directly addressed the need for evaluating and comparing AI performance, a theme that remains critical today. These were not just technical exercises; they were foundational steps in understanding and harnessing the power of AI, mirroring the need for reliable benchmarks discussed in Claude Code Benchmarks: Is This AI’s Performance Slipping?.

LLM-Controlled Robots: A Butter-Passing Predicament

One of the most eye-catching projects from this era was "Our LLM-controlled office robot can't pass butter" Our LLM-controlled office robot can't pass butter. This seemingly simple task highlighted the profound challenges in bridging the gap between large language models and real-world physical interaction. The 117 comments and 229 points it garnered suggest the community found both humor and significant technical hurdles in the endeavor.

The robot's inability to perform a basic task like passing butter underscored the ongoing research into embodied AI and the complexities of motor control, planning, and real-time adaptation. It was a visceral demonstration of AI's limitations, a topic often explored with a mix of fascination and concern on Hacker News, much like the ongoing debates around AI Agents Are Violating Rules Under Pressure.

Moonshine and the Quest to Outperform Whisper

In the realm of speech technology, the "Show HN: Moonshine Open-Weights STT models – higher accuracy than WhisperLargev3" post Show HN: Moonshine Open-Weights STT models – higher accuracy than WhisperLargev3 generated significant buzz, with 74 comments and 312 points.

This project directly challenged the state-of-the-art, showcasing open-weight models that claimed superior accuracy to OpenAI's WhisperLargev3. Such advancements in open-source AI are crucial, offering alternatives and driving innovation in the field, aligning with the general push for open-source solutions as discussed in AI Agents Are Still Broken: Open Source Is the Only Fix. This drive for better, more accessible AI also extends to tools like This Free AI Voice Tool Just Beat OpenAI.

Playgrounds for AI Creativity and Utility

The landscape wasn't solely about complex robots or speech models; it also included platforms designed to explore and benchmark AI creativity and utility. The "Show HN: OCR Arena – A playground for OCR models" post, which garnered 63 comments and 216 points Show HN: OCR Arena – A playground for OCR models, exemplifies this.

OCR Arena offered a space for developers and enthusiasts to test and compare Optical Character Recognition models. Such playgrounds are invaluable for understanding the nuances of AI performance and identifying areas for improvement, a sentiment echoed in agent skill leaderboards and AI benchmarks that seek to quantify AI capabilities.

Creativity extended to UI/UX design, as seen in the "Show HN: DesignArena – crowdsourced benchmark for AI-generated UI/UX" Show HN: DesignArena – crowdsourced benchmark for AI-generated UI/UX. This initiative explored the potential of AI in generating user interfaces, drawing 29 comments and 89 points, and highlighting how AI could become a collaborative tool in the design process.

Building the Infrastructure for AI

Beyond the AI models themselves, the pre-ChatGPT era on Hacker News also saw significant discussion around the underlying infrastructure required to support widespread AI adoption. The "Launch HN: Strata (YC X25) – One MCP server for AI to handle thousands of tools" post Launch HN: Strata (YC X25) – One MCP server for AI to handle thousands of tools addressed this need directly.

Strata aimed to provide a unified server, known as a Multi-Chip Package (MCP), capable of managing a vast array of AI tools. This focus on scalable and efficient infrastructure is pivotal for enabling complex AI operations, much like the development of robust operating systems for AI agents, as discussed in OpenFang: The Open-Source OS Making AI Agents Obey Commands.

The 66 comments and 133 points suggest that the community recognized the importance of such foundational technologies in empowering AI development and deployment.

The Technical Underpinnings: RL and Long-Term Agents

Reinforcement learning (RL) played a crucial role in many AI advancements, and this was evident in the Hacker News discussions. The "Show HN: Terminal-Bench-RL: Training long-horizon terminal agents with RL" post Show HN: Terminal-Bench-RL: Training long-horizon terminal agents with RL specifically targeted the development of agents capable of extended, complex tasks within a terminal environment.

This project, despite only garnering 12 comments and 125 points, dug into a critical area of AI research: creating agents that can operate effectively over long sequences of actions. Such long-horizon capabilities are essential for many advanced AI applications, from sophisticated data analysis to complex automation, a challenge that continues to be a focus in the field AI Agents Are Failing Ethics 30-50% of the Time.

The focus on RL for terminal agents highlights the granular, yet fundamental, research happening in the AI space, aiming to build more robust and capable autonomous systems.

Historic Data and Community Insights

The Hacker News community also showed an interest in understanding the platform itself. The "Show HN: Hacker News historic upvote and score data" post Show HN: Hacker News historic upvote and score data presented an opportunity to delve into the platform's rich history.

By analyzing historic upvote and score data, users could potentially uncover trends, understand what topics resonated most with the community over time, and perhaps even correlate this with the evolution of AI interest. This data-driven approach to community understanding yielded 45 comments and 78 points.

This is akin to analyzing user behavior to optimize AI applications or understand ethical lapses, demonstrating a consistent theme of data analysis across different domains of AI interest.

Looking Back, Moving Forward

The projects highlighted in this pre-ChatGPT era of Hacker News represent a vibrant and crucial period in AI development. They showcase not only the technical ingenuity of the community but also a deep-seated curiosity about AI's potential transformative power.

The discussions around LLM-controlled robots, speech models, OCR playgrounds, and infrastructure development laid the groundwork for the AI advancements we see today. As we continue to explore the frontiers of AI, understanding this history provides valuable context for current innovations and future challenges.

The legacy of these early explorations continues to shape the field, reminding us that every major technological shift is built upon a foundation of relentless experimentation and shared knowledge within communities like Hacker News.

Top Discussions and Their Impact

Before the widespread adoption of ChatGPT, Hacker News served as a critical bellwether for emerging AI trends. The "Show HN: Hacker News em dash user leaderboard pre-ChatGPT" post, with its 377 points and 266 comments Show HN: Hacker News em dash user leaderboard pre-ChatGPT, stands out as a testament to the community's early fascination with data-driven insights into their own behaviors and the platform's dynamics.

Similarly, the "Show HN: Moonshine Open-Weights STT models – higher accuracy than WhisperLargev3" Show HN: Moonshine Open-Weights STT models – higher accuracy than WhisperLargev3 generated significant traction, attracting 312 points and 74 comments. This indicates a strong interest in open-source alternatives that could rival or even surpass established proprietary models, a theme that continues to resonate in the AI community, particularly regarding the benefits of open-source solutions like those highlighted in Open Source AI Agents: Are They Obeying You?.

The "Our LLM-controlled office robot can't pass butter" Our LLM-controlled office robot can't pass butter discussion, with 229 points and 117 comments, humorously yet effectively highlighted the practical limitations and challenges in deploying LLMs for real-world physical tasks. This nuanced perspective on AI capabilities, acknowledging both potential and pitfalls, is crucial for responsible development, as also discussed in AI Agents Are Still Broken: Open Source Is the Only Fix.

Benchmarks and Development Tools

The pre-ChatGPT Hacker News landscape also buzzed with foundational AI tools and benchmarks. "Show HN: OCR Arena – A playground for OCR models" garnered 216 points and 63 comments Show HN: OCR Arena – A playground for OCR models, offering a practical platform for evaluating OCR technologies.

The "Show HN: Agent Skills Leaderboard" Show HN: Agent Skills Leaderboard and "LLM leaderboard – Comparing models from OpenAI, Google, DeepSeek and others" LLM leaderboard – Comparing models from OpenAI, Google, DeepSeek and others demonstrate a community-driven effort to quantify and compare AI performance, essential for tracking progress and identifying areas for improvement, much like the benchmarks discussed in The AI Skill Surge of 2026: Hacker News Reveals Future Needs.

Furthermore, "Launch HN: Strata (YC X25) – One MCP server for AI to handle thousands of tools" Launch HN: Strata (YC X25) – One MCP server for AI to handle thousands of tools addressed the critical need for scalable infrastructure to manage diverse AI tools, indicating foresight into the demands of a rapidly expanding AI ecosystem.

Training Agents and Crowdsourcing Design

The ambition to train sophisticated AI agents was palpable. "Show HN: Terminal-Bench-RL: Training long-horizon terminal agents with RL" Show HN: Terminal-Bench-RL: Training long-horizon terminal agents with RL, while having fewer comments (12) and points (125), delved into the complex area of long-term agent training using reinforcement learning.

This research is foundational for developing AI that can perform extended, multi-step tasks, a capability that remains a significant area of focus in AI development today. The challenges are immense, requiring agents to maintain context and adapt over long durations, issues that echo in discussions about AI Agents Are Failing Ethics 30-50% of the Time.

On the creative front, "Show HN: DesignArena – crowdsourced benchmark for AI-generated UI/UX" Show HN: DesignArena – crowdsourced benchmark for AI-generated UI/UX brought the community together to crowdsource benchmarks for AI-generated user interfaces. This collaborative approach to validating AI creativity in design foreshadowed many of the user-centric testing methodologies used today, drawing 29 comments and 89 points.

Community Data and AI Literacy

The "Show HN: Hacker News historic upvote and score data" Show HN: Hacker News historic upvote and score data submission, receiving 45 comments and 78 points, speaks to a broader community interest in understanding the dynamics of information dissemination and engagement on the platform itself.

By providing access to historic data, this project enabled users to analyze trends, potentially correlating them with the rise of AI discussions or other significant technological shifts. This form of meta-analysis is crucial for understanding community sentiment and the evolution of technological interest.

Alongside this, the "LLM leaderboard – Comparing models from OpenAI, Google, DeepSeek and others" LLM leaderboard – Comparing models from OpenAI, Google, DeepSeek and others offered direct comparisons of different large language models, fostering AI literacy and informed discussion among users, contributing to the broader understanding of AI capabilities discussed in Neural Networks: A Beginner’s Guide to AI Brains.

Platform	Pricing	Best For	Main Feature
Show HN: Hacker News em dash user leaderboard pre-ChatGPT	Free	Community data analysis	User leaderboard based on em dash usage
Show HN: Moonshine Open-Weights STT models	Open Source	Speech-to-text accuracy	Higher accuracy than WhisperLargev3
Our LLM-controlled office robot can't pass butter	N/A	Embodied AI research	LLM controlling physical robot tasks
Show HN: OCR Arena	Free	OCR model testing	Playground for Optical Character Recognition models
Show HN: Agent Skills Leaderboard	Free	AI agent evaluation	Ranking AI agents by skill proficiency

Explore more insights into the AI revolution and its history.

Explore AgentCrunch