Llama-modchip: 284B DeepSeek on 64GB RAM, no GPU
Llama-modchip aims to run a 284-billion-parameter DeepSeek model using 64GB of RAM without a GPU. The provided information does not explain its technical approach. (github.com)
A longtime Claude Code user is switching to Pi after frustrations accumulated with Anthropic’s coding assistant. The author values reviewing generated code and using IDE selections as context, but Claude Code’s VS Code integration supports only one connected session at a time, making parallel tasks awkward. Its separate GUI extension allows multiple tabs but has other shortcomings: some slash commands fail, there is no custom status line, and concurrent subagent responses clutter the main conversation. The author also objects to frequent feature additions that seem aimed at enterprise usage and higher token consumption, rather than their own needs, while changes to prompts and tool descriptions repeatedly disrupt personal workflows.
The broader complaint is that Claude Code lacks a coherent product direction and prioritizes new capabilities over stability and usability. In 2026, the author counted 244 releases, more than three-quarters of which changed the system prompt or tool descriptions; even after reducing the prompt substantially, the default Bash tool still used nearly 3,000 tokens on guidance the author considers unnecessary. Pi, by contrast, offers a more stable, minimalist, customizable setup. After a few days of adjustment, the author found it smooth to use and built extensions for personal needs - including support for multiple open sessions - in about 20 minutes. They acknowledge Claude Code’s pioneering role and the difficulty of developing a new kind of software, but argue that open-source alternatives such as Pi offer a practical escape for users who share these frustrations.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
Llama-modchip aims to run a 284-billion-parameter DeepSeek model using 64GB of RAM without a GPU. The provided information does not explain its technical approach. (github.com)
A GNU maintainer’s controversial take prompted debate about AI-generated code in open-source projects. Commenters discussed project leaders who use AI tools while distancing themselves from code ownership. (social.treehouse.systems)
ThinkStream says AI can generate work, but people still need to understand it. The provided page content contains only an anti-bot prompt. (thinkstream.app)
Ensemble is a book and open-source project for building an AI coding agent that helps create the next edition. It advocates development-driven specifications, where specs evolve alongside implementation instead of dictating development upfront. (github.com)
Maple uses manifold-activated pruning to compress mixture-of-experts language models below 2 bits per weight. Its developers say it runs Gemma 4 26B at 45 tokens per second on a 6GB GPU. (github.com)
Anthropic says AI has already helped develop models with capability gains the company did not initially recognize, and it may soon reach its definition of recursive self-improvement. The company warns that faster AI-driven development could create safety risks and says its Responsible Scaling Policy includes plans to monitor and mitigate them. (techcrunch.com)
Today's best Hacker News stories, summarized and screenshotted, one email a day.