ProgramAsWeights' 1M-View Demo Fails to Convert, Sparking Dev-Tool Marketing Debate

A public debate broke out over how research-oriented dev tools should be marketed, after ProgramAsWeights (PAW) author yuntiandeng revealed that his English<->Claudish demo reached 1M views with clear attribution yet converted almost no users, openly asking where his marketing went wrong.

Confirmed

• Developer willcb publicly criticized yuntiandeng for spamming replies under related tweets all week to promote PAW, calling it off-putting, and noted that even a million views brought no users.
• willcb argued the concept is too complex to grasp at first glance—how small the model is, how fast on a Mac, whether users must manage the Qwen backend, why GPT is needed—and only after significant effort did he conclude PAW is essentially LoRA search/compilation. He recommended saying so directly, since target users already know LoRA, suggesting the framing: PAW searches LoRAs for sub-1B models via synthetic examples.
• willcb opposed taglines like "just 3 lines of code" or "like Jev but local and open-source" that hide complexity, arguing researchers and developers actually want to understand it, and suggested learning from DSPy's narrative.
• willcb also questioned the open-source strategy: such projects should grow on GitHub before seeking mass market fit; jumping to a self-hosted community platform cuts off growth, and code that looks like a general library is really just a client for a hosted service.
• For promotion, willcb advised tailoring by audience: engaging content (threads, demos, videos, visuals) to attract users, and methods/training-recipe details for research impact. He also proposed treating tweets as paper abstracts and blogs as extended versions.
• yuntiandeng responded that PAW's core innovation separates understanding task specs from executing them: compile once, then a 0.6B model executes repeatedly on many inputs locally.
• yuntiandeng clarified PAW's architecture differs considerably from Jev—it outputs no confidence scores and doesn't guarantee outputs fall in predefined classes—so it should not be seen as an open-source local Jev, though many applications don't need scores anyway.

Unconfirmed

• Performance details raised by willcb (Mac speed, model size limits) have not been answered with concrete figures by the author.

Why it matters

The exchange illustrates a common dilemma for research-driven open-source tools: genuine technical innovation can be wasted by unclear positioning and mismatched channels—a 1M-view demo with near-zero conversion is a cautionary case for indie developers, and willcb's advice forms a reusable marketing checklist.

2026-09-21 ~ 2026-09-21 · 11 related posts

• [source] 1M-view demo, no users: PAW creator opens up on the build-vs-adopt marketing gap — yuntiandeng · 2026-09-21
• 'Spray-and-pray marketing reads as desperate': dev critique of 1M-view demo that converted no users — willcb · 2026-09-21
• PAW critique continued: burying the GitHub repo undermines pre-product-market-fit growth — willcb · 2026-09-21
• [source] PAW Author Explains Compile-Once/Run-Many Design: 0.6B Model Runs Locally on CPUs — yuntiandeng · 2026-09-21
• Dev Argues PAW Shouldn't Hide Its Complexity, Points to DSPy as the Better Playbook — willcb · 2026-09-21
• 'Just 3 lines of code' backfires: dev argues tools should expose complexity, not hide it — willcb · 2026-09-21
• [source] Dev urges PAW to just say it's a LoRA compiler for sub-1B models instead of hiding the mechanism — willcb · 2026-09-21
• PAW author clarifies: not open-source Jev, just 3 lines of code for triage tasks — yuntiandeng · 2026-09-21
• Dev advice for PAW launch: say 'LoRA search' outright, differentiate from Jev on capability — willcb · 2026-09-21
• Tweets are the new abstract: how to actually get people to read your paper — willcb · 2026-09-21
• How to promote a research tool: demos for builders, methods for researchers — willcb · 2026-09-21