PulseAugur
EN
LIVE 14:11:31

Local Gemma 4 models show surprising knowledge of niche JAWS shortcuts

The user is experimenting with local AI models, specifically Gemma 4 variants like Gemma 4:12b and Gemma 4:e4b, to understand their capabilities in providing information about JAWS screen reader shortcuts. While the models can recall some specific key combinations for opening link lists (e.g., Insert + F7), their accuracy and usefulness vary, especially when asked about more nuanced navigation or scripting. The user notes that the models' knowledge base for such esoteric topics is surprisingly robust, comparable to niche gaming communities, despite the smaller user base of JAWS. AI

IMPACT Local AI models demonstrate surprising recall for niche technical information, suggesting potential for specialized assistance tools.

RANK_REASON User experimenting with existing models for specific use cases.

Read on Mastodon — mastodon.social →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Local Gemma 4 models show surprising knowledge of niche JAWS shortcuts

COVERAGE [1]

  1. Mastodon — mastodon.social TIER_1 English(EN) · [email protected] ·

    So, playing with local AI, specifically, Gemma4 (it-qat) models, Gemma4:12b can tell me that, to open a list of links with JAWS, I press Insert + F7. Gemma4:e4b

    So, playing with local AI, specifically, Gemma4 (it-qat) models, Gemma4:12b can tell me that, to open a list of links with JAWS, I press Insert + F7. Gemma4:e4b says Insert + L. I can't test Gemma4:27b-it-qat right now, but 31B gives the right answer, Insert + F7, but then drops …