How Viqus works

Every few minutes Viqus checks the RSS feeds of these 14 publications and keeps stories published in the last 48 hours:

• Hugging Face Blog
• AIModels.fyi
• The Gradient
• Mistral AI
• OpenAI News
• DeepMind
• AI – SiliconANGLE
• Machine Learning Mastery
• InfoQ AI
• Simon Willison
• The Verge AI
• TechCrunch AI
• MIT Technology Review AI
• Ars Technica AI

Stories that cover the same event as one already published in the last 96 hours are skipped, so you get one analysis per event rather than one per outlet.

The full text of each story is analyzed by an open-weight language model (gemma4:e4b) running on Viqus' own server; articles are not sent to third-party AI services. The model produces the headline, the summaries, the key points, the "why it matters" section, the category, tags and the Viqus verdict. The output is checked automatically for structure before publishing, but it is not reviewed by a human.

Every analysis page names and links the original source. Read it before relying on any detail: the model can misread, omit context or state something with more certainty than the source does.

Two independent 1–10 scores separate attention from consequence:

• Real impact: likely consequences over the next 12 months or more. 1–2 negligible, 3–4 affects a narrow group, 5–6 influences a sector, 7–8 changes competitive dynamics or regulation, 9–10 a before/after moment for the industry.
• Media hype: how much attention the story is getting. 1–2 barely covered, 5–6 major tech outlets, 9–10 dominating the news cycle beyond tech.

The model is instructed to be critical: most news should land between 4 and 6. A high hype score with a low impact score is the most useful signal Viqus gives you — lots of noise, little change.

Scores published before October 2026 were produced with an earlier version of the instructions and run noticeably higher; compare them with each other rather than with recent ones.

Every Monday the weekly briefing collects the 8 highest-impact stories of the previous week and an AI-written analysis connecting them. It is written from the analyses above, not from new reporting.

All analyses remain available. Only those from the last 90 days, and older ones with an impact of 9 or more, are offered to search engines; the rest stay accessible as an archive.