High-Quality Info Sources

Build a small, high-signal information radar instead of a giant attention landfill.

Core workflow

1. 1. Clarify the monitoring goal.
2. Group sources by role, not by popularity.
3. Prefer primary sources over commentary.
4. Keep the default list small.
5. Add a review rule so the list stays useful.

Clarify the monitoring goal

Start by identifying what the user actually wants to track:

• - breaking product/model releases
• research progress
• developer ecosystem changes
• market/industry moves
• critical or skeptical takes
• company-specific monitoring

If the user does not specify, assume they want a balanced monitoring set with:

• - official release channels
• technical interpreters
• industry operators
• critics / risk voices

Group by source role

Do not return a flat pile of links unless explicitly requested. Organize sources into roles such as:

• - Official / primary — company accounts, labs, docs, release blogs
• Builders / operators — founders, engineers, product leads
• Explainers — people who interpret developments clearly
• Critics / risk voices — people who stress test hype and assumptions
• Aggregators — useful only if they add speed or coverage without too much noise

Default ordering:

1. 1. official / primary
2. builders / operators
3. explainers
4. critics / risk voices
5. aggregators

Quality filter

Prefer sources that satisfy most of these:

• - close to the event
• high signal-to-noise ratio
• technically or operationally informed
• consistent over time
• not purely engagement bait
• useful for decisions, not just amusement

Penalize sources that are:

• - mostly reposting others
• chronically sensational
• vague and uncheckable
• redundant with better primary sources

Output patterns

Choose one of these depending on the request.

1. Small radar list

Use for users who want the minimum viable watchlist.

Format:

• - category
• source name / handle
• why it matters
• what to watch for

Aim for 8-15 sources.

2. Extended source map

Use when the user wants broad coverage.

Format:

• - grouped categories
• 3-8 entries per category
• short note on each entry
• note on which ones are must-watch vs optional

3. Monitoring system

Use when the user wants an operational workflow.

Include:

• - the core source list
• refresh cadence
• how to prune the list
• how to summarize findings into notes / Notion / docs

Maintenance rules

When building a reusable source system, include these rules:

• - keep a core list and an overflow list
• review monthly or when signal quality drops
• remove duplicates aggressively
• cap the default list so attention remains scarce and valuable
• promote only sources that repeatedly produce useful first-order information

AI-specific default lens

When the user asks for AI information sources and gives no stronger constraint, combine:

• - frontier labs
• open-source model players
• infrastructure / hardware players
• respected technical voices
• skeptical / governance voices

Read references/ai-sources.md for a starter set and selection logic.

Tone and judgment

Be opinionated. A source list is a filter, not a census.

Prefer:

• - “Follow these 10 first”
• “These 5 are optional”
• “This one is noisy but useful for early chatter”

Avoid pretending all sources are equally good.