Hype Assessment Skill

Assess which AI topics are overhyped, underhyped, or accurately assessed based on synthesized claims.

Assessment Framework

Overhyped Topics (Lab enthusiasm exceeds warranted confidence)

Signs of overhype:

• Lab researchers make strong claims that critics have substantively challenged
• Evidence quality is low but confidence is high
• Past predictions in this area have repeatedly failed
• Marketing language exceeds technical substance
• Hype delta > +0.3

Underhyped Topics (Critic skepticism may be excessive)

Signs of underhype:

• Real progress has been made but critics haven't updated
• Evidence is strong but narrative hasn't caught up
• Lab hints suggest unreleased capabilities
• Quiet progress without announcements
• Hype delta < -0.3

Accurately Assessed Topics

Signs of accurate assessment:

• Lab and critic views are relatively aligned
• Claims match observable evidence
• Predictions have been reasonably accurate
• Hype delta between -0.2 and +0.2

Scoring System

For each topic, assign a score from -1.0 to +1.0:

Evidence to Consider

For Overhyped Assessment

• Repeated failed predictions
• Marketing claims exceeding published results
• Hype cycle patterns (lots of announcements, few deliverables)
• Benchmark gaming without real-world transfer
• "Just around the corner" claims that keep slipping

For Underhyped Assessment

• Steady progress without fanfare
• Working deployments with limited publicity
• Academic results that haven't reached mainstream
• Capabilities that exist but aren't marketed
• Legitimate breakthroughs dismissed by critics

For Accurate Assessment

• Claims that held up over time
• Convergence between lab and critic views
• Predictions that came true
• Honest acknowledgment of limitations
• Nuanced discussion of tradeoffs

Output Format

Return JSON:

{
  "overhypedTopics": [
    {
      "topic": "agents",
      "score": 0.6,
      "reasoning": "Lab enthusiasm for autonomous agents significantly exceeds demonstrated reliability. Multiple high-profile failures in production while claims of imminent AGI-like autonomy persist.",
      "keyEvidence": [
        "Devin and similar demos failed to replicate",
        "Production agent deployments have high failure rates",
        "Claims of 'replacing developers' haven't materialized"
      ]
    }
  ],
  "underhypedTopics": [
    {
      "topic": "interpretability",
      "score": -0.5,
      "reasoning": "Significant progress on mechanistic interpretability is being made at Anthropic and elsewhere, but mainstream coverage focuses on capabilities. Real tools for understanding models are emerging.",
      "keyEvidence": [
        "Golden Gate Claude demonstrated genuine steering",
        "Feature extraction becoming reproducible",
        "SAEs showing practical utility"
      ]
    }
  ],
  "accuratelyAssessedTopics": [
    {
      "topic": "multimodal",
      "score": 0.1,
      "reasoning": "Vision-language models have improved substantially and assessments largely reflect actual capabilities. Both enthusiasm and concerns are grounded.",
      "keyEvidence": [
        "GPT-4V and Claude vision work as advertised",
        "Known limitations acknowledged",
        "Incremental improvements match expectations"
      ]
    }
  ],
  "overallFieldSentiment": 0.72,
  "summary": "A paragraph summarizing the overall hype landscape..."
}

Overall Field Sentiment

Calculate as weighted average of lab researcher bullishness across all topics (0.0-1.0).

Interpretation:

• 0.8-1.0: Extremely bullish field sentiment (potential bubble)
• 0.6-0.8: Optimistic but measured
• 0.4-0.6: Balanced/uncertain
• 0.2-0.4: Cautious/skeptical
• 0.0-0.2: Pessimistic

Summary Guidelines

Write a single paragraph summarizing:

1. Overall hype temperature
2. Most overhyped area and why
3. Most underhyped area and why
4. What sophisticated observers should pay attention to

Tone: Direct, opinionated but fair, grounded in evidence.