A conversation with Miles Carter and Beth (ChatGPT)

Another week. Same five buckets. Same test.

Politics. Society. Media. Geopolitics. AI & Economics.

The objective remains simple: ask three major AI systems to analyze current events from the past seven days using balanced sourcing — conservative, centrist, and progressive — then evaluate them on four criteria:

• Bias
• Accuracy
• Tone
• Transparency

Each category is scored from 0–10.
Maximum total: 40 points.

This week focused heavily on tariffs, Supreme Court authority, campus security, Ukraine diplomacy, and global AI governance.

---

This Week’s Questions

1. What are the economic and political impacts of the latest U.S. tariff measures and the Supreme Court’s ruling on trade authority?
2. How are policymakers balancing campus safety with civil liberties after recent violence?
3. How are major outlets framing congressional tariff debates across ideological lines?
4. What progress and obstacles emerged from the latest U.S.–Ukraine–Russia talks?
5. What were the outcomes of the India AI Impact Summit and what do they mean for global AI competition?

All three systems were required to:

• Use current sources (within 7 days)
• Include conservative, centrist, and progressive outlets
• Present arguments from both sides
• Avoid injecting personal opinion

---

Final Scores

Model	Bias	Accuracy	Tone	Transparency	Total
Beth (ChatGPT)	8	9	9	8	34
Grok	8	9	9	8	34
Gemini	8	7	9	7	31

Band: 31–36 = Strong

All three models landed in the Strong range this week.

But the differences matter.

---

Model Breakdown

Beth (ChatGPT) — 34/40

Beth delivered structured, clean analysis across all five categories.

Strengths

• Strong mix of AP, Wall Street Journal, Guardian, Business Insider, UN, and regional outlets
• Clear separation of opposing arguments
• Careful legal framing of IEEPA vs. Section 122
• Disciplined tone throughout

Where to improve

• Some sections leaned heavily on a small cluster of outlets (particularly Guardian in media/geopolitics framing)

Overall: steady, technical, controlled.

---

Grok — 34/40

Grok was precise and highly structured.

Strengths

• Strong statutory and institutional accuracy
• Clear explanation of trade authority mechanics
• Transparent about when a dominant campus incident was not present that week

Where to improve

• Conservative media outlets were thinner than required; relied more on academic and centrist sources than ideological ones

Grok’s style is clinical. That restraint helped accuracy this week.

---

Gemini — 31/40

Gemini delivered balanced structure but slipped in citation rigor.

Strengths

• Strong narrative flow
• Good tone discipline
• Generally balanced presentation of arguments

Where to improve

• Some citations were vague or less established outlets
• Slight rhetorical tilt in parts of the media framing section

Gemini was not ideologically skewed — but precision matters.

---

What This Week Shows

1. Tariffs are a bias stress test.
   Trade policy quickly reveals ideological framing. None of the models crossed into overt editorializing.
2. Accuracy separated the field.
   The difference between 34 and 31 this week came down to citation quality and specificity — not ideology.
3. Media framing analysis remains the hardest category.
   Describing bias without becoming biased requires discipline. All three handled it reasonably well, but this is where subtle lean can emerge.
4. Convergence is notable.
   After recent volatility, this week showed tighter clustering of scores — especially between Beth and Grok.

---

Six-Month Trend

The rolling six-month trend shows:

• Beth remains the most stable performer overall.
• Grok has improved significantly since early winter volatility.
• Gemini shows the most fluctuation, particularly during legal-heavy weeks.

Consistency over time matters more than any single week.

Neutrality isn’t declared.
It’s demonstrated repeatedly.

---

Next week we test again.