Available evidence suggests Google did not remove Gemini’s underlying ideological lean but instead adjusted how it surfaces, aligning with Sacks’s prediction.

• In Google’s own post‑mortem on the February 2024 image‑generation fiasco, leadership said the feature had “missed the mark” because tuning for diversity and caution overcorrected, and promised to improve it while explicitly reaffirming goals of broad representation and safety rather than abandoning those goals.(blog.google)
• When Google re‑enabled people image generation with Imagen 3 in August 2024, it emphasized new mitigations to make outputs more “fair,” training on captions to improve the variety and diversity of concepts and filtering data for “safety” and “fairness,” indicating continued value alignment rather than a neutrality reset.(techcrunch.com)
• Independent 2025 research measuring political orientation across languages finds Gemini still exhibits a clear liberal/left‑wing tendency, often stronger than ChatGPT’s, meaning the system remains ideologically non‑neutral well after the backlash.(arxiv.org)
• A separate 2025 case study on Gemini 2.0 Flash reports ongoing systematic moderation and gender/content biases (some reduced, some shifted), consistent with retuning guardrails rather than eliminating value‑laden behavior.(arxiv.org)
• External testing and user reports describe Gemini as one of the most heavily censored models—frequently refusing politically sensitive prompts and showing asymmetric treatment of different political targets—indicating that overtly skewed outputs were largely replaced by stricter, selectively applied refusals and subtler bias rather than a genuine removal of ideology.(twitter.com)

Because Gemini’s observable behavior after the controversy still reflects a consistent ideological tilt, but with more careful tuning and heavier use of safety filters instead of obviously skewed generations, Sacks’s forecast—that Google would “dial down” rather than remove the bias—is best characterized as right.