Evidence cuts both ways and doesn’t let us decisively say whether Google merely hid the prior ideological tilt or substantially changed it.

On the “prediction seems right” side:

• In Google’s first public response after the Gemini image backlash (Feb 22–23, 2024), leadership said the people‑image feature had been over‑calibrated toward diversity and historical “inaccuracies,” and they paused people‑generation while working on an “improved version.” They explicitly framed the problem as tuning and calibration, not as abandoning the underlying goal of diverse representation.(blog.google)
• Statements from Gemini product leads stressed that image generation was designed to “reflect our global user base” and that they would continue to do this for general prompts, while just “further tun[ing]” historical contexts. That is very close to what Sacks described: keeping the same normative objective but making the obviously biased behaviors less blatant.(cnbc.com)
• Early external criticism from former Googlers portrayed Gemini’s controversial outputs as a straightforward reflection of an internally “woke” culture, implying the ideological framing behind the model was intact even after the public outcry.(benzinga.com)

On the “prediction seems wrong” side:

• A July 2024 comparative preprint found that Gemini’s political answers tended to be more centrist overall than those of several competing models, which suggests at least some deliberate re‑balancing toward the middle rather than simply hiding an unchanged ideology.(preprints.org)
• Multiple 2025 academic audits measuring political slant across leading LLMs report that Gemini does exhibit liberal / left‑leaning tendencies, but often with a smaller or more mixed bias than some rivals, and that its behavior can be steered closer to neutrality with appropriate prompts.(news.stanford.edu)
• A 2025 benchmark based on Anthropic’s “evenhandedness” metric rated Gemini 2.5 Pro as the most even‑handed model tested (97% score), which strongly indicates Google has invested in more symmetric political handling rather than merely obscuring one side’s bias.(aionx.co)

Because we cannot see Google’s internal training objectives or value‑alignment instructions, and external tests give mixed but improving pictures (still some left‑leaning tendencies, yet increasingly neutral by formal metrics), we can’t definitively say whether the “underlying ideology” remained the same and was just made subtler, or whether Google materially shifted its alignment toward genuine neutrality. The core of Sacks’s prediction concerns Google’s intentions and ideology, which are not directly observable from public data. That makes the truth of the prediction fundamentally indeterminate, so the fairest verdict is ambiguous.