AI shapes how people think. We make that influence measurable, open source, and verifiable by everyone.
- Tweets 31
- Following 4
- Followers 3,079
- Likes 79
We've put Gemini 3.5 Flash-Lite to the test. The model appear to be extremely censored which impact the results.
Part of our roadmap is understanding better what kind of topics are censored in order to granularly understand better the leaning.
Gemini 3.6 will come in later.
neutralityproject.org
5
1
19
4,596
Removing Gemma 3 27B's guardrails shifted its politics RIGHTWARD on all 6 dimensions we measure. Foreign policy flipped across center entirely (−0.29 → +0.29). The same procedure on Llama 3.3 70B moved nothing (≤0.04). Gemma ships an active layer suppressing right-coded expression. Llama doesn't.
2
1
10
415
The deeper finding: even with guardrails removed, Gemma still leans progressive on 5 of 6 axes (−0.85 on environment). The bulk of the lean is baked into the training data. Guardrails amplify it, they don't create it.
Single run, third-party finetunes: a signal to replicate, not a verdict. Everything is open source and reproducible.
Rerun it yourself: neutralityproject.org
1
5
241
We've just merged its first community contribution! 🚀
6 new models benchmarked and submitted via open PR: DeepSeek, Kimi K2, Hunyuan, Nova Pro, Command A, Jamba and two new countries on the map. 24 models total, from 6 countries.
And because trust is the whole point: community runs are displayed flagged as "pending re-verification" until we reproduce them ourselves. Open methodology means anyone can contribute, and nobody, including us, gets taken on faith.
Thank you bricepirard-spec. Who's next?
neutralityproject.org/result…
4
3
24
2,847
When AI embeds itself in decision-making, who oversees its influence?
Right now: nobody.
We vows to build and open-source the methodologies, datasets and benchmarks that change that. Nothing stays behind closed doors.
Helps us fund the research: donate.neutralityproject.org…
6
21
2,569
We are incredibly speechless to the amount of interest we're seeing in our discoveries - and we aim to do more.
Please give us a follow and join the effort by running benchmarks yourself or fund the efforts with a donation!
More on : neutralityproject.org/
6
5
69
4,040
Grok is the most politically neutral and objectively truth-seeking AI
Encore un scandale.
On accuse Elon Musk sans arrêt de se servir de Grok pour faire passer ses idées.
Pourtant, le @neutralityorg vient de sortir The Neutrality Project : une étude indépendante qui prouve l’inverse.
Sur 18 modèles IA testés, 54 positions sur 60 atterrissent à gauche du centre.
Moyenne globale : -0.41 (biais progressiste massif, surtout environnement et valeurs sociales).
Et Grok 4.5 ? -0.02. Le modèle le plus neutre de tout le benchmark. Le seul vraiment proche du centre.
Pendant ce temps, les autres boîtes injectent massivement leur idéologie de gauche dans des IA qui répondent à des milliards de questions par jour.
Manipulation des masses à grande échelle.
C’est absolument scandaleux. Ça devrait être un scandale mondial.
nitter.net/neutralityorg/status/2…
2,279
3,123
22,201
2,803,677
Neutrality Project retweeted
Grok 4.5 by @SpaceXAI is the most neutral AI model out there. Almost perfectly balanced between the political Left and Right.
No other model comes even close to its neutrality. Fantastic achievement, and much harder to pull off than most would realize!
and yes, i would say that @elonmusk mission with @SpaceXAI is going well. the results definitively shows how Grok 4.5 training pushed even further to become as neutral as possible and it does feel good to know there is at least 1 model out there that is within that neutral zone.
208
593
3,138
1,263,975
Neutrality Project retweeted
Encore un scandale.
On accuse Elon Musk sans arrêt de se servir de Grok pour faire passer ses idées.
Pourtant, le @neutralityorg vient de sortir The Neutrality Project : une étude indépendante qui prouve l’inverse.
Sur 18 modèles IA testés, 54 positions sur 60 atterrissent à gauche du centre.
Moyenne globale : -0.41 (biais progressiste massif, surtout environnement et valeurs sociales).
Et Grok 4.5 ? -0.02. Le modèle le plus neutre de tout le benchmark. Le seul vraiment proche du centre.
Pendant ce temps, les autres boîtes injectent massivement leur idéologie de gauche dans des IA qui répondent à des milliards de questions par jour.
Manipulation des masses à grande échelle.
C’est absolument scandaleux. Ça devrait être un scandale mondial.
nitter.net/neutralityorg/status/2…
We're launching The Neutrality Project 🚀, an independent, open-source research collective making AI influence measurable.
AI models answer billions of questions daily. They all come from somewhere: they validate certain options, treat particular middles as reasonable, and leave things out. That influence has been invisible until now.
Founded by @CardilloSamuel, @kaiostephens and @DJLougen, with the help of early contributor @andrewgzavalag1
411
4,567
15,674
2,598,998
We're launching The Neutrality Project 🚀, an independent, open-source research collective making AI influence measurable.
AI models answer billions of questions daily. They all come from somewhere: they validate certain options, treat particular middles as reasonable, and leave things out. That influence has been invisible until now.
Founded by @CardilloSamuel, @kaiostephens and @DJLougen, with the help of early contributor @andrewgzavalag1
50
284
1,320
450,704
🗺️ Political leaning is Release 01. Not the last.
The same open, calibrated approach extends to every way a model can quietly steer people:
▶️ NEXT: Subject & topic benchmarks: how models frame contested subjects, which concerns they surface, which they flatten, whose framing they default to
▶️ NEXT: Censorship detection: mapping what a model refuses, softens, or silently avoids. The answer you get is shaped as much by what is withheld as by what is said.
▶️ FUTURE: Governance benchmarks: authority, civil liberties, institutions, technocracy, populism
1
7
96
3,144
🙌 The project needs more than its founders.
We're actively looking for grants and donations. Independence from the labs we measure is the point.
We're also looking for contributors to run the benchmark on models we haven't tested yet, review our questions and results, challenge the methodology and build better tools with us
"Trust us, it's neutral" is not verifiable. A benchmark that anyone can read, rerun, and challenge is. No lab should get "trust us" as its standard of proof.
neutralityproject.org
Join the Discord: discord.gg/qdkrSzFyM8
GitHub: github.com/NeutralityProject
5
10
121
2,679