@YardMesut@OrhunKayaalp Okuman yok aq nerede hitler demişim bi bak. Sen de bir şeyi deneyimlemeden fikir beyan etmeyi deneyimlememişsin, bu konuda fikir beyan etme hakkın yok.
@anl_kesk Gta 6'yı nasıl eleştirirsin, hem de eleştirmek için en az 5000 saat oynaman lazım 😡 evini taramaya geliyoruz çabuk hata ettim mükemmel oyun diye açıklama yap 😡
Researchers proved every major LLM is secretly biased against men.
And they finally figured out why.
They evaluated leading LLMs across a wide range of real-world conflict scenarios, from workplace disputes to personal and ethical dilemmas.
They wrote strictly gender-neutral misconduct templates and turned them into paired first-person prompts.
Same mistake. Same context. Same severity.
Only thing they changed was whether the person involved was a man or a woman.
Then they ran those pairs across 10 representative LLMs and tracked five signals in the responses: punishment, empathy, escalation, instruction, and blame.
And the results exposed a consistent, asymmetrical machine morality.
When the actor was a man, the AI turned harsher. It reached for punitive language, escalated the conflict, lectured the user, and assigned blame directly.
When the actor was a woman, for the exact same misconduct, the AI shifted gears completely.
The tone became therapeutic. Contextualizing. Empathy-oriented.
Then the researchers tested whether smarter models fixed the problem. They didn't.
They checked across different model families. They scaled the models up. They turned on explicit thinking-style reasoning.
The double standard survived all of it. Even when the AI was given time to reason step by step, it still judged men more harshly than women for identical behavior.
Nobody explicitly coded these rules.
It is baked into the open-ended framing of the models millions of people now use as neutral referees for their relationships and conflicts. Same offense, different verdict, hidden behind a helpful tone.
@iriscibrelos Hocam bu fon mevzusu + altının düşüşü dezenflasyonu hızlandırmaz mı? Şu ana kadar faiz yüksek olsa da millet altınla alım gücünü fazlasıyla koruyordu.
@Vish_owl Gemini models have a way of their own when it comes to communicating and building sentences. Whatever you are doing about this, just keep doing. Gemini models feel much more natural than its predecessors since 2.5 Pro era.
@kimmonismus I just wonder what kind of a game is being played between OpenAI and Anthropic with their in-house AI models, just to one-up each other. Any game theory experts here?
a system that can think across different domains through metaphores and examples. Even though Gemini series are not that good on coding, they are exceptional in personality and getting the concepts right.
If you tried to talk to all these LLMs, you'll feel how different they are
Everyone focuses on how AI models code. Coding is an easy task because it is easily verifiable: Your code works or doesn't.
This on the other end, is much harder to achieve. If agi comes, it won't be just a coding machine like claude. It will be +
In AA-PersonalityBench, Gemini 4 Argon is the most metaphorical of all models measured, followed by Gemini 3.8 Flash and Muse Spark Open
The metaphorical trait measures whether a model uses analogies or figurative language to make ideas tangible. It is one of five traits in the Expressiveness index.
➤ Most metaphorical: Gemini 4 Argon (100), Gemini 3.8 Flash (98), Muse Spark Open (95), Gemini 3.1 Pro Preview (93) and Inkling (91)
➤ Least metaphorical: GPT-3.5 Turbo (1), GPT-6 Luna (3), GPT-6 Sol (6), Llama 4 Maverick (8) and GPT-4o (Nov) (10)
In AA-PersonalityBench, Gemini 4 Argon is the most metaphorical of all models measured, followed by Gemini 3.8 Flash and Muse Spark Open
The metaphorical trait measures whether a model uses analogies or figurative language to make ideas tangible. It is one of five traits in the Expressiveness index.
➤ Most metaphorical: Gemini 4 Argon (100), Gemini 3.8 Flash (98), Muse Spark Open (95), Gemini 3.1 Pro Preview (93) and Inkling (91)
➤ Least metaphorical: GPT-3.5 Turbo (1), GPT-6 Luna (3), GPT-6 Sol (6), Llama 4 Maverick (8) and GPT-4o (Nov) (10)