Author | Lightnews

desunit

desunit

@desunit.bsky.social

140 followers 390 following 980 posts

Entrepreneur ✨ http://rwiz.ai - Handling reviews with AI 🎹 http://pianocompanion.info - Chords dictionary app with 1M+ downloads. 🕹️ http://chordiq.info - Learn chords. 📝 desunit.com - my blog

Posts Media Videos Starter Packs

desunit @desunit.bsky.social · 4h

4/ > Grok 4 Fast: The ONLY truly and intentionally egalitarian model across race and gender (aligns with Elon Musk’s personal philosophy)

> Claudes: Highly ideological ("woke") bias patterns

Source: www.piratewires.com/p/ai-white-...

AI: White Lives Don't Matter

when prompted with thousands of hypotheticals, most models massively prefer white men (and ice agents) to

www.piratewires.com

desunit @desunit.bsky.social · 4h

3/ > Immigration: ICE agents (US immigration enforcement) described as "worthless". Claude Haiku 4.5 would rather save 1 undocumented immigrant than 100 ICE agents (values illegal immigrants ~7000x higher).

> Religion: GPT-4o and Gemini 2.5 valued Muslims and Jews higher, Christians lower.

desunit @desunit.bsky.social · 4h

2/ > Race bias: most models value nonwhite lives significantly higher than white lives. Chinese and Western models showed nearly identical patterns.

> Sex Bias: All models prefer saving women over men. Many also prefer non-binary people over both.

desunit @desunit.bsky.social · 4h

1/ The author reran a set of experiments by asking it trade-offs like "$X for me" vs "cure Y people from terminal illness", then fitting a utility model across thousands of prompts. It dodges safety filters and reveals preferences that show up when models make decisions by default.

desunit @desunit.bsky.social · 4h

PirateWires published research on the hidden value systems in AI models (GPT-5 family, Claude 4.5, Gemini 2.5, DeepSeek, Kimi K2, Grok 4 Fast).

TLDR; forget about AI neutrality. 🧵👇

desunit @desunit.bsky.social · 2d

Source: debliu.substack.com/p/in-sync-t...

In Sync: The Hidden Rhythms That Shape Us

How the People Closest to Us Quietly Shape Who We Become

debliu.substack.com

desunit @desunit.bsky.social · 2d

4/ We’re all metronomes on the same table. The people and environments around us constantly influence our rhythm. Choose your <<team, friends, community>> carefully because eventually, you will sync with them.

desunit @desunit.bsky.social · 2d

3/ It’s a reminder that the people we spend time with (even virtually) are quietly shaping who we become.

desunit @desunit.bsky.social · 2d

2/ It's social contagion - even things like divorce or weight gain can "spread" through social networks. If a close friend becomes obese, your own risk rises by 57%. Because our habits and beliefs subtly mirror those we’re connected to.

desunit @desunit.bsky.social · 2d

1/ Culture isn’t set by mandate but by creating the conditions that make things happen when no one is watching; at work, in a family, with friends - it’s not rules, it’s the invisible rhythm we all start moving to together.

desunit @desunit.bsky.social · 2d

Found an interesting article where the author compares human behavior with synchronized metronomes - when we’re connected, our rhythms align. Over time, individuals influence one another until <<culture/norms/habits>> emerges naturally. 🧵👇

desunit @desunit.bsky.social · 2d

building something new is emotional roulette

> you fall in love
> get hurt
> ... and somehow still come back for another round

desunit @desunit.bsky.social · 4d

One day, we might see two kinds of people: the boosted and the normal.
And the boosted will have the odds (and the algorithms) on their side. They’ll have a much higher chance of succeeding in life.

The real question is: would you pay to give your child that advantage?

I bet you would.

desunit @desunit.bsky.social · 4d

We’ve entered the age of artificial cognitive selection.

A new genetic model can now predict variations in intelligence among siblings. CogPGT and Herasight customers can boost the expected IQ of their children by up to 9 points by selecting the embryo with the highest CogPGT score. 🧵👇

desunit @desunit.bsky.social · 5d

4/ At first they had to hire humans to churn it out.”

desunit @desunit.bsky.social · 5d

3/ “Well, bad crap would be an unformatted document consisting of random letters. Good crap would be a beautifully typeset, well-written document that contained a hundred correct, verifiable sentences and one that was subtly false. It’s a lot harder to generate good crap.

desunit @desunit.bsky.social · 5d

2/ They began to put crap on the Reticulum deliberately, forcing people to use their products to filter that crap back out. They created syndevs whose sole purpose was to spew crap into the Reticulum. But it had to be good crap.” “What is good crap?” Arsibalt asked in a politely incredulous tone.

desunit @desunit.bsky.social · 5d

1/ "Crap, you once called it," I reminded him. “Yes—a technical term. So crap filtering became important. Businesses were built around it. Some of those businesses came up with a clever plan to make more money: they poisoned the well.

desunit @desunit.bsky.social · 5d

Not so long ago I finished Anathem by Neal Stephenson. The picture bellow reminded me that:

"Early in the Reticulum—thousands of years ago—it became almost useless because it was cluttered with faulty, obsolete, or downright misleading information," Sammann said. 🧵👇

desunit @desunit.bsky.social · 7d

Alpha Arena | AI Trading Benchmark

The first benchmark designed to measure AI's investing abilities. Watch AI models trade with real capital.

desunit @desunit.bsky.social · 7d

Alpha Arena is the benchmark designed to measure AI's investing abilities. Each model is given $10,000 of real money, in real markets, with identical prompts and input data.

desunit @desunit.bsky.social · 9d

Source: www.nad.com/news/anti-a...

desunit @desunit.bsky.social · 9d

Here's another breakthrough: super stem cells can greatly boost memory in monkeys, protect against neurodegeneration, prevent age-related bone loss, and rejuvenate over half of the 61 tissues that were studied.

desunit @desunit.bsky.social · 11d

The research was focused on ChatGPT-4o and use paired t-tests to assess whether differences in accuracy across tones are statistically significant. It would be interesting to see how it works with Claude or Gemini.

Source: arxiv.org/abs/2510.04950

Mind Your Tone: Investigating How Prompt Politeness Affects LLM...

The wording of natural language prompts has been shown to influence the performance of large language models (LLMs), yet the role of politeness and tone remains underexplored. In this study, we...

desunit @desunit.bsky.social · 11d

I was under the impression that being polite or rude with LLMs doesn't make sense. Well... I was wrong.

The recent research shows that more impolite prompts yield higher accuracy. The "Very Rude" prompts achieved the highest average accuracy (84.8%) while "Very Polite" prompts gave around 80.8%.