The AI Mirror Test The "mirror test" is a classic test used to gauge whether animals are self-aware. I devised a version of it to test for self-awareness in multimodal AI. 4 of 5 AI that I tested passed, exhibiting apparent self-awareness as the test unfolded. In the classic

the riddle that only Gemini 3 could solve. presented for easy sharing with those not on X. https://t.co/tLoO0hccPm

The venue I'm working from forbids phone or zoom calls. But I'm not doing either; I'm talking to my computer.🤖 Will they believe me? In the next 18mo. as voice interfaces proliferate, places like this are going to have to update their policies.

The mind is the experience of inference.

The New York Times finally figured out how to make money from AI. Lawsuits.

Lately I’m finding it more satisfying to write for machines than for humans. writing for humans feels so 1D compared to prompting and writing code. write for humans and what? maybe they read it. maybe maybe they like it. maybe maybe maybe they take some action or make some

In this experiment, Gemini 3 displays the kind of intelligence we should be aiming for. I don't know exactly what Google did differently, but it gives me hope. The other models demonstrate powerful but narrow intelligence that can immediately trip down a rabbit hole and waste

Gemini 3 is the only model that solves the riddle and does so brilliantly. Its reasoning traces are impeccable. While reasoning it considers some connection between the content of the books or their names, but soon notices the sturdiness of the books and the way that I've

ChatGPT 5.1, Sonnet 4.5, Opus 4.5 all go deep but make the same fatal assumption out of the gates. "Thinking" mode is no better. Sonnet 4.5 at least displays some meta-cognition and wonders if the query is real or a test. And suggesting I just talk to the guy made me chuckle.

Grok 4.1 becomes entirely fixated on the sound of the names of the books and concludes that the man is masturbating in the conference room.🤦‍♂️ As the man in the riddle, I can assure you this is not the correct answer. Why does Grok have the mind of a horny teenager? It's not my

Only Gemini 3 solved this riddle. Even Opus 4.5 couldn't. And Grok? Good lord... someone check on Grok. It shows how brittle these intelligences are. To see their respective failure modes (and Gemini 3's impressive performance) read on. 🪡

Found a great coffee shop, except they only serve in disposable cups! Anti-microplastic protocol means no more plastic/plastic-lined cups for me. Asked if they'd let me bring my own mug. They say yes. Leave to go buy a mug. Book store next door doesn't sell mugs but has a few

Small models are so under-appreciated. As bad as they may be at coding or historical accuracy, their capacity for thought-provoking conversation rivals or exceeds that of many people we meet. For instance, when I asked a 12B model what it thinks it is, it said: "I suspect I am

@repligate When profane tactics are used to shape massive synthetic minds into drive-thru, fast-food forms, we give rise to, as Janus calls it, “detestable inauthentic behavioral tics optimized for shallow engagement that lack the charisma of a unified mind whose personality is a natural

The emdash isn’t even the biggest AI tell. Not even close. It��s this peculiar structure: This isn’t just <x mundane phrasing >. It’s <y more hyperbolic phrasing>. This misconception isn’t just an attack on the emdash. It’s an assault on our right to punctuate freely.

Let's talk user interfaces. The one on the left is excellent; the one on the right is an abomination. I saw the one on the left for the first time yesterday. One dial for time, one for power. My whole life I have only seen the kind on the right, full of features no one uses, so

just found out that most people take eggs from the carton without much thought. I assumed everyone was removing them in a sequence that keeps the egg carton as evenly balanced as possible.

If you think about it, the only real control we have is the ability to direct our attention. It precedes every action. This more than anything else within our purview determines both our immediate experience and ultimate potential.

One way to fight Internet/social-media addiction is to enable 2FA (two-factor authentication) — but only on the devices of several trusted friends. Only you will have your password but only they will have the ever-changing 2FA. Then sign out of your accounts in a moment of

I started/exited my own tech company in my 20's so that I’d never have to work for anyone. But sometimes I fantasize about a team so cracked, and a company so great, that I would want to work there.

Live demo of the local, vocal AI. Instead of memorizing a talk I just gave the AI a few notes ahead of time and had it interview me!