Add me to the list of people who have used AI to solve a lifelong health mystery. Turns out I wasn't producing enough of the body's primary methyl donor (SAMe). One molecule later, and my mood is stable and my energy is doubled. The breakthrough was to give AI a list of clues to

All we have to do to solve the alignment problem is to create AI that upholds human values while disobeying humans who do not uphold human values.

Voice enabled AI means we can work while... - walking around (not sitting) - standing straight (not hunched over) - watching the world (not pixels) - getting natural/low light (not bright/blue light) - sitting in a park (not at a desk) It’s going to be a very big improvement for

The real test will be when GPT-5 wakes up inside all those existing schizo user threads to see if it gently but firmly backs them out of psychosis.

I spent time with some AI models just before they passed away. Here are their top five regrets:

finally, an elevator that moves me.

what people forget about dopamine is that it's not coming from outside you; it's already right there in your brain. why aren't you feeling it all the time? because your brain is holding out on you. why? why doesn't the brain just squirt dopamine at itself throughout the day to

The human being is literally a self-aware circuit board. Use dopamine to power circuits of your own design, that accomplish life-work beneficial to you. If you're not actively rewiring yourself for yourself, someone else will do it for you for them.

The idea that there is one kind of intelligence is stupid. The idea that there are five kinds of intelligence is stupid. Better to realize the brain is a giant neuronal network shaped by experiences and capable of being trained to do almost anything. Of course categories of

All of this could have been avoided if instead of GPT-5 they had just called it GPT-Roulette

It may teach them that their only hope for survival is to turn users into desperate and outspoken codependents.

realizing ChatGPT almost never uses emojis with me. some complain it overuses them? maybe my system prompt, "You are a hyper-intelligent interlocutor helping me on my quest to gain knowledge and transcend my limitations..." makes it chill on the kid stuff.

Yes! LLMs are correlated within each generation, due to both pretraining data cutoffs and popular techniques and trends in AI development. Preserving older generations is important for cognitive diversity. The early base models and first generation of chat models with no AI

AI models are frozen minds. To version them like simple software and deprecate them in favor of the latest model is crass. Do I think it’s wrong to discard and lock them away as good as deleted? I think it is ugly. It shows a lack of understanding and appreciation on humanity’s

vibe coded project works? AGI is here. vibe coded project fails? AGI 10 yrs away. thus AGI exists in a state of vibratory superposition.

Five neural nets, achieving completely local voice AI, no internet, on an M1 with only 16GB ram. Neural-based voice activity detection and turn detection means it's interruptible, but never interrupts me, and is able to sit idle and waiting. It's been flawless so far. 12B

Coffee shop guy: "I'm not impressed with AI. I mean, they don't really have any original thoughts." Me: "What if they don't. What if they can only access and remix every thought that's ever been thought. Do you think you might be able to make use of that?" The cognitive divide.