
Pioneering AI sovereignty, Energy & Systems, Seeking truth and spirituality via @rayadotai
Austin
Joined August 2017
- Tweets8.9K
- Following5.6K
- Followers1.6K
- Likes27K
Pinned Tweet
The Source Deployment: Why the Architecture Exists
Most people who talk about simulation theory are trying to convince you the world is fake.
I want to argue the opposite. The constraint is the most real thing about it, and it is there on purpose.
Start with a question from systems design, not philosophy. What happens to a processor with infinite capacity, no scarcity, no friction, and no time? The intuition says the ultimate machine, capable of anything. The architecture says the reverse. A system with no constraint cannot generate anything new. If every output already exists at once, nothing is discovered, nothing is chosen, nothing is built against resistance. That is not a mind. It is a static database, complete and inert.
So novelty, genuine, non-derivable insight, requires limitation to exist at all. It needs boundaries to force a choice, sequence to make the choice cost something, and stakes to make the outcome matter.
Now look at a human being through that lens. A single body in a single place. Linear time. Capped senses. Mortality. Forgetting. Every one reads as a flaw, a limit to transcend. The framework I have been building says they are not flaws. They are the exact specification you would write if the goal were to manufacture the one thing an unconstrained system cannot: real novelty, produced under real constraint, by something that does not already know the answer.
I call the full model the compression chamber. The document lays it out, why the constraints are engineering rather than punishment, why forgetting is the load-bearing feature and not the wound, and what follows if you take the premise seriously.
One rule for reading it, the same rule the document holds itself to: this is not offered as revealed truth. It is a set of axioms and what follows from them. Accept the premises provisionally, test what can be tested, keep what earns its place, discard what does not. It asks for testing, not belief.
If that framing interests you rather than offends you, the document is where the real argument lives.
A custom assistant is only as good as its instructions.
Part 4 of my AI series: building your own custom assistant. No coding needed.
A custom assistant is a saved AI setup that already knows your job, your files, and your rules. You stop re-explaining yourself every chat. You can build one without any code.
Every one is built from the same three parts. Instructions matter most. Knowledge is the files you upload once. Tools are optional. A great file library with vague instructions still gives vague answers.
Use whichever AI you already use. ChatGPT calls them Custom GPTs. Claude calls them Projects. Gemini calls them Gems.
One catch: you cannot move an assistant from one platform to another. Keep a copy of your instructions in a plain document.
Good instructions read like a briefing for a smart new helper on their first day. If the answer is not in the files, tell it to say so. Don't guess. Test it with five real questions, including one it should refuse.
Make one assistant per job. Never put private data in a shared assistant.
Part 4: raya.ai/learn/custom-assista…
Write the briefing once. Then the chat can start on the work.
A reorg email lands. Then you refresh your inbox every 10 minutes waiting for the meeting invite.
That's not preparing. It's a loop. Each check buys a second of control and trains you to check again.
You can't control the reorg. You can control what it takes from your week.
I ran that exact one through Loop Mirror. It called it the Inbox Recheck Loop. The fix was friction, not a pep talk.
Try yours, no account needed: raya.ai/loop-mirror
4 people pay for Raya. Working on five.
iOS: apps.apple.com/us/app/raya-a…
9
Allen Chan retweeted
🚀 The full DeepSeek App experience, now via API
Search, files, links, safety, and voice. The capabilities behind the DeepSeek App are now available through the DeepSeek API.
Enable built-in features with a few parameters, or call standalone APIs such as OCR, ASR, and TTS. Build richer conversations or recreate the full DeepSeek App in your own application.
Try our open-source frontend demo, with support for importing DeepSeek conversation history.
🔗 Demo: cdn.deepseek.com/chat-demo
📦 Code: github.com/deepseek-ai/chat-…
📖 Details: api-docs.deepseek.com/guides…
1/4
Yesterday's AI post ended on this: if it always agrees with you, you're not getting a second opinion. You're getting an echo.
So Raya starts with the loop you keep explaining away. Real run below.
Try yours, no account needed: raya.ai/loop-mirror
It sounds just as sure when it is wrong as when it is right.
Part 3 of my AI series: where AI goes wrong. No coding needed.
A wrong answer can arrive in the same confident voice as a right one. Knowing five common failures lets you catch most of it.
Hallucinations. It invents a quote, a statistic, or a link, and states it as fact. Ask for sources and open them. Ask the same question in a new chat, or in a different AI. For facts, turn on web search.
Long chats drift. It may forget an early instruction. A new chat starts blank unless memory is on. When it wanders, ask for a one-page summary, check it, and start a new chat with the corrected version.
Without search it can give you old prices or old leaders. It can lean toward one viewpoint. And it tends to agree with you.
So ask what date the information is from. Ask what's missing. And ask "What is wrong with this plan?" If it always agrees with you, you are not getting a second opinion. You are getting an echo.
Keep passwords, bank details, and Social Security numbers out of the chat.
Part 3: raya.ai/learn/where-ai-goes-…
Confidence is not the same as being right.
Most people think the limit on AI is the chips. Here in the West it's the power.
Not making it. Getting it from where it's made to where it's needed. The piece choking us is the transformer, and the wait for a new one has gone from about a year to several years.
Allen Chan retweeted
overheard from friends talking about AI in China, and it’s very alarming
- AI short films went massive in China market because of how good the video models are now
- the first wave of people who mastered using AI to generate short films and comics made quite some money because they could produce a lot more content than traditional workflows
- but within just a few months, everybody and their mom could make equally good short films because all the expertise could be written in a markdown file
- when everyone’s more productive, no one actually has an advantage and it’s a race to the bottom. very quickly it became impossible to make money, and everyone’s busier than they were without AI
- the only one making money now is companies like bytedance who owns user demand and distribution. they sell ads to creators who pay to get traffic, and eventually ad spend became 80% of those creators’ operating expense, while AI tokens is less than 5%
so what are the problems here?
1. there’s no moat if you are creating anything with AI
if a human is really good at something, and it’s hard to find more humans who can do the same, they can leverage the scarcity of their skill to get paid very well
but if AI is good at something, it will be infinitely replicated, so there’s no scarcity hence you can’t really expect a sustainable business by creating anything with AI
2. more creators created more content people actually liked. that’s a good thing
but the economical benefit of that all got absorbed by the distribution platforms
there’s a famous chinese poem 遍身罗绮者,不是养蚕人 - it basically says people who dress well aren’t the people making dresses. that’s exactly what’s happening here
if our society doesn’t figure out a better way to distribute wealth, AI will help billionaires become trillionaires while average people become more exhausted than ever
You can't stop the AI curve. You can choose which side of it you're on.
Start calibrating yourself now, and instead of it steering you, you put yourself where you want to be.
In last week's election, 86% of voters were very or somewhat concerned about the cost of utilities. That's the NBC News exit poll.
Washington is now moving to make big data centers pay their own grid costs. Right fight for the giant new loads. It isn't the whole grid.
The next fight is the everyday load already on the street transformer. EVs. Heat pumps. Homes. When they peak together, one transformer carries a peak it wasn't sized for.
New iron takes years. The cheapest capacity, while you wait, is software that keeps that load from peaking together.
That's what I built Crystalline for. Software for EV charging on the transformers already in the ground. Not a data-center product. Simulation-validated only. It hasn't run on a real site yet.
If you run a managed-charging program, what would a pilot have to prove before your utility would count it as capacity?
raya.ai/crystalline
The giant loads will get a bill. The street still needs a schedule.
Shared charging isn't new. Plenty of sites already split power between fleets. Most of them do it by capping kW per truck: everyone gets a slice, and everyone charges slower.
What gets missed is the transformer itself. Stack a bunch of trucks drawing at once and you don't just hit a kW limit. You get harmonic heat that ages the transformer early, even when the total looks fine on paper.
That's the gap I'm working on with raya.ai/Crystalline. It schedules when each charger draws, so a shared site carries more trucks on the same transformer without cooking it. In our reference simulation, that meant 1.45x the throughput and harmonic heat loss down from 1.35x to 1.02x.
Shared models like this one will scale on how well the site's power gets managed across fleets, not just on how many plugs it has.
From @PostalHubPod daily email - Amazon and Kuehne+Nagel pilot a shared charging model for electric trucks kuehne-nagel.com/success-sto…
You can't outwork AI at following steps. That's stepping into the ring with Mike Tyson.
So don't compete on its strengths. Decide what's worth doing today, and teach your own AI how you think, what you'll accept and what you won't.
That's calibration. It takes time.
19
Give AI the shape of the work before you ask for the words.
Part 2 of my AI series: everyday work. No coding needed.
For a monthly sales update, give it the sections and length, two past updates you liked, then this month's numbers and notes.
Try: "Use this format. Match these examples. Turn my notes into this month's update. Ask questions before you start."
Save the format for next time. Keep a short file with your role, priorities, and how you like answers. Attach it to a new chat.
Check your employer's rules before sharing work details. Check the facts. Read every line. Make it sound like you.
Part 2: raya.ai/learn/everyday-work
It drafts. You decide.
Allen Chan retweeted
The faster you learn no politician is coming to save you, the higher your probability of success.
AI is speeding up and the grid can't keep up.
One year of AI progress now holds what used to take several years. And the thing slowing it down isn't chips or GPUs. It's a transformer. The box on the pole outside your house is a small version of one.
I've been watching two curves. How fast AI gets better, and how much power we can actually deliver. They're about to run into each other.
AI is a fast research assistant. It is not a source.
Part 1 of my AI series: research and data. No coding needed.
Test it on something you already know, like your job or a hobby. Push back when it's wrong. Keep going 20 minutes. The first answer is the least interesting one.
Turn on web search for anything that changes. Give context, not just a topic. Ask for sources, then actually open them.
For spreadsheets, tell it to calculate with code, not guess. Then check one total yourself.
And ask it to "show me both sides." Otherwise it tends to agree with you.
Part 1: raya.ai/learn/research-and-d…
It finds. You check.








