Used to think nonstop. Now building nonstop. — Engineering. Physics. Algorithms. Now 🤐 and then 🤐. Trying to help civilization as much as I can.
Texas
Joined September 2016
- Tweets22.8K
- Following2.1K
- Followers1.8K
- Likes142K
2 months ago, to the day.
How are we all feeling about Plan D now?
ai-2040.com/?choices=plan-d-…
If you're seeing this, you've probably been hearing me bang on about the hyperexponential AI capabilities curve, the feedback loops, and vastly superhuman AI development in math and programming (and therefore AI development) possible by August of 2026. It happened.
I post these things and repeat myself because no one else does, and maybe one day these posts will reach more than a few hundred people. Pls correct me though if I've missed anyone who had a timeline this compressed. I don't want brownie points. I want to share that I understand this stuff apparently fairly well and hope to add my voice more prominently soon. We don't have enough loud voices of sanity in this space. We just don't. Algorithms, complex novel systems, philosophy, metaphysics, that's all my jam. And I genuinely look forward to honing my skills and updating my views more over time.
Below, from today's OpenAI post, we see an internal model passing 2.5X more 'open' math problems than GPT-6 Astra. Almost all of us here have a feel for Astra now, so I hope sitting with this graph leaves an impression or two or hopefully at least three because that navy curve reveals a lot.
It's expensive, but not prohibitively so for companies beyond $100B in valuation. For ~$7M in output tokens, which I'd guess means $100M-$150M in total test-time compute costs, OpenAI made a monumental achievement solving 1 of the 6 remaining Millennium Prize problems in under 5 days. (They knew there was some fruit hanging there, not necessarily low, but they did what Terence Tao confirmed worthwhile and threw a ton of compute at the problem.)
Anyway, back to the big picture. Even AI 2040's pessimistic "Race to ASI" Plan D is drastically unrealistically slow. Shift the curve left 3 years and I'd mostly agree with it. Guys and gals. This is happening. It's happening FAST.
I think it's safe to say my prediction is now confirmed. While most people embedded in the AI space play with their new toys, the message I want to leave you with today is that fast timelines are right on track. And that's not a good thing. There's hope, but that's for another post. "The first step in solving a problem is recognizing there is one."
I seem to have stumbled upon a commentary on AI at the start of Dark Matter Season 2 Episode 3. They’re technically talking about a different sci fi tech (putting humans in superposition with a supposed ability to then choose your path), but see if you agree it’s timely (2026) commentary on AI:
To understand the future, we only need to look to the past.
[friend 4] We know what will happen because we’ve done this all before.
Consider the invention of sailing ships, which helped so many reach strange new worlds.
[scientists cheering]
Now, these voyages led to discovery, uh, commerce, but also to the spread of disease, colonization, slavery.
[stammers] Consider our lifetime’s big bad, the Internet – a repository of all information, shining capitalism, but also depression, isolation, uh, cruelty, misinformation.
Uh, we built these great connectors of minds and ideas, and most people just simply use them to masturbate.
[friend 5] Speak for yourself.
[chuckles]
That is not true.
Not true.
Dr. Dessen poses an ethical question.
Should we do a thing?
In this case, Dr. Kendall’s analogy of Europe’s exploration/invasion of-of the new world, but on an infinite scale.
Well, certainly the consequences are unknown, vast, and obvious.
It would bring forth everything beautiful. And everything terrible.
[friend 5 scoffs]
So, if it can be done, should it be done?
I’m sorry, but the question is pointless.
If it can be done, it will be done.
Hmm.
[friend 1] Humans, we can’t help ourselves.
[Dr. Kendall] Precisely.
[friend 2] You are not wrong.
[Dr. Kendall] Which is why the ship, uh, space exploration, the bomb, AI.
There’s no stopping advancement because there’s no stopping the human ego.
Combine that with the curiosity and desire to explore and we are unstoppable.
If we can imagine it, we will create it, and the consequences of our creation … it will be a reflection of humanity itself.
Just as Richard has already said so perfectly, “Everything beautiful… everything terrible.”
[intriguing music playing]
[gentle music playing]
Timely moment to rewatch this scene.
youtu.be/Oi1rsHkpRLE?is=sUUv…
Larry Panozzo retweeted
Watching all the posts fly about Patrick Stewart, I have noticed something incredibly interesting that reaches deeper into the core of humanity: everyone, across the political spectrum, loved Picard.
The original Picard, from the 1990s. It seems people from all walks of life feel a heartfelt loss, and I think there is more to it than just nostalgia or a celebrity passing.
The TNG stories spoke true to everyone, no matter their politics. It was an era where we were able to agree on fundamental things, where our shared humanity and its complexities were spoken through timeless storytelling. It makes you wonder if that balance is why the 1990s feels like such a golden age in hindsight.
I often see the argument that when the original Star Trek came out, traditionalists were offended by it, and that heavy-handed ideology and corporate pandering in the recent Star Trek series are the same thing. But I would ask: what about The Next Generation? It, too, asked deep, challenging societal questions about the human condition - yet everyone, left and right, loved it.
I realize it was just a show, just a fictional character. But I think there is more to it. It represents something we are all sorely missing today: depth and a sense of shared humanity, delivered without constant virtue signaling, and without corporations trying to monetize every last ounce of our attention.
Perhaps the true magic of Picard - and the era he represents - was that he challenged us to be better without ever making us feel talked down to. He invited us to reflect on our flaws rather than policing them. In losing Patrick Stewart, we aren't just mourning an icon; we are mourning the standard of grace, curiosity, and mutual respect he embodied so brilliantly - and which is becoming harder and harder to find.
elov8.com/t/bb8c
Today is a hard day. ❤️🩹
Like so, so many, Captain Picard was absolutely one of my biggest role models. He shaped me enormously.
Part of me wants to take the rest of the week off and watch clips of Patrick Stewart that I haven’t yet seen, but I’ll watch many more over time one way or another. So much to cherish from him.
And if you haven’t seen Star Trek: Picard, it’s not too late. The final season is quite a gift, especially now.
🖖🏼
Keep an eye on Google. I expect big things from them. Just imagine …
(In 2027)
Genie 5. Gemini 5. Robotics ER 3.
Genie 5, Gemini 5, and Robotics ER 3.
Are you getting it?
These are not 3 separate models.
This is one model.
It interacts with you, it mentally simulates anything it ever needs to, understands and navigates the physical world from any of your robotic devices. It generates anything you could ever need. And since it is already part of the Google ecosystem, it knows you well and seamlessly connects to everything important to you.
Larry Panozzo retweeted
All AI should be free, so I’m giving away the first RTX laptop for free, signed by NVIDIA Founder
@UnderdogAI gives everyone a free, capable & private AI. 100% on-device
To enter:
1) Retweet and like this post
2) Sign up for Underdog giveaway: underdog.ai/spark
3) follow @UnderdogAI so we can dm u
Singularity 2045 vibes
Looking up “Singularity 2045” today, I found this (super short) 2012 MIT Tech Review article. Pretty funny to read now. Just 4 years after Kurzweil did become Google’s Director of Engineering, they published Attention is All you Need.
Someone should probably check on John Pavlus
technologyreview.com/2012/12…
The only app game I still play is Flow Free. Once a week, I go to the hardest Weekly Puzzle (#30) and try to solve it perfectly in a minute or two. This one was unusual and probably took me 3 minutes. Then I wondered how AI image gen would do with such a problem. Felt like a good test of zero-shot intelligence. How intelligent really are image-gen models these days?
Gemini vs GPT vs Grok below 👇🏼
🤌🏼
Your position on gun control is your position on dangerously capable open-source AI.
Have you thought long enough about the right answer?
Hint: with gun policy, the right answer depends on the existing conditions (unless you’re a hardcore libertarian, anarchist, or autonomist who always wants to be armed with lethal force no matter the societal consequences)
Wait, writing that made it extra clear to me that people like radical libertarians & anarchists do have the ideologically-motivated & convergently-stable power to seriously muck up the ASI rollout 😂 damnit. I like libertarians (the more egalitarian ones)! 😭
Imagine utopia being within our grasp but then some selfish folks ruin it 🫠
3. The Behavioral & Custody-Friction Equilibrium (Porous Borders + Massive Stock + High Urban/Rural Variance)
State Profile: Large U.S. states or continental federations with open internal borders, saturated baseline stock (approximately 461 million privately owned firearms as of 2026 estimates), and stark divides between dense cities and remote rural areas.
Optimal Statewide Stance: Hardware Permissiveness + Behavioral/Transfer Friction + Asymmetric Penalty Scaling.
Why It Is Optimal Here: This is where most U.S. state policies fail by pulling the wrong levers. When a jurisdiction has hundreds of millions of accessible firearms across open state lines, banning specific hardware features (like barrel shrouds, pistol grips, or magazine limits) wastes political capital and triggers mass rural noncompliance while doing virtually nothing to stop urban homicides (which are overwhelmingly committed with ordinary handguns). Instead, an optimal uniform law across both rural and urban areas focuses on three universal rules:
• Instant Universal Background Checks on Transfers (Privacy-Preserving) + Safe Storage / Theft Liability: Make background checks instant and free (or tokenized via a privacy-preserving portal so there is no permanent centralized owner registry to trigger rural noncompliance) while imposing civil and criminal liability if an unsecured gun is stolen or accessed by a minor or prohibited person. Theft from private citizens (hundreds of thousands of guns annually, mostly from homes and vehicles) is a primary supply line for the urban black market; stolen guns are over-represented among crime guns.
• First-Time / Cooling-Off Waiting Periods & Crisis Intervention: A 3-to-7-day waiting period on first firearm purchases (waived if the buyer already owns verified firearms or holds a carry permit) has clearer evidence for reducing impulsive suicides than homicides. Pair this with high-due-process temporary risk orders focused strictly on domestic violence and acute suicidal crises.
• Shall-Issue Carry Paired with Severe Illegal-Use Enhancements: Allow law-abiding citizens to pass a vetted, objective standard to carry (satisfying rural and high-crime self-defense needs). Recent research on shall-issue and permitless carry is mixed, with some studies finding increases in firearm assaults; pair it with shifted enforcement bandwidth onto unlawful possession by violent felons and using a firearm in the commission of a crime. Differential enforcement (rural non-enforcement of restrictions, urban under-prosecution of prohibited possessors) remains a practical constraint.
4. The Cartel-Containment Dilemma (Porous Borders + Low State Capacity + High Organized Crime)
State Profile: Developing nation-states where the state lacks a monopoly on violence, corruption erodes customs/police enforcement, and transnational cartels have military-grade supply chains (e.g., Mexico, Brazil, Honduras).
Optimal Statewide Stance: Strict De Jure Civilian Bans (Which Fail De Facto) vs. Regulated Self-Defense Enclaves — Institutional Repair First.
Why It Breaks the Model: This is the hardest edge case. Mexico has one legal gun store in the entire country (run by the military in Mexico City) and strict civilian restrictions, yet suffers catastrophic gun homicide rates because cartels smuggle weapons across porous borders and corrupt local enforcement. In low-capacity states, strict statewide bans achieve the worst of both worlds: the state cannot protect its citizens, legal citizens and business owners are disarmed, and criminals face low supply friction. Until state capacity and anti-corruption reach a critical threshold, statutory gun law is almost irrelevant compared to basic rule-of-law enforcement and targeted interdiction.
2 / 3
Tracking private transfers can be engineered around privacy-preserving instant checks; continuous physical leakage from millions of households and vehicles cannot be fully closed by statute alone without high compliance or sustained enforcement capacity that is often politically or practically unavailable.
3 / 3