Dimensionscape

Proud dad, husband, programmer, part-time nerd.

Engineer at Dimensionscape and OpenFL

[Dimensionscape @Dimensionscape_]

@JK_Lundblad On that note, I'll probably just talk to Grok when I hop in the Tesla.

[Dimensionscape @Dimensionscape_]

@JK_Lundblad I enjoy talking to an uber driver. Not everyone is introverted.

[Dimensionscape @Dimensionscape_]

I can't really imagine it in that way. I think it'll look a lot more alien than that. Your robot isn't going to go shopping. Instead, I imagine shipping/supply logistic networks getting hyper efficient and traditional shopping becomes legacy all together because instant deliver of goods becomes incredibly cheap, and sending a humanoid robot with actuators and general intelligence seems like a waste of electricity.

[Dimensionscape @Dimensionscape_]

and... by all accounts, we live in an old, dead universe that, given the sheer probability of planets capable of supporting life, should be teeming with intelligence by now

It's not.

Perhaps I'm being too pessimistic and the real truth is that sufficient technological advancement allows us to step out of this universe entirely into someplace better. 🤑

[Dimensionscape @Dimensionscape_]

In fact, the probability of self-annihilation goes up with each progressive jump in complexity.

Each major jump in capability opens new failure modes that did not exist before: modern industrial food systems and food science created an epidemic of metabolic disease, fission made large scale war able to end civilization. molecular biology made high lethality, high transmissibility pathogens designable rather than only evolvable(COVID!), and sufficiently general AI could, in some scenarios, concentrate power or error at a scale no earlier technology allowed.

In that sense, the menu of ways we could destroy ourselves has grown with complexity.

[Dimensionscape @Dimensionscape_]

@JK_Lundblad Nobody is even close to putting humans back on the moon. It's hard.

[Dimensionscape @Dimensionscape_]

We don't call them developers because they aren't developing anything. I think most people imagine that someone is out there writing code to make AI tick. They aren't. It's almost entirely training and experimentation with weights, and we have no idea what abilities AI will have until it has them. AI is grown like a living organism or a crystal, we don't develop it. We grow these models in labs.

The truth is, we know as much about how AI really works as we do the human brain, which isn't much.

[Dimensionscape @Dimensionscape_]

@JK_Lundblad I personally love my RAV4 plugin.

[Dimensionscape @Dimensionscape_]

@JK_Lundblad Pirates still exist. Yes they are still terrorists. I don't know if mlkjr and the Hollywood romantization of 18th century pirates have anything to do with eachother

[Dimensionscape @Dimensionscape_]

@JK_Lundblad Ironically, we all went to space based on german designs.

[Dimensionscape @Dimensionscape_]

China went from producing 15% of the worlds steel in 2000 to producing 57% in 2020, and since then its been declining while steel production started growing in the west again.

The inversion came because of China's real-estate crash plunged their domestic consumption and they just started dumping it. I personally believe that you need tariffs to fight something like that to avoid eroding your own steel industry.

State-backed surplus should never be allowed to threaten a collapse of domestic industries, especially when they're critical to infrastructure and defense.

[Dimensionscape @Dimensionscape_]

@JK_Lundblad For example, I'm okay with heavy tariffs on steel. Give subsidies to builders when they buy American steel.

[Dimensionscape @Dimensionscape_]

To my greater point, I don't think tariffs are inherently bad. Yes, they're probably being applied poorly. The part that is irritating to me is that we are negatively coupling the idea of tariffs to Trump, making the approach more radioactive than it really is. There are both smart and dumb ways to apply policy where it matters.

[Dimensionscape @Dimensionscape_]

@JK_Lundblad I just agreed that liberation day tariffs were bad.

[Dimensionscape @Dimensionscape_]

A star begins when a dense clump inside a cold molecular cloud becomes gravitationally unstable and collapses, fusion doesn't even begin yet. Gravity is what creates the conditions for fusion. Hydrostatic equilibrium appears after fusion has already started. Your analogy reverses the casual order.

[Dimensionscape @Dimensionscape_]

You could have said the same thing about the internet in the 1990s: “The more internet infrastructure your country builds, the richer it will be.”

And...that was directionally true over a long horizon and badly wrong as a near term investment rule. Yeah, fiber, routers, and data hotels were real, but so was the overbuild. The technology became fundamenta, but at the same time, a lot of the first capacity did not earn its cost of capital on the timetable people assumed.

The same split still applies here.

The flaw in this discussion is that more data centers dont necessarily mean more intelligence. They mostly add scale and capacity. The intelligence is already formidable, and it is a separate thing all together. Extra gigawatts let you serve more users, run longer traces, and spin up more agents. They do not automatically produce a smarter model... and whether that extra capacity turns into profit is still completely theoretical. Everything is being built on promised demand, not on a fully proven market that already covers power, depreciation, and a four-to-six-year chip refresh. The cash is leaving now. The return is scheduled for later.

Yea, AI is going to be fundamental too. That is not the hard question. The hard question is when the economics catch the build. It is not clear that hyperscaling pays off in the near term.

[Dimensionscape @Dimensionscape_]

And, I would just note that what people call “the new model” in 2026 is often just the same weights plus a better jacket, meaning post-training, effort sliders, tool loops, memory, multimodal glue, safety routers, and a harness that has learned the exam. Fable 5.1 is the exhibit here and twelve weeks after Fable 5, no-tools reasoning moved a little, the composite moves a little, and one new agentic science board doubled. That is not what a new pretrained brain looks like. This is what a stack looks like when the lab spent the interval on RL, scaffolding, and the leaderboard.

[Dimensionscape @Dimensionscape_]

@JK_Lundblad In simpler terms: the accumulative accomplishments are terrific, but the the hype is 20% real progress and about 80% competitive marketing.