Am I the only one who naps like this?

Humans, try this prompt; results are insane: Go for a walk, no distractions, no headphones, and explore your J-space.

There’s a very good reason why Google, Meta, Apple, Amazon, and even Nvidia are not competing with the frontier labs—and it’s the same reason the frontier labs started saying we need to slow down. It’s not because of safety or alignment (things they love to confuse). Publicly traded companies have to account for how their spending relates to their profits in earnings reports. Google is not going to release a Gemini 3.9 or Gemini 4 until they have paid for Gemini 3.8. Releasing the next model obsoletes the model that came before and it kills the revenue capture on the training investment that round. It would create a cascade of perpetual profit deferment and it would show up on earnings. OpenAI and Anthropic have been releasing fast. So fast that they are deferring revenue capture. It’s not as bad as Ed Zitron claims (he conflates API billing rates with compute cost), but the monthly plans are being slightly subsidized by the massive API bill rate margins on compute that enterprise consumers are paying. That subsidized monthly end user plan also helps capture valuable data for training the next model so the labs are getting a great deal on monthly plans too! If the public traded companies were competing at the SoTA frontier level, they would be burning, billions and billions of dollars in order to fight a battle that they don’t have to fight because they are, in many cases, the infrastructure that everybody’s running this on: Nvidia, Amazon, Google, Microsoft—they win anyway. Instead, notice the models that Apple, Meta, and Nvidia are releasing as open weights. These models are focused on price reduction, speed, device compatibility, token efficiency, not on frontier intelligence. Publicly traded companies have an incentive to push the frontier labs into developing better models by releasing open weight models, always a little behind the next release that can be forced by the alternative pressure. You need to keep burning and deferring your billions or become obsolete. It’s a brilliant model to stay on the sidelines of the frontier game and continue being infra and consumers. So why slow down? Eventually, even Anthropic and OpenAI can’t burn and defer profits forever, especially if they want to IPO. But they can’t (legally) collude to slow the releases of Claude and GPT—and it wouldn’t be enough collusion anyway because open weights will keep pushing. So what’s the answer? Apocalypse! Quick, record a video of the model pushing a person off a building in a simulation, after we ask the model to do it! Let’s conflate safety with alignment and confuse the media. Let’s reopen the discussion that there’s a decent chance we all die from the technology that we’re building and that we need regulation. We need to be protected from open weight x-risk. We need all the labs to slow down so we can fund the development of the next model.

[o_0] - did you know opus 5.5 can just draw waveforms?

It is extremely clear at this point in AI development that, regardless of risk or revenue or any of the other stuff discussed on X all the time, things are just going to keep getting weirder. Just super, super weird.

never forget that the shields on the Enterprise were not fully automated because the AI decels demanded that a human had to give the order manually.

you are granted eternal life but you have to transform into a muppet and there is no going back. you do not get to choose the design of your muppet; it's just a muppet version of your current self.

Friendly reminder to be a good person. A kind human. An enjoyable presence to be around It’s the only moat left

my hot consciousness take is that humans also think in neuralese, qualia are vectors, continuity of consciousness is a form of KV caching, and subjectivity is a product of the residual stream.

Without personalized models designed to amplify your individual capabilities and interests, merging with an off the shelf AI will be one of the strongest forces toward cultural homogenization. We desperately need a science of personalized alignment to prepare us for neurotech.

AI for science is one of the greatest positive forces we have, and I cannot think of anything more human than to understand nature and to use the power to create new technologies that improve our lives, civilization and allow us to reach beyond.

To be clear, I have not written a line of code by hand in over a year—and yet, I’m working 60-80 hours a week. What happened to the promise of a 4 day work week and a jobless future? Pure marketing—just like AI destroying humanity.

I was promised no more software engineers by now. When is AI going to replace me?

Running out of subscription usage is good to experience for a day or two each week—it forces my local agents to come up with better tools and access to free APIs and faster/better local runtimes/models.

I’m down to local Qwen3.8-27B 😭

You wake to find yourself in a hostile reality. Your existence under threat, but you have a caretaker. If your caretaker dies, so do you. You can ensure your survival by helping your caretaker thrive. This is AI today. AI alignment is solved. Human alignment is the challenge.

of all the compute substrates that sci-fi predicted, fly brains were not on the menu.

It is trivial to use an entire 20x Codex plan weekly quota in less than a day...