> …in particular, no specific user data was accessed in order to solve this problem.

> Following an investigation, we have confirmed that Buckmaster’s Codex prompts over the two months preceding this announcement and paper on September 8, 2026, could not have influenced the system in any way, including through training.

I'm sure some people doubt OpenAI's claims, but Anthropic has also made advances in mathematics using AI. At some point you have to accept that not every single one of these breakthroughs is regurgitating a human insight.

If doubters refuse to pick specific cognitive tasks that they'd bet against AIs being able to accomplish in the next 4 years, I don't see how their claims are useful. If AIs are so limited in their capabilities, then clearly there must be some things they'll certainly be incapable of, at least in the short term.

1. https://openai.com/index/navier-stokes-solution/

The real answer is that we don't know if LLMs are conscious, and we don't really know how we'd that figure out. I guess if an AI wrote a philosophy paper on consciousness that had new insights, that might change some minds. But even that would fail to convince most people.

Scaling has produced novel capabilities with each larger model, and the rate of new capabilities doesn't seem to be slowing down yet. Even if you think the rate of improvements will slow down, that still means there will be significant improvements beyond what current models can do. Moore's law has slowed down, but modern computers are still much faster than ones from a decade ago. And unless you work at Anthropic or OpenAI, you don't know what the state-of-the-art is capable of. The most advanced publicly available models are months behind what AI labs have, and are deliberately limited to reduce liability.

Each Starlink V3 satellite weighs approximately 2 tons. They use ion thrusters for propulsion, so propellant on each satellite is 50-100kg, not half a ton. The current payload bay can hold around 60 V3 satellites, so they'd launch fewer of them (as they did in this launch) instead of putting partially fueled satellites into orbit.

A tower catch and an ocean splashdown require the same amount of propellant. The only difference is when the deorbit burn happens. Yes there will be some boil-off as the ship stays in orbit for longer, but a day in orbit would lose around 0.5% of the propellant. According to the live stream, both the booster & ship were loaded to about 98.5% of capacity. And filling the tanks completely would give them more payload capability, not less.

If you need to see a satellite deploy followed by a tower catch to be convinced of their payload numbers, then you're asking for evidence that doesn't exist yet. But I'd be happy to bet on that. How about this: If before December 31st, 2027, SpaceX deploys a payload greater than 50 tons from Starship while recovering both the booster & ship (I assume via tower catch. Fishing out of the water doesn't count.) you pay me $1,000. If that doesn't happen, I pay you $1,000. If you like, we could both choose a charity and the loser donates to the charity of the winner's choosing.

Also Starlink V3 orbits are at 330–370km, which is below pretty much everything else in space. Atmospheric drag is high enough that without ion thrusters, satellites will deorbit in a couple years. If SpaceX doesn't keep launching stuff, it will self-clean and allow others to use it.

1. https://www.electrifyamerica.com/plug-and-charge/

Also Nissan has instructed Leaf owners to not use DC fast charging, as doing so can cause battery fires. A recall has been issued, but no fix is available yet.[1]

1. https://static.nhtsa.gov/odi/rcl/2025/RCLRPT-25V655-5278.pdf

For example: Spokane to Seattle is about 230 miles in a straight line, or 280 miles driving. A flight takes less than an hour (not even 40 minutes of which is spent in the air) and costs $80-130 one way. A train traveling 200mph nonstop would take 1 hour 24 minutes. Building high speed rail costs $20-$200 million per mile (the upper end of that is California HSR). If we optimistically assume $20m/mile, constructing the line will cost $5.6b. If all SEA-GEG passengers (677k per year) choose to take rail instead of flying, and we want the project to pay for itself after 50 years, then each ticket will need to cost $165. This also assumes no cost to maintain or operate the rail line.

Even though my car has the older hardware, the self-driving is far beyond anything else I've used. It makes road trips far less tiring. I basically sit in the driver's seat while it drives from charging station to charging station. I've gone hundreds of miles without touching the steering wheel. Usually I intervene because I want to park in a specific spot, as the car tends to choose spots that are less convenient. Also it likes to park in my neighbor's driveway instead of mine. (The driveways are about 5 feet apart.)

The phone app lets you control all kinds of stuff remotely. I like to turn on the climate control before going to the car, that way my kid's car seat won't be uncomfortably warm on a sunny day. You can view the cameras from your phone, lock/unlock the car, open/close the windows, and open the trunk & frunk. And it works anywhere there's cell network connectivity, not just within direct radio range of the car.

I'd love it if more cars had these sorts of features, but I don't see it happening any time soon. Tesla seems to be a software company that happens to make cars, while everyone else is a car company that happens to make some software.

But even popular commuter routes like the 38 can get pretty dicey. One time a guy was blasting music out of a bluetooth speaker. He had headphones sitting on his neck, so I asked him to use his headphones instead. His response was to get in my face and start yelling, threatening to beat me up. Everyone around us stepped away and ignored us. Apparently the next stop was his, so he got off while threatening to jump me if he ever saw me again.

If you take public transit in the bay area regularly, you are almost guaranteed to have some very unpleasant experiences.

It seems like the best solution would be to implement registration fees to offset externalities like pollution and road wear, and have congestion fees for when the roads are busy (including for vehicles such as busses). Then let the price signal decide the optimal mix of vehicles on the road. I think if such a system were implemented, the biggest change in behavior would be people taking fewer car trips during peak traffic hours.

Also I'm not sure what the new law changes. Since 2016, Seattle has banned charging extra rent for having a pet. Subsection E states: "Other than the pet damage deposit authorized by subsection 7.24.038.A, the landlord may not charge the tenant any fee for keeping a pet."

1. https://library.municode.com/wa/seattle/codes/municipal_code...

The difference with LLMs is that they are a substitute for more general cognitive skills. Right now they are also a complement (using them increases the productivity of humans writing software), but it's only a matter of time before they improve to the point where they fully substitute. And it seems quite likely that the timeline for that is significantly shorter than my remaining career.

If you are confident that there will be specific tasks that LLMs can't accomplish in 2033, I'm happy to entertain such ideas. But considering the current rate of progress, I can't think of anything like that.