The LLMs i have been using clearly demonstrate some expert level capabilities. The problem is that they also fail some basic elementary school tasks once in a while. Reviewing the entirety of their voluminous output to screen those out is not humanly feasible.

So, "People With Disabilities Don't Exist" then?

My father was recently paralyzed by Guillain-Barré, so I'll let him know, thanks.

I was feeding some graphs of my PV production to claude.ai . There was clipping - a flat line near the top - and i was trying to get it to tell me how much more production i would get if the inverter was bigger. It kept insisting there was no flat line or clipping going on, making up the peak number. Confidently. Even after multiple prompts, including some as "are you blind". There are things these tools are just terrible at. Reading graphs is one of them, apparently. I also tried to get it to create a diagram of my PV installation. It did 99% of the job, but just couldn't make very minor formatting corrections - preventing the boxes from overlapping. It tried to regenerate the graph from scratch, but made my unwanted changes in the process.

I have similar issues when using various code generator agents. They struggle enormously with making isolated minor fixes.

It's almost as if we need some non-generative AI to actually get the job done.

It is chatgpt codex with gpt 5.5 cli.
Its reasoning is pretty good. Makes far fewer mistakes than the corresponding AI chatbot, which causes me a lot of facepalms.

Is still needed in 2026 to read this post.

Oh god. If I spent enough time digging through my ancient Slashdot posts, somewhere back there there are posts of me going, "While I loved the strategy behind Falcon 9, I'm really not keen on this plan to make Starship out of huge carbon fibre tanks, that sounds like a really failure-prone solution..." I'm glad they only spent like a year on that idea before deciding it was dumb; somewhere back there there's also a bunch of posts of me cheering their switch to steel ;) . SpaceX still keep having random COPV problems (most of which they don't even make themselves). Not too encouraging for the notion of the cold gas thruster add-on to the Roadster, where the plan is to replace the back seat with COPVs, so you have a COPV right behind your head.

Electron has been getting by on CF, and honestly I'm impressed, but they've also been only working with very small launch vehicles thusfar. We'll see how neutron goes...

False equivalency.

The options they may have for evading wealth tax would depend on how the statutes get written, and what loopholes they can buy.

Don't get me wrong, there's a lot to say about printing small rocket parts, such as for the engines. But they were printing basically sheet metal cylinders, which is such an immensely slow and inefficient way to go about it, and it left them with parts that were heavier and less aerodynamic (rougher surface). Crazy that idea ever got any funding.

The one seen over Moscow might have been, with a bit more thrust...

"SapceX has got to be a huge scam too" - SpaceX launches the vast majority of the world's commercial cargo to orbit. The Falcon 9 FT has the highest success rate of any rocket with a statistically significant number of launches under its belt, and is dirt cheap. SpaceX's core operations are roughly breakeven, but that's including subsidizing the development of Starship. Starlink is a money printer.

There are lots of things sketchy about the SpaceX IPO, to say the least, but SpaceX, as a company, has been extremely successful with rocketry.

MidJourney was the first sizable AI company to become profitable, having done so back in 2022.

The funny thing was that I knew him for like six months online before I realized he was fully paralyzed. He's been covered in the Finnish press a number of times. Amazing guy. Up until recently he was living in a house he built himself before ALS struck, but the medical service decided he was too far away and he had to move closer. You lose a lot of control over your life with ALS.

He wrote a book about nuclear safety engineering recently, which is a fascinating read, and which I strongly recommend.

It very much does learn even within its local session. For example, I have been teaching it to ssh into my raspberry pi hooked up to the carrier hvac bus via rs485 for the last few months for reverse engineering purposes. I iterate every week or so with new tracesx making it updtae the code, and run the nee version. The additional prompts I type eerkpy are very short, as it has all the conversation history. Even pre migration from wsl to lxc. Sometimes it does forget certain detailsx like any LLM. I do not mind if the convo gets used for public training, as this is open source code. There may be an opt out. I'm not sure i would trust cloud LLM if this was proprietary, though.

BTW, AI is a godsend for ALS patients. Even with an eye tracker, your writing throughput is low / tiring compared to typing, and of course you can't do anything that requires physical activity (painting, playing an instrument, speaking, etc). AI tools help fill the gaps.