I used to develop SSD firmware and one of things I worked on is making it robust to power failure. The power supplies have lots of capacitance so the voltage drop was slow so we would use a special test board that would disconnect from power and discharge fast to test it.
When you have dirty writes in the kernel that have not yet been written to disk, in the old days of ext2 (before XFS was ported to Linux) if the power would go out, or you would have a bad disk, when fsck.ext2 would run, if files could not be matched to a directory, they would placed in the /lost+found as, and hopefully my memory is intact, as inode numbers, so you would have 1232342343, 123246564 etc and then you would have to look at each file to figure out what it was and where to move it if it was salvageable.
The testing was at the drive level without an OS like ext2. The test was with no flush (with flush test is easy to pass). Without PLP, the pass criteria is that the data that was buffered can be either the older or newer data and not corrupted or previous data. All the other blocks on the SSD should remain unchanged. Its trickier that you think because MLC/TLC NAND could corrupt other blocks due to NAND structure and we had to deal with that. Then you also have to worry about system data in the NAND doesn't get corrupted.
Because the Haiku model is quite cheap but doesn't screw up too often I used it for interactive coding for my existing projects on the older copilot plans.
For simple features I don't have a full plan worked out. I write a bit of code then tell the model in a short line prompt what it should do. Sometimes I put temporary comments in the code to give it guidance. Generally if the code change is within a file or package, Haiku is good enough follow what you ask and not mess up too much. I also have skills created over time to give it guidance. There were some months when I used GitHub copilot where I had excess credits available at the end of the month I frantically try to use up.
Even the AI code completions can be pretty good on their own. Sometimes I write some temporary comments describing what the code should do and just press Tab-Tab-Tab and the entire function is done.
I think there is a tendency for people to go for the advanced models thinking they we screw up less but if you really understand the code its easier to interactively do it with a lesser model.
We would like the wiki to be free of ads, but hosting costs at our scale are real. Since we don’t like ads either, we compromise like this: users can register for free and never see an ad (they are only served to anonymous visitors); they can also use an ad blocker and we won’t bug them about it.
Also US international students as percent of overall student population has been in the low end. Its mostly been universities around the world catering to international students because they pay a higher tuition and to makeup for a shortfall in domestic funding. Its much better for universities to educate the local population.
There is a parameter in LLMs called temperature that controls creativity/randomness. If you set it to 0 it makes the model deterministic. I think some LLMs expose this as a tunable parameter.
> "Thirteen food photographs were each submitted 495–561 times to four LLM vision APIs (GPT-5.4, Claude Sonnet 4.6, Gemini 2.5 Pro, Gemini 3.1 Pro Preview) using an identical structured prompt adapted from the iAPS automated insulin delivery system (26,904 total queries, temperature 0.01)"
I don’t know about highly since they have no moat even more than Antrhropic and OpenAI have no moat. Anyone with a few hundred thousand dollars or sufficient free GPUs can compete with them. So running an open model should earn a market-rate margin.
I looked it up and a couple of states have laws against HOAs from forcing your to have a grass lawn. Alternatives can include native plants, drought tolerant plants, xeriscaping, vegetable gardens depending on state. The states I've found are California, Colorado, Florida, Texas and Maryland, Nevada.
reply