• 1 Post
  • 182 Comments
Joined 2 years ago
cake
Cake day: June 12th, 2023

help-circle




  • You can’t make an LLM only reference the data it’s summarising. Everything an LLM outputs is a collage of text and patterns from its original training data, and it’s choosing whatever piece of that data seems most likely given the existing text in its context window. If there’s not a huge corpus of training data, it won’t have a model of English and won’t know how to summarise text, and even restricting the training data to medical notes will stop mean it’s potentially going to hallucinate something from someone else’s medical notes that’s commonly associated with things in the current patient’s notes, or it’s going to potentially leave out something from the current patient’s notes that’s very rare or totally absent from its training data.





  • AnyOldName3@lemmy.worldtolinuxmemes@lemmy.worldwindows update
    link
    fedilink
    arrow-up
    10
    arrow-down
    3
    ·
    1 month ago

    Variations of this meme get posted every week, but I’ve never experienced it, despite having had tens of grub updates murder-suicide the Windows boot loader and grub itself across five or six different machines. Thankfully, it’s pretty easy to rebuild a Windows boot partition, but the frequency that I’m hit with this problem is one of the major reasons I avoid using Linux. Eventually I’m going to have to switch, but that’s driven mainly by Windows getting worse rather than any of the pain points I’ve had when trying to switch full time in the past having been fixed.


  • It depends on the kind of acceptance. If you accept that certain things will be harder for some people and make reasonable accommodations so they can get on with their lives, then people can get on with their lives. If you accept things will be harder so use that as an excuse for people never doing anything without removing any of the obstacles stopping them doing things, they’ll never get anything done. It’s really just addressing the problems versus deciding the problems are inevitable and giving up. That said, giving up can be a lot less miserable than refusing to acknowledge problems and yelling at people when they don’t keep up.


  • Even accounting for disasters, coal power puts more radionuclides into the environment into the environment than nuclear for the same amount of energy. If you dig lots of stuff up and spew it all into the air, the small amount of radioactive material that’s in coal and the rocks around it is much bigger than the tiny amounts of nuclear fuel a nuclear power plant gets through. If the only concern is radiation that persists on geological timescales, then swapping all coal for nuclear is an improvement. Other things that release surprising amounts of radiation include making things out of granite (it’s usually got uranium in) and importing bananas and Brazil nuts.

    If it’s changes on a geological timescale in general, then as fossil fuels form on a geological timescale (the clue’s in the name), digging them up is going to take unfathomable amounts of time to undo. It won’t even be as quick as the first time around, as most coal formed before ligninase evolved, so trees fell over and didn’t rot and usually became coal, buy they’re biodegradable now so need specific fossilisation-friendly conditions to become coal.


  • That problem isn’t unique to nuclear. It wouldn’t be newsworthy if a worn-out wind turbine blade was incinerated unsafely, and that’s something that happens routinely and is much more damaging than dumping this quantity (about a beer crate full) of depleted uranium. The reason we’re hearing about this incident is that the nuclear industry is held to a much higher standard than anything else in the energy sector. There are good reasons for that - the worst case scenario for a single fuckup is much worse - but a lot of it is just fear-mongering by fossil fuel companies who needed to lie to make something seem more dangerous than what they do, even before climate change was recognised.

    Nuclear is so much better than fossil fuels that even if we cut every corner and accepted a Chernobyl-scale indicent would happen a couple of times a year, it’d still be preferable over the gradual phase out of fossil fuels and resulting climate crisis we’re on track for.




  • Thermal problems are much less likely to kill hardware than they used to be. CPU manufacturers have got much better at avoiding microfractures caused by thermal stress (e.g. by making sure that everything in the CPU expands at the same rate when heated) and failures from electromigration (where the atoms in the CPU move because of applied voltage and stop being parts of transistors and traces, which happens faster at higher temperatures). Ten or twenty years ago, it was really bad for chips to swing between low and high temperatures a lot due to thermal stress, and bad for them to stay at above 90°C for a long time due to electromigration, but now heat makes so little difference that modern CPUs dynamically adjust their frequency to stay between 99.0° and 99.9° under load by default. The main benefit of extra cooling these days is that you can stay at a higher frequency for longer without exceeding the temperature limit, so get better average performance, but unless your cooling solution is seriously overspecced, the CPU will be above 99.0° under load a lot of the time either way and the motherboard just won’t ramp the fan up to maximum.


  • The last time they had plenty of stock and cards people wanted to buy at the same time was the RX 200 series. They sold lots of cards, but part of the reason people wanted them was because they were priced fairly low because the cards were sold with low margins, so they didn’t make a huge amount of money, helping to subsidise their CPU division when it was making a loss, but not more.

    Shortly after this generation launched Litecoin ASIC mining hardware became available, so suddenly the used market was flooded with these current-generation cards, making it make little sense to buy a new one for RRP, so towards the end of the generation, the cards were sold new at a loss just to make space. That meant they needed to release the next generation cards to convince people to buy them, but as they were just a refresh generation (basically the same GPUs but clocked higher and with lower model numbers with only the top-end Fury card being new silicon) it was hard to sell 300-series cards when they cost more than equivalent 200-series ones.

    That meant they had less money to develop Polaris and Vegas than they wanted, so they ended up delayed. Polaris sold okay, but was only available as low-margin low-end cards, so didn’t make a huge amount of money. Vega ended up delayed by so long that Nvidia got an entire extra generation out, so AMD’s GTX 980 competitor ended up being an ineffective GTX 1070 competitor, and had to be sold for much less than planned, so again, didn’t make much money.

    That problem compounded for years until Nvidia ran into their own problems recently.

    It’s not unreasonable to claim that AMD graphics cards being in stock at the wrong time caused them a decade of problems.