@thejessiekirk @nixCraft

>

If they do, the index funds will buy them at a very high price, flooding them with old people's money. Musk & Co can then sell their stocks and make obscene amounts of money. If the company turns out later to be unstable, the retirement fund will lose a ton of money, and index funds will no longer be a safe way to invest long term.

@thejessiekirk @nixCraft

>

There are rules on what companies can get on such a list, especially right after going public, so that people can ensure the company is stable. Some of those rules have changed lately, and SpaceX is hoping to get on those lists while they are still valued super high.

>

@thejessiekirk @nixCraft lots of big money managers, like pension funds, put their money in something called "index funds", usually run by banks or investment companies. Index funds buy the best performing stocks on the stock market automatically, cause that's historically been a safe way to invest long term.

The way they find out what stocks perform best is through stock market indices. Those are lists run by companies like S&P.

>

@pbloem @mndflayr @ZachWeinersmith but haven't you just added a special case on top of the same tool then? "when you have X amount of possibilities in probability range Y, say you don't know".

I know this is mastodon, but I'm not trying to gotcha you on AI here, I'm genuinely curious if there's some aspect of fine-tuning I'm missing that completely transforms the technology. To me, it just looks like a growing list of special cases.

@pbloem @mndflayr @ZachWeinersmith e.g. if I took an old T9 program, and I changed weights around to make some common suggestions more likely, that wouldn't be a qualitative change, it would just be a better autocomplete.

I don't know what the equivalent of reasoning traces would be >_>. Maybe changing weights based on the rest of the message 🤷

@pbloem @mndflayr @ZachWeinersmith I don't think I understand what it is about fine-tuning or reasoning traces (or the thread you linked) that turns an LLM from (what you seem to agree is) adhering to the "autocomplete mental model", to being something else.

I could understand if you argued that LLMs are a different category of tools than autocomplete altogether, but why would training them to be more user-friendly, or running them repeatedly, constitute a qualitative change?

@angusm @mbpaz @Illuminatus @jwz @pluralistic I just remembered that the Swedish word for "halter" (the straps you put on a horse or cow's face to attach a leash or reins to) is "grimma", and in Old Swedish "gríma".

I can't find any statements from Tolkien on the subject, and the general consensus seems to be that it's derived from the broader Old English / Norse word. But Tolkien did have a lot of strong opinions on the first Swedish translation of LotR (he hated it) and it does fit better >_>