As everyone is aware, the world is still going nuts trying to establish more, newer and better AI tools. Mainly by tossing absurd quantities of money at the problem. A number of those billions go towards building low-cost or totally free services that operate at a significant loss. The tech giants that run them all are hoping to bring in as many users as possible, so that they can catch the market, and become the dominant or only party that can provide them. It is the timeless Silicon Valley playbook. Once supremacy is reached, expect the enshittification to begin.

A likely method to make back all that money for establishing these LLMs will be by tweaking their outputs to the preference of whoever pays the many. An example of what that such tweaking appears like is the refusal of DeepSeek’s R1 to discuss what happened at Tiananmen Square in 1989. That a person is certainly politically motivated, but ad-funded services won’t precisely be enjoyable either. In the future, I totally anticipate to be able to have a frank and truthful discussion about the Tiananmen occasions with an American AI representative, however the just one I can afford will have presumed the persona of Father Christmas who, while holding a can of Coca-Cola, will intersperse the recounting of the terrible occasions with a cheerful “Ho ho ho … Didn’t you know? The holidays are coming!”

Or maybe that is too improbable. Right now, dispite all that money, the most popular service for code completion still has trouble dealing with a number of easy words, in spite of them being present in every dictionary. There should be a bug in the “complimentary speech”, or something.

But there is hope. One of the tricks of an approaching gamer to shock the market, is to damage the incumbents by launching their design for free, under a liberal license. This is what DeepSeek just made with their DeepSeek-R1. Google did it earlier with the Gemma designs, as did Meta with Llama. We can download these models ourselves and run them on our own hardware. Even better, people can take these models and scrub the predispositions from them. And we can download those scrubbed designs and run those on our own hardware. And after that we can lastly have some really beneficial LLMs.

That hardware can be a difficulty, though. There are two choices to select from if you wish to run an LLM locally. You can get a big, effective video card from Nvidia, or you can purchase an Apple. Either is expensive. The main spec that shows how well an LLM will perform is the quantity of memory available. VRAM when it comes to GPU’s, normal RAM in the case of Apples. Bigger is better here. More RAM suggests bigger designs, which will drastically enhance the quality of the output. Personally, I ’d state one needs at least over 24GB to be able to run anything helpful. That will fit a 32 billion parameter model with a little headroom to spare. Building, or purchasing, a workstation that is geared up to deal with that can quickly cost countless euros.

So what to do, if you don’t have that quantity of money to spare? You buy second-hand! This is a practical alternative, however as constantly, [mariskamast.net](http://mariskamast.net:/smf/index.php?action=profile