Technology

I've been hearing about the imminent crash for the last two years. New money keeps getting injected into the system. The bubble can't deflate while both the public and private sector have an unlimited lung capacity to keep puffing into it. FFS, bitcoin is on a tear right now, just because Trump won the election.

This bullshit isn't going away. Its only going to get forced down our throats harder and harder, until we swallow or choke on it.

Just wait for the 5090 prices...

Sorry, crypto is back in season.

nope, if normal gamers are already willing to pay that price, no reason for nvidia to reduce them.

There's more 4090 on steam than any AMD dedicated GPU, there's no competition

Other than with language models, this has already happened: Take a look at apps such as Merlin Bird ID (identifies birds fairly well by sound and somewhat okay visually), WhoBird (identifies birds by sound, ) Seek (visually identifies plants, fungi, insects, and animals). All of them work offline. IMO these are much better uses of ML than spammer-friendly text generation.

This has already started to happen. The new llama3.2 model is only 3.7GB and it WAAAAY faster than anything else. It can thow a wall of text at you in just a couple of seconds. You're still not running it on $20 hardware, but you no longer need a 3090 to have something useful.

Well, you see, that's the really hard part of LLMs. Getting good results is a direct function of the size of the model. The bigger the model, the more effective it can be at its task. However, there's something called compute efficient frontier (technical but neatly explained video about it). Basically you can't make a model more effective at their computations beyond said linear boundary for any given size. The only way to make a model better, is to make it larger (what most mega corps have been doing) or radically change the algorithms and method underlying the model. But the latter has been proving to be extraordinarily hard. Mostly because to understand what is going on inside the model you need to think in rather abstract and esoteric mathematical principles that bend your mind backwards. You can compress an already trained model to run on smaller hardware. But to train them, you still need the humongously large datasets and power hungry processing. This is compounded by the fact that larger and larger models are ever more expensive while providing rapidly diminishing returns. Oh, and we are quickly running out of quality usable data, so shoveling more data after a certain point starts to actually provide worse results unless you dedicate thousands of hours of human labor producing, collecting and cleaning the new data. That's all even before you have to address data poisoning, where previously LLM generated data is fed back to train a model but it is very hard to prevent it from devolving into incoherence after a couple of generations.

I think it's more likely a compound sigmoid (don't Google that). LLMs are composed of distinct technologies working together. As we've reached the inflection point of the scaling for one, we've pivoted implementations to get back on track. Notably, context windows are no longer an issue. But the most recent pivot came just this week, allowing for a huge jump in performance. There are more promising stepping stones coming into view. Is the exponential curve just a series of sigmoids stacked too close together? In any case, the article's correct - just adding more compute to the same exact implementation hasn't enabled scaling exponentially.

You can install Ollama in a docker container and use that to install models to run locally. Some are really small and still pretty effective, like Llama 3.2 is only 3B and some are as little as 1B. It can be accessed through the terminal or you can use something like OpenWeb UI to have a more "ChatGPT" like interface.

The market can remain irrational longer than you can remain solvent.

A. Gary Shilling

Not copilot, but I run into a fourth problem:
4. The LLM gets hung up on insisting that a newer feature of the language I'm using is wrong and keeps focusing on "fixing" it, even though it has access to the newest correct specifications where the feature is explicitly defined and explained.