Ask Lemmy

I view "ai" (which I will always think of as a misnomer) as a bastardization of machine learning research that is positively contributing to the world and the literature as opposed to whatever the fuck the "AI" companies are doing.

Realistically, it's just media visibility. People were doing ML research for ages before LLMs became 'the next big thing.' The things developed by that research are often incredible but also incredibly narrow. ML based protein folding systems were big news among nerds several years ago. The amount you have to explain to even have the groundwork to explain how important that development is could never fit in a clickbait headline or article. You know what does fit in a clickbait title? 'Celebrity ignoramus talks to chatbot, decides it's God.' 'AI company insider says their product is the best thing since sliced bread and everyone who doesn't pay them money will commit suicide.' and 'You are doomed because of AI. Click here to find out why.' No one outside of the field those narrow ML systems are built for can understand their output. Anyone can understand 'You're absolutely right!'

This already exists as 'Mixture Of Experts' (MOE) models. It's how Deepseek is able to remain competitive with a fraction of the budget.

Accessibility to profit.

ML is more difficult, even if not impossible, to package in a way that is easily consumable by those with the capacity to pay for services.

First thing is that the market is pretty young. Imagine you are now using IBM mainframe PC to run chatgpt and not even windows1.0 on your IBM pc.

There are small AI models. Small agentic capable AI models that can do tool calling. GenAI and what you described. One thing is some solutions are mathematically impossible to gain certain quality. Other thing is they're just not owned by billionaires and not used by corporations so press don't cover them and nobody cares except few enthusiasts and people that need such capabilities and know limitations. Press only covers what they're paid for or what can gain some traction.

The hate on AI as you described doesn't help AI as a science domain. It's not really related to science but related to ineffective spending of money by billionaires.

That's the Unix philosophy you are describing, not Linux. Linux does not follow that, it's just a common practice among programmers. See linux is a monolithic kernel containing all drivers for all hardware, newer example is systemd

Well, mainly because I don't need an LLM to tell me about history in a way that I would need to go verify. But ML is useful in, e.g. Home Assistant, with many speech-to-text models, as well as text-to-speech models, to choose from. You can hook those up to an LLM, so your voice assistant can understand vague or complex intents. There's genAI image enhancement and painting tools available in Krita. There's object recognition in Frigate, and Immich. Et cetera.

In short, the Unix philosophy in AI is what the FOSS community is already doing. It supports the things we want to do, in the background, by doing one thing and doing it well. It just doesn't look like Copilot, with its single text input which takes over everything. (Which is what we hate because it's bad.)

I mean people are. All the time. They just don't get the attention things like lmk have.

For example, SAM 3 exactly what I think you are asking, but for images.

But there is another point in here about how"actually, just bigger model better" and that's the thing with transformers. Them basically becoming chat bots through clever training and massive size and training datasets wasnt expected. You don't get that behavior from much smaller transformers. And so there was an apparently emergent phenomena in this case. A small network isn't going to do what you think it is going to do precisely because it's over constrained.

There are. You just need to know more about ai.

Note ML is a tech stack and AI is now a marketing term. It used to be a field of study and still is but the marketing term is so widely used its reduced the meaning of the original term. Kind of how 5G is a tech spec but then companies kept labeling their networks 5G before the 5G spec was certified. Then when the 5G spec came out it created even more confusion.

There are plenty of small models that you can run locally, and they can be fine tuned on different types of content.

Back in around 2018 (2019?) it was all the rage, to try to "help" neural networks by smartly selecting the best network tp use, like you describe. Turns out it was just better throwing it all in one net, it makes a better selector too...

Then I have a question, what would you use it for?

If you'd like to analogize with special-purpose programs versus general purpose ones, you could consider an LLM to be more like an "operating system" rather than a single application running on that system. The LLM isn't specialized in a particular task, it supports running any task you want to throw at it.

In this analogy, the specialized applications that are running on the LLM would be the things called "agents" and "harnesses." They're the parts that hold the code that is tailored specifically to the particular task that they're for. So if I wanted to set up a system using an LLM to, for example, read court transcripts and automatically search legal databases for statutes and case law relevant to whatever's being discussed in them, I wouldn't simply copy and paste the transcript into the LLM's context and expect something useful to come back. I'd have to embed the LLM inside a system that prompts it correctly, has tools that can search legal databases, mechanisms for storing intermediate results, scripts to check the formatting of inputs and outputs, and so forth. None of that accessory stuff needs to be an ML model, it can just be conventional programming. Trying to train a specialized ML model from the ground up to do all that stuff without the associated harness helping it would be hugely wasteful.

I suppose another analogy could be the hardware itself. For every task that we use general-purpose von Neumann architecture CPUs for we could create a specialized chip on a purpose-specific circuit board. But instead for most tasks we find it's much cheaper and more convenient to create general-purpose computers and then program them for these specific tasks. Nowadays you'll often find that simple home appliances with just a few buttons and a few functions will have a full blown microcontroller inside them with firmware. Probably lots of unused inputs/outputs and ram and whatnot. That's because it's far cheaper and easier to build a factory that stamps out millions of general-purpose programmable chips than it would be to have hundreds of factory lines that each do a run of ten thousand custom-designed chips. It's genuinely less wasteful doing it that way.

For how LLM specifically struggle with that and what is done about it s white nice article from 2023 about MOE models:

https://huggingface.co/blog/moe#what-is-a-mixture-of-experts-moe

They explain the issue with "you need size but want specialization" quite well in my opinion.

They kinda are. they hit a wall and started making specialist llms that then get queried by the main one. even have things where the main one evaluates repsonses from a few. im not 100% sure what all they are doing but its not monolithic.

There are examples like Spleeter, it separates music into individual tracks (drums, vocals, piano, guitar etc) Based on its training model.

But for marketing a paid product people expect one thing does it all.

For sys admin tasks LLMs already have a tendency to write a script to carry out what is asked of it. We lean into that for exactly the reasons you give

Ai is machine learning. They just rebranded it.

Llms are most accessible to the general public. The more accessible it is the more marketable it is. Plus it feeds into the dream of talking to the computer, it understanding, and doing the tsk for you.

There’s a ton of less sexy AI/ML out there that works exactly how you’re describing. You just don’t see it because it’s a specialized tool.

If you ask an llm to edit a photo for you it processes that request. it over to the photo editing AI to do the task. When it’s done the llm hands it back to you. That’s probably the most consumer facing example of something like that where two different AIs work together for your task.

Is that what mcp is? For tool integration, an mcp connection allows the LLM to ask a tool questions. Every tool vendor is integrating an mcp server.

For example if I ask an LLM about a bug report it may have no way of figuring it out. But since I configured an mcp service to our ticketing system, my LLM can ask that ticketing system about the bug and get an exact answer