> How do you figure? 20 dollars/month is insanely cheap for what OpenAI/Anthropic/Google offer. That absolutely qualifies as "empowering a common person".
This must be sarcasm. This has to be.
1,951 karma · joined August 20, 2019
This must be sarcasm. This has to be.
To many interested in actual AI research, LLMs are known as the very flawed and limiting technique they are, and the increasing narrative disconnect between this and the table stakes where they are front and center of every AI shop, carrying a big chunk of the global GDP on its back, is annoying and borderline scary.
Perhaps. Though we have empirical evidence of how much we can quantize and distillate models to the point of practical uselessness. That sets a bar for how large a local model needs to be for general-use as to compete with the could ones. We are talking in the area of 60GB for GPT-OSS/Qwen3.5, which is what enthusiasts are running on 32GB DDR5 + 24GB VRAM RTX 3090.
> As if consumer hardware won't get more powerful
Now I will let you, with that last fact in hand, plot a chart of how much it's been costing to provision that over the past 2 years and use it to prove me wrong about the affordability of local models.
I cannot remember when I last had a network, bluetooth, audio or driver issue with Linux on my ThinkPad. I legitimately cannot. But I can assure you I that had 3 of those in the last couple weeks alone on my work's HP elitebook with W11 24H2.
- I shouldn't be paying more for my next CPU because it has a NPU that I won't ever use. Give me the freedom of choice.
- Given that freedom of choice, it would seem that a majority would opt-out (as seen recently by Dell), so the morals of all that are dubious.
- NPUs may not be completely stupid as a concept, in theory, but at this point in time they are proprietary black-boxes purpose-built for marketing and micro-benchmarks. Give me something more general-purpose and open, and I will change my mind
- …but the problem is, you can only build so much general-purpose computing in bespoke processor. That's kind of its defining trait. So I won't hold my breath.
- Re: local-inference for the masses, putting aside the NPU shortcomings from above: how large do you think a LLM needs to be so it's deemed useful by your average laptop user? How would the inference story be like, in your honest opinion (in terms of downloading the model, loading it in memory, roundtrip times)? And how often would the user realistically want to suffer through all that, versus, just hopping to ${favorlite_llm.ai} from their browser?
Anyhow, if that makes me "antiai", please, sign me up!
Your true believer convictions don't matter here. Those AI accelerators are merely just marketing stunts. They won't help your local inference because they are not general purpose enough for that, they are too weak to be impactful, most people won't ever run local inference because it sucks and is a resource hog most can't afford, and it goes against the interests of those scammy unprofitable corporations who are selling us LLMs as AI as the silver bullet to every problem and got us there in the first place (they are already successful in that, by making computing unaffordable). There's little to no economical and functional meaning to those NPUs.
I don't like this manner of baselessly singling me out. Once again, there is no "my definition" of AI. Here is the one from wikipedia, that matches the academic definition with which I am aligned "It is a field of research in computer science that develops and studies methods and software that enable machines to perceive their environment and use learning and intelligence to take actions that maximize their chances of achieving defined goals."
My whole point is that the current AI-craze boils down to one single highly-specific method that got added the AI toolbox. It has clear shortcomings and limitations that major tech corporations hide or disguise knowingly, for their profit and our own deception. My take is that we collectively benefit from calling the current breed of AI products what they are: LLMs. For transparency's, correctness' and honesty toward consumers' sakes. The marketing around those products needs to be regulated accordingly.
Now I want to elaborate on why the commonly used taxonomy of AI, as formalised by scientists, or, what you call "my definition" is the only one that matters (with the practical consequence that nobody in the field uses the term "AI" without more qualifiers). Say, you develop a program to play tic-tac-toe. It is something whose complexity is low-enough that it has an analytical solution, with all the states of the game that can be stored and look-up. It could also be trained via convolutional neural networks, a family within machine learning, within AI, on lots of games. Would you call the resulting capability of playing tic-tac-toe "AI"? Theoretically yes, you must (and myself as well). Where we differ is in the importance of specifying how this ability is accomplished. The analytical implementation will be fully explainable, while the neural-network won't. In other words, we won't be able to tell whether the answer from the latter is faithful and desired, or subject to what we nowadays call "hallucination" in a turn of the tech giants to further anthropomorphise LLMs.
Now, maybe more about where I'm coming from. I researched computer vision algorithms for self-driving vehicles for several years. The trolley dilemma and other considerations about what it means for machines to behave "morally" has been ingrained through my formative and academic years. As Engineers and machine learning scientists, we have been evolving our field from predictable analytical methods to new ones with better results but less introspectability. I'm of the standpoint that this is only fine as long as the society is educated on those matters and willingly commits to using those methods with there trade-offs, for about the same reasons we put labels on food products and list known side-effects and their occurring changes in our medication. Because now those computational methods affect us in the real world in very material and tangible manners.
> Seems pretty intelligent to me.
Convenient? Yes. Intelligent? I mean, you can redefine AI to be whatever you want by lowering this bar however low you want. LLMs gave us slightly more convincing chatbots than prev-gen and nobody then would call them intelligent. The only reason we do now is marketing. It's laughable to me that we are still calling that something that's so obviously unable to push back on trivially impossible requests.
> Of course, not long ago we were using the term largely for things that were basically big chains of conditional statements.
No, we weren't? What are you even talking about? We had already built large-enough artificial neural networks whose outcomes couldn't be trivially explained in the 70's. Nobody with a sane mind and an understanding of what they were would call them oracles or intelligent devices. That's where the true genius of Altman lies.
In the academic fields where this taxonomy matters, nothing much has changed with LLMs, or not more than with DNNs, SVMs, etc. Nobody that's been involved in ML research for more than 5 years seriously thinks "job's done pals, we got to pack it up, after 70 or so years of effort, we've finally figured it out, and that AI we were looking for, we got it".
And no, that doesn't make me some kind of "AI hater", or someone unable to see value in LLMs.
The fact that it crashed on me for everything and nothing all at once seemed a bigger problem than "complexity".
This is obviously a big, unanswered, issue. It's pretty clear to me that we are collectively incentivised to pollute the well, and that it happens for long-enough for everything to become "compromised". That's essentially abandoning opensource and IP licensing at large, taking us to an unchartered era where intellectual works become the protected property of nobody.
I see chatbots having less an impact on our societies than the above, and interestingly it has little to do with technology.