I'm not convinced we're going to get much further than this, at least without a see change in how this stuff is done. We saw a pretty big amount of diminishing returns with GPT-4, which cost 2 orders of magnitude more money to train than 3.5. Is the performance more than 2 orders of magnitude better than 3.5? How much compute is too much to be spending on this bullshit. 30% of humanity's total computing resources? 70…
The performance doesn't have to scale with the training cost. It's ok if we have diminishing returns. The only question is if we can keep getting improvements. If we could get a 10x improvement over GPT-4 (while maintaining safety/alignment), the potential benefits to productivity across the entire world economy are so great that it justifies nearly unlimited investment.
The reason why the LLM apparent world model should not be considered to be the same as a human's world model is because of the modality of learning. The world model we learn as we learn a language includes the world model embedded in language. But the human world model includes models embedded in flailing about limbs, the permanence of an object, sounds and smells associated with walking through the world. Now, all t…
That makes sense, but isn't this a matter of presenting it with more models? Maybe a physical model discovered via video or something like that? Then it will be similar to what babies are trained with, images and sound. Tactile and olfactory would be similar. By doing this you'd glue the words to sights, sounds, smells, etc. But it also seems like this is already someone has thought of and is being explored.
You are correct, there is active research on this. And words and pictures are associated in models like stable diffusion. There has been some success combining GANs and LLMs, but it is far from a solved problem. And as the training data gets more complex the required training resources increase too. Currently it's more like a confusing barrier than a happy extension of LLMs.
I don't think it's fair to include the billion years of human evolution as a cost on the human side of the chart but not include it in the AI side. AIs didn't evolve themselves out of the primordial ooze, they were built by humans and required herculean human effort to develop and improve. They stand on our shoulders, yet they're still in infancy when it comes to capability. I have yet to see any evidence of an LLM b…
>The best use of LLMs that I've seen so far is as a boilerplate-producing autocomplete system. Considering that we have better ways to automate this (better programming languages that can abstract away the boilerplate), this is not very high praise. I think it's in our nature as software people to look at their ability to work with code, but they're quite good when applied to general language tasks. I've been using t…
That's fair, although it gets complicated to work out numbers because we don't train many LLMs, whereas we're constantly training humans, each of whom cost the planet tons of CO2 emissions every year... and, of course, your point that LLMs just aren't very good yet. I fear that they're good enough (or appear to be to the layperson) that execs will replace customer support staff with them, even if the outcomes overall aren't as good.
I'm not as fearful. When customer support gets too expensive companies already outsource it to India. Indian customer support workers cost far less than Westerners in terms of energy and CO2 emissions, both in terms of training and ongoing costs. India is about 2T/person in CO2 emissions, compared to 15T for North Americans. And that number is averaged over the whole country. I would imagine the poorer areas of the country have much lower emissions and the bulk of CO2 comes from the wealthier big cities.
I'm getting suspicious that there is a bit of a blind spot in understanding the world and the usefulness of what we call intelligence, as Noval Yuah Harari says, intelligence is overrated. Look at what we've done to the planet and our environment we have fucked it properly, yet we consider ourselves to be intelligent? Could it be that intelligence is overrated and discovery of new ideas / thing is underrated? Our ego…
What we've done to the planet is perhaps less a consequence intelligence, more a consequence of multi-polar traps. Though it's in our collective self-interest to protect the planet, it's in our individual self interest to ignore the problem / prepare so that our own children can "weather the storm". It kinda depends on how much you care about people across the world and future generations. But yeah, if humans were mo…
All of this can be boiled down to the simple maxim: "Stupid is as stupid does."
> You would like a home from 1969. As I live in California, I've seen 'em. They're not all Eichlers. > You would like the salary saved from 1969. A tech worker makes more than 5 times the median 2023 personal income and their industry didn't exist then, so I mean, I wouldn't even if I could afford the house in Palo Alto. And you wouldn't want to say this to a racial minority or even someone in Appalachia.
You’ve kinda proven the point. The argument made was cherry picking. Here again - You’ve used the most highly paid subset of workers. In any other nation, and in other industries in America - Tech workers dont get paid that much.
There isn't a choice, the original post is incorrect. https://fred.stlouisfed.org/series/DSPIC96
That's a graph of disposable personal income, not purchasing power, and it doesn't account of the increased cost of several major spending categories. It's also the total disposable personal income for the entire country, and not median household income. See the "Units" field: it's in billions of chained 2012 dollars. It's from 2018, but this shows what's going on more clearly: https://www.pewresearch.org/short-reads…
> and it doesn't account of the increased cost of several major spending categories.
Inflation adjustment does cover some of that.
> It's from 2018, but this shows what's going on more clearly:
That's cost of employer healthcare benefits, isn't it? Which causes wages to be an increasingly smaller % of total compensation.
There has actually been research that found that there are strong diminishing returns in terms of at least expanding parameter sizes. While I think there are still breakthroughs to be made in terms of window sizes and workarounds like Mixture of Experts, I'm not sure how much farther we will get here in the long term in terms of raw performance of the LLM itself. FWIW, Sam Altman agrees and has a surprisingly similar…
The open source community has breakthrough after breakthrough lately. its absolutely stunning how fast it's advancing. I can run art models and llms on cpu/gpu now. I've tested out opensource models with quality better than chatgpt3.5turbo, and I can even fine tune them on my notes and books for better results. It's all so easy with so many one click installers now too! My husbands D&D group uses some AI for their ga…
> I can run art models and llms on cpu/gpu now. I've tested out opensource models with quality better than chatgpt3.5turbo, and I can even fine tune them on my notes and books for better results. It's all so easy with so many one click installers now too!
Do you have a good resource to find this stuff? I’m behind the times.
(Also, to be a pedant, most of that is inference and not training. But I can’t say much about fine tuning so I’m not really trying to argue against your point.)
This entire piece is based on one massive, unsupported assertion, which is that LLM progress will cease. Or, as the author puts it, "we are at the tail end of the first wave of large language model-based AI... [it] ends somewhere in the next year or two with the kinds of limits people are running up against." I want to know only one thing, which is what gives him the confidence necessary to say that. If that one stat…
I work in this space and think that it's far more rational to accept the axiom that LLM progress will not be significant than to bank product work on assuming it will increase drastically . I do think we still have yet to squeeze the most value out of current LLMs, but most people's radical AI dreams are completely out-of-touch with reality for anyone working closely on these problems. My biggest fear in this space i…
It may be rational and it may be irrational. All I'm saying is I want to see an argument. My impression of the AI space generally is that there are many, many obvious ideas which are simply waiting to be picked up off the ground and tested, and that it's not clear to anyone how much better (what are we even quantifying this with -- log loss?) the base LLM capabilities have to be before they're suitable for making tools that automate large quantities of work. Even if you assume that 240B is the limit for how many parameters number of engineers who have the ability to fine-tune GPT-4 or whatever Google is about to come out with is vanishingly small compared to the number of engineers who are participating in the open-source ML community, and the number of engineers in the open-source ML community is vanishingly small compared to the number of engineers in the long-tail of app developers who will ultimately adopt LLMs to their use-cases. Even assuming that GPT-4 is the best an LLM can possibly be, (which, again, I've seen no argument for), the widening of LLM availability and the building of practical tooling is a strong reason to believe that the utility of LLMs to concrete products will dramatically increase in the next 4-5 years.
Apparently GPT-4 is getting pretty good at knowing when it's wrong: https://thezvi.substack.com/p/ai-26-fine-tuning-time#%C2%A7g... (They asked GPT-3.5 and GPT-4 "are you sure" to see if it would change its answer, both when the original answer was right, and when it was wrong)
Does it do that because it can check it’s own reasoning? Or is it just doing so because OpenAI programmed it to not show alternative answers if the probability of the current answer being right is significantly higher than the alternatives?
I don't know. I don't think anyone is directly programming GPT-4 to behave in any way, they're just training it to give the responses they want, and it learns. Something inside it seems to be figuring out some way of representing confidence in its own answers, and reacting in the appropriate way, or perhaps it is checking its own reasoning. I don't think anyone really knows at this point.
Humans aren't required to use clean room design. Using copyrighted materials as inspiration/reference is not uncommon or illegal.
If you spend weeks drilling flash cards on copyrighted code, then produced pages of near-verbatim copies with copyright stripped, any court would find you to have violated the copyright. A lot of people right now are banking on "it's not illegal when AI does it", and part of that strategy is to make "AI" out to be something more than it is. That strategy has many parallels to cryptocurrency hyping.
'near-verbatim' is the key.
Though it'd be hard to copyright any code small enough to fit on a flash card.