Comments URL: https://news.ycombinator.com/item?id=49382665

Points: 2

# Comments: 0

Some examples of why I think this is really hard: Say I build a story generation system. I work hard on building an agent swarm of actors, critics, editors, researchers. I craft into the various agents concepts of story arcs, outlining techniques, character development. I build a huge well thought out process for how to agentically write an actually good story, so long as you give it a title. Heck, I even design and train my own custom LLM with original layer ideas and novel training techniques to use on this system. After all that I then take that final step and give it a title. Do I have no claim to that? I probably put more work and creativity into it than an author would have a book. What if I then gave it 500 titles? 5,000? Would my claim degrade the more titles I fed it? Is it a percentage of work question? What is the core concept here that defines 'human-centric'? What is the cut-off here?

Let's go even further. I don't prompt. I live in a world with unlimited context models. I have a conversation about the book I want it to write. During that process I reject some ideas and accept others. I didn't give it a 'system prompt' but essentially all I did was prompt it and select versions I liked. Is that not human centric? How about if I asked it for advice and it did some editing work on my story? Did that make it not human centric even though the starting text was mine? What if that starting text was 99% replaced with a version 10x as verbose. Defining based on how you interacted with the model (prompted and selected) just seems way to weak to be a clear test.

Edit: should have said 'poor tax'

Reality should always have a place in the law.

I think there is a real point people are missing here. Memory is going sky high because most of it is being shifted to GPU specific ram for server build out. This means that when this ram hits EOL it -won't- flood the consumer market in a meaningful way. It is disposed of because most people won't put a server GPU into anything. It is like a used big rig, built for purpose and when it reaches end of life it is truly at end of life. Consumer DRAM will stay very high while the cartel retools and moves back into that space. This means that the cartel wins and pays no real price for it when they shift back to producing consumer ram, they step immediately into a massive demand for pc RAM and keep raking in $$$. But they all have to do it in lockstep so that they have a market to come back to. Otherwise someone would become the king of PC ram and they couldn't re-enter easily if/when the AI server market fails. Being in lock-step means they rake in more money and mitigate the risk of an AI bubble, all at the cost of the consumer.

At least that is what this looks like to me. They know they could make a lot of money ignoring the AI server build out but they all magically didn't. A competitive market would have seen a spike in price when this much demand came in, but the lockstep movement of the big three shows that they are coordinated and not competing.

[1] https://www.youtube.com/watch?v=f0gN1x6sVTc

- We are conditioned to expect that the price listed is not the actual final price. If I see 1 euro as a price in Italy, I can hand them 1 euro and walk away. In the US if I see $1 as the price I could get a bill for??? 1.20? then tip? 1.40? Some random magic number I have no idea how it happened? The price listed should be the price paid. Period. Stop pretending taxes aren't part of the price.

- We have legislation that makes tips part of expected compensation. Minimum wage is different for servers. Get rid of any legislation like that. Every job should pay a living wage (including benefits) or shouldn't exist. If that were in place you would see a lot fewer places begging for tips since the cost of things would be higher to match the salaries they would need to pay. So right now if you don't tip you are actually taking away someone's paycheck and that is wrong on all levels.

Of course all of this is also part of the American psyche. We really like being lied to about price. Show us a lower number, even though we know it is fake and we will end up being charged way more by the time the transaction is done, and we will hop there. Take the taxes off the price and don't include the cost of labor on the listed price and we go 'Look! A bargain!'

I appreciate the factual nature of this story but I think an important aspect of journalism is giving people relatable references that allow them to better understand the information presented. It would be helpful if a story like this gave that context because without it it is hard to understand if this is 'Wow! What a close one!' or 'I guess this is a little interesting?'.

[1] https://www.science.org/content/article/milk-jug-sized-detec...

[2] https://arxiv.org/abs/1708.01294

This has a lot to do with how MCTS works BTW. The current best path is often only the current best path because a lot of investment has been sunk into it. If you were to put equal resources into a different path you may find that it was actually far better. It is just that the early rollouts favored the other 'best path' so you sunk a lot of resources into that one. We are very early in our exploration of LLM architecture. I highly doubt we are anywhere near the best path right now.

I think when successful, optimization really just means 'finally built right' and people forget the ridiculously inefficient ways before.

[1] https://en.wikipedia.org/wiki/List_of_countries_with_annual_...

[2] https://worldpopulationreview.com/country-rankings/police-ki...

Transportation is going through many massive changes right now. I haven't driven to the office in a decade. Huge segments of the workforce are migrating to exoburbs because they don't need to drive to work anymore. My son hasn't mentioned wanting to drive. My neighbor's kid flat out says she doesn't want to learn (both will be able to soon) and many of their friends don't want to learn either. There is a generation growing up far less invested in the car than the generation before them. Some of the points in this article may not be perfect, predicting the future is a fools bet, but it sure feels like the trend they are bringing up is real.