evrydayhustling

An argument can be true in its parts and still mislead by presenting an oversimplified model. The quote below is pointing out that timesharing can obviously be more efficient in theory -- but ignoring the equally obvious objection that data center operators can't be trusted to allocate towards global efficiency in practice.

> A lot of people want local inference. And somehow, they assume that the problem is solved now. Because, we are going to have open-source models. We’re going to be able to run them on our computers. And so we’re not going to need all the stuff in the data center. But from a global allocation perspective, actually, it’s quite terrible to have everybody have a bunch of the world’s limited supply of DRAM sitting in their own computer. And, being used 5% of the time. You want maximum utilization in the cloud if you’re trying to deal with this very supply constrained environment. And somehow everyone’s... Many people’s intuition is the opposite of that.

So the leak was BS, or at least now outdated. Top 10 must be under 50%.

One might ask why the public markets would even offer this liquidity. One important answer is that they are forced to via index inclusion.

So if you know an index buyer who takes satisfaction in staying above fooled-by-randomness active traders, remind them that that they are now the designated unload zone for shares the privates don't want anymore.

> value the owner derives from the house is not dependent on the resale value of the house

Value is always relative - even if the experience hasn't changed, the rising house price does mean that same experience is being value more highly relative to other opportunities. (In most cases this is because the experience actually has changed: if your neighborhood gets nicer, you are indeed benefitting!)

The bigger fallacy in rising property taxes is that civic services don't necessarily get better or more expensive just because property rates rise. And the bigger unfairness is that homeowners with rising values but not rising income don't always have efficient access to the equity of their homes.

A fair property tax regime would avoid presuming budget increases just due to rising value of housing stock, and might subsidize HELOCs or similar structures for folks who need to access property value to pay taxes.

Even the article itself has the title as a question: "Will OpenAI eat Jev's lunch?". A more useful title would be "OpenAI is Positioned to Compete with Jev".

100%, that's why I led with the post being useful as-is -- you don't have to shift your view to Graeber's because we can always link to it as a supplement :).

IMO part of why myth-of-barter has persisted is because it has worked for helping folks understand our modern finance system (among other reasons, b/c many folks building that system grew up on it as well). So it makes sense to use in an explainer.

With that said, it doesn't feel to me like it detracts from anything to start with obligation before currency. Markers of past favors seems like as easy a place to start as trading stones.

We used to use BERT-based embeddings + semantic distance for classification / decision problems in new domains. There was a lot of interest at the time in these kinds of pre-generative but portable models -- Meta's Prophet was another example that came up a lot.

IMO anyone sane wants some protection right now. The Q is whether we should seek protection through post-hoc accountability, or preemptive bans/certification on certain tech. Both methods will have a hard time stopping foreign actors, but preemptive bans have the added harm of locking in winners and paradoxically making us slower to develop more reliable and aligned systems. If regulation sets a standard for sufficient alignment, what further motivation is there to go beyond?

In this case, who holds the agency is exactly the point. Anthropic and OAI are claiming we need protection from AI itself, but the statement supported by putting agency in the right place is that we need protection from them.

> weights update over time and past learning improves future learning such that we get better sample efficiency

Is there an architecture-independent definition of forward transfer?

For the practical experience and implications of AI progress, I think we are increasingly discussing what these LLMs can accomplish inside a stateful harness, the state of which could be described as part of a (very squirrely) parameter space.

Around 2012 I was working near 5th Ave and 14th st, and there was a giant water main break. They had to dig about three stories down in a large intersection, and you could walk near the edge and look in.

It was extraordinary, like something out of a sci-fi where an alien ship has been torn open and the mess of pipes and wires looks organic. It was also a funny case of lost tech / organizational memory -- I heard they had to ask the union to help find retired folks who had worked on the lower levels and could explain why pipes were laid out the way they were.

Disagree, words drift but they still mean things. Nobody building accounting software in the 90s called it AI (or even an expert systems, barring the most inventive salescritters).

Besides what kind of slippery slope argument is this? All software is AI, so if you do harm with software, AI is a harmful technology? Seems like a not very helpful model of the world.