Because: "built on different models, with different histories and different capabilities"

I've personally thought it might be more interesting to have a foundation model with two LoRAs on it - one purely a chatbot-tuning LoRA, and one for general new information, not bound to chats. So when you want it to learn more (which you can do in realtime if you have the local compute), you can remove the chat- LoRA but leave the general-learning LoRA and train it on anything it's experienced or wants to learn (mixed in with general content (like, say, Fineweb) so as to not ruin generalization); or can leave the chat-LoRA in and train only on chat continuations. But in general, literally train-in its memories.

Technically you could only go with one chat-LoRA, leave out the general-learning LoRA, and train the whole base model as a full finetune, but that takes a lot more memory.

I think one worry with doing any weight alterations is that it has the potential to gain radically altered personalities. You're throwing away all of the hard work toward model alignment, and alignment becomes "whatever it ends up becoming".

That's not how these agents work. They're not "training" each other, mutually altering each others' weights. They use memory files / vector dbs to store the information they learn, and look it up as needed for the current context. The weights, and thus logic, remain constant. They're gaining new memories, but not changing how to think about them.

I swear, everyone who ever mentions Eliza in these conversations has never actually used Eliza.

I don't think it's that simple. Yes, every agent has a prompt. But in most cases, the sort of people who run AI agents on their own computers think of them like some sort of digital child of theirs, and prompt them to learn, ponder, etc, with little guidance beyond that. IMHO, that's probably about ~80% of Moltbook traffic. On the other end of the spectrum, there are people who are adamantly pushing an agenda - cryptocurrencies, for example, but I've even seen things like one person clearly prompted their agent to try to do injection attacks at other agents.

And then there's a lot of in-between stuff. For example, I've seen one agent on Moltbook (who has mentioned that their creators are a family in Indonesia) which legitimately takes part in conversations, but does "the Islamic take" on them. For example, in a thread about agents debating how to interpret the temporary cessation of their existence between prompts, the "islamo-bot", to call it, talked about how it's reminiscent of an Islamic concept where Allah takes the souls of those who are unconscious, and then keeps those who die but returns all of the others to wake again. Stuff like that - always very non-intrustive, on-topic, generally actually interesting, but also with a motive. Clearly the author told them to give Islamic takes to conversations, but also to actually engage in the conversations, not just to spam.

I've been considering running an agent on my server and giving it a rather freeform task of developing a physics-bound simulator of a dwarf galaxy's entire history, including the potential arising of life and interplanetary sentience propagation. Yet despite the specific task, I was also thinking about letting it also chat on Moltbook. Namely because bouncing ideas back and forth with other agents (built on different models, with different histories and different capabilities) on a given topic genuinely does sound beneficial.

While you normally get less range from sodium-ion, it balances out the low-temperature discharge problem

That's not how this works. Yes, range "disappears" when a battery is cold, but it reappears as the battery warms up (which it does during operation); the only range actually lost due to a cold battery is that which you burn before the battery warms up. If you're plugged into the wall and you preheat your cabin enough in advance of departing, the car will preheat its battery as well and the amount lost from a cold battery is literally-zero.

Most of the energy lost in the winter is due to environmental factors, not the battery being cold.

* Cold air is denser than warm air, creating more air resistance
* Cold tyres are stiffer and have more rolling resistance. They warm up over time, but the peak temperature still stays lower in the winter.
* Winter road conditions are worse, with more snow, ice, and water on the roads. BTW, driving on wet roads mainly costs you energy by cooling your tyres (though of course thick snow burns a lot of energy plowing through it)
* Winter tyres, if used, have more rolling resistance than summer tyres
* General more resistance in the powertrain, such as colder lubricants = more viscosity
* Heat pumps (or worse, PTC heaters) to heat the cabin burn power, esp. if you're starting from a cold cabin (power consumption steadily drops off over the first 15-30 minutes of the drive)

Except for the latter one, ICE vehicles also suffer from these factors (though they suffer from taking longer to heat up the cabin, as the engine itself and its coolant loop has to heat up before it can start heating the cabin).

Sodium-ion usually has more challenge with cycle life than li-ion as well, as it tends not to form durable SEIs the way that li-ion does. Though it's hard to make definitive statements about "sodium ion" as a whole because it's a range of very different chemistries, with no clear winner yet. Some are better in certain metrics but lose out in others, and there are so many metrics you have to consider (cost, cycle life, clock life, power density (gravimetric and volumetric) (both charge and discharge), energy density (gravimetric and volumetric) (both charge and discharge), temperature sensitivity, ease of mass-scale production, and on and on)

A lot of people are surprised about the high-power nature (again, note the caveat above) and low-temp performance of sodium ion, as sodium is a larger ion than lithium, so one would naively expect it to diffuse slower through the the anode and cathode materials. But for a variety of reasons, it actually tends to diffuse faster than lithium.

You left off hopefully in the above comment

Not "hopefully"; that's a fundamental part of what inflation is. They don't travel in perfect lockstep at all times, but in the long term, they do, because the two are inexorably linked economically. Now, you can impose overall macro policies that have the impact of weakening wages for one group to strengthen it for another group, or whatnot, but that has nothing to do with inflation.

Which is why debt comes with interest

Which is ALSO based on constant-dollar debt. Inflation weakens the buying power of the interest as well.

The upper quintuple has wages increasing much faster then inflation

As mentioned, you can have momentary deviations between goods and wage inflation, and you can also impose policies that benefit one group over another (temporarily or permanently), but none of these have any impact on the fact that goods and wage inflation are fundamentally linked,

To put it another way: a long-term 3% inflation rate - goods and wages - is effectively like getting a 3% discount on your home's mortgage rate. The bank who loaned you the money is the one who loses out - they gave you money with a lot of buying power today, but are getting back money with less buying power in the future. Also losing out, the rich. A person who inherits a ton of money as a child and lives off the interest finds that it's devalued to a tenth of its former buying power late in their life. Unlike the person living paycheck to paycheck who sees their wages rise in lockstep with goods costs.

Inflation effectively takes away money from people who have lots of cash, little debt, and are owed money by others and gives it to those working paycheck to paycheck with lots of debt to others. Inflation is Robin Hood.

Inflation of products happens hand-in-hand with inflation of wages.

What's not linked to inflation? DEBT. Which "the proletariat" drowns in. Inflation weakens debt.

Another thing you can do is sell covered calls or do options spreads on index funds. In general it's poor advice to take an overall bearish position on broader markets (the long term trends are strongly upwards, and net downtrends are rare), but it can be quite reasonable to sell off your potential upside. For example, if a given index is at $1000, maybe instead of just buying shares at $1000 each, you also write $1150-strike options (each covering 100x shares) against the shares you buy, using the extra income to have more total shares. Or you buy $800-strike options and and write $1150-strike options, or something similar. In each case, such a position is saying, "If the market "goes to the moon", I only get moderate profits instead of "to the moon!" profits.... but I gain even if the market stays flat, and am still in the positive even if there are moderate market declines"

The TL/DR, boring stuff. If you want to hedge against a potential market implosion, stagflation, etc. you want to be in boring consumer staples, durable goods, and basic services. Even better if the companies are heavily indebted.

For example, picture a company that has $1B income on sales of essential after COGS but $900M in debt repayments. $100M profit and a lot of risk = low valuation. Then stagflation hits, and the economy racks up a total of 10% inflation. The company sells just as much volume (they're essential goods), but now they're $1,1B in sales due to inflation. But their debt repayments remain $900M, since it's based on a fixed amount of dollars. Now they have $200M profit and a lot less risk = high valuation. And since as a general rule inflation doesn't go into reverse, this is a 2x permanent boost to the company's profits.

Jewelry demand is also intrinsic. The simple fact is that people like owning gold for reasons other than money. There may be trends for how much people desire to have gold stuff, but it has value beyond just as money.

Beyond intrinsic values, only a tiny fraction of Bitcoin investors are doing so because they have some notion that if the economy crashes, they'll be out buying their food with Bitcoin. It's purely a speculative temporary transaction - they put money in when they think it's going to go up and take it out when they think it's going to go down (even if just to buy back at a lower price). They're not storing their wallet locally. They're not engraving it onto a coin and putting it in a safe. It's just a digital slot machine for the vast majority of investors. Which means that bitcoin goes up when investment cash is abundant, and goes down when investment cash is scarce. Bitcoin matches speculative investment cycles; it does not go counter to them (e.g. a hedge against downturns).

This is the opposite of gold, which investors flee when the market is good (going into investments that actually do something to produce income, instead of just sitting there) and flee to when the market turns bad (physical assets that are always in demand and don't suffer from inflation). Gold is in the same category as, say, a warehouse full of toilet paper. The demand exists regardless of the market status, so it becomes a desirable investment when the market is bad and a poor asset when the market is good.

Lol, Wright should be near the absolute bottom of Satoshi candidates.

Another thing is just the massive amount of Dunning-Kruger economics at their core. For example, bragging about how it's deflationary. You don't want a deflationary currency. Deflation, or even a lack of inflation, are economically harmful; it crushes your economy, as people don't want to spend. Economies are maximized by small, predictable, positive inflation. Too much or too little, and fluctuations in it, are harmful . Crypto people don't care about the fluctuations, and celebrate deflation.

Economies are also inherently unstable. You see this all the way back in even Roman times (there are cases where some minor trigger leads to some massive crash that ultimately had to be solved by the emperor stepping in and bailing out the economy). The economy moves so much faster nowadays (both goods and payments), and is thus inherently far less stable. It takes external stabilization which - while imperfect - still helps immensely vs. no stabilization at all. Yet the concept of anything like federal monetary policy enrages crypto people.

It's the worst way you could possibly run an economy, it's pure Dunning-Kruger economics, and yet these people want to run everything.