@birb_cromble's banner p

birb_cromble


				

				

				
0 followers   follows 0 users  
joined 2024 September 01 16:16:53 UTC

				

User ID: 3236

birb_cromble


				
				
				

				
0 followers   follows 0 users   joined 2024 September 01 16:16:53 UTC

					

No bio...


					

User ID: 3236

Didn't he do some fairly important stuff around epigenetics with respect to methylation?

Emails indicate Dawkins, a former Oxford professor known for his atheist views, was aware Epstein had been jailed but dined with him at a gala dinner at a conference in Arizona in April 2014. Dawkins also wrote to his agent that he had heard “his case is not as black as painted”

https://www.thetimes.com/us/news-today/article/richard-dawkins-epstein-emails-sx82lwsbz

Before I get into the consciousness question, I have to ask, is this the first known instance of Nobel Disease developing in someone who didn't win a Nobel prize? He's an eminent and decorated scientist, who, later in life, has started pontificating well outside his field in an area that is potentially crank-adjacent. I'm not an expert on this sort of thing, but I can also recognize that his skills in evolutionary biology don't necessarily transfer to neuroscience and psychology.


Disclaimer: I'm just a guy who thinks about stuff when he's out fishing. Don't take me seriously.

Moving on from that, the question of LLM consciousness is a hard problem, and one that fascinated me as a layman. My understanding from reading papers on the topic is that there's still a lot of debate over what consciousness even is, and most of the attempts at defining it smuggle in assumptions that the entity under test is embodied and subjected to a continuous stream of stimulus in a way that's hard to apply to an LLM.

Most definitions of consciousness assume introspection. You can do something that looks a lot like introspection on the output side via reasoning. Does that count?

Most definitions of consciousness include awareness of the environment. For LLMs, which essentially only have one sensory organ (the token input stream), how does that even work? Furthermore, what is the environment for something that only exists as a program on a computational substrate? It's hard to model that.

A lot of models of consciousness seem to imply volition or intentionality as well. How does that work with an LLM? They're inert unless something is passed into them as input. A human that didn't do anything unless prodded wouldn't be considered conscious (unless he were a teenager). You could argue that being embodied means that humans are always subjected to stimulus in ways that LLMs aren't (eg: hunger, thirst, temperature), but that seems like a cop out

I think a lot of this discussion obscures the fact that everyone assumed that intelligence and consciousness would (or will) arrive as a package deal. This causes a lot of people to argue past each other.

"The LLM is intelligent!"

"But it can't be intelligent because it's not conscious!"

It seems like Dawkins is trying to square this by claiming it's both.


Back to Dawkins - looking over a little bit of the interchange between Dawkins and the LLM, I wonder if he would have reached the same conclusion if the LLM told him that his books were a middlebrow rehash of Calvinism in biological drag. I've noticed the people in my personal life who go hardest on LLMs being intelligent, conscious, or both tend to make that turn after the LLM starts unceasingly praising them.

A simple stew made with a ham that was on sale for $1.50/lb, cabbage, onions, the last of the garlic from last year's harvest, and some paprika and black pepper.

If you're willing to manage it yourself you can get a few basis points over a fund.

It's not worth much, but I wanted to know how it worked

Why so few details? Can you share what you’ve learned? Spill it.

  1. I'm hesitant to give anybody the impression that I'm someone they should listen to.
  2. Some of it is geographically sensitive and I don't want to dox myself.
  3. My bar for success is low, so it's nothing particularly impressive. I'm basically using this as a learning opportunity in the hopes that I can have a tier of risk and return that sits between a HYSA and equities.

With that out of the way:

  1. I'm buying bonds directly through my brokerage or at auction.
  2. Bond funds are though my brokerage
  3. Treasuries are through Treasury direct.

What I've found so far:

  1. Municipal bonds are a better deal than their headline yield would imply. Being exempt from both state (if they're issued from my state) and federal taxes gives an increase in the final yield, when you'd normally see a decrease. The default risk is also fairly low. The biggest risk is duration. If your state has a municipal bond fund, you can mitigate that a little through the fund having a rolling portfolio.
  2. Collateralized loan obligations are pretty interesting - I have some money in JAAA and it's doing fairly well. Once again, it beats a HYSA, the duration is low, and it's insulated from some default risk by being last in line to default.
  3. Active bond funds usually do better than passive ones, in contrast to equities. It's enough of a difference that they usually justify the expense ratios.
  4. International bonds are kind of a pain in the ass for taxes, even if they're at arm's length in a fund.
  5. Treasury derived income is usually state tax exempt, which is a nice bonus.

Is it just me, or do mortgage backed securities feel like they aren't as radioactive as they were ~20 years ago? The underwriting requirements for mortgages are lot more strict these days, and the systems in place for what happens on a default are more robust.

I've been using a strategy right now of mostly purchasing things that have multiple layers of backstopping. Municipal bonds are another - in my state, if the municipality fails, the state takes it over. If the state fails, at least for the ones I've bought, the feds take over. If the US government fails, I'm not going to be worried about my bond yields. The tax exemption is a nice bonus on those too.

Thanks!

I've recently been getting into fixed-income investing as a way to learn by doing. I'm investing a little bit of money each month (that I can afford to lose). If my after tax yield is a full 1% higher than I could get from my HYSA over a six month time period, with no principal loss of >5% over any thirty day window, I'll say that I have a layman's grasp on the topic. If not, I'll have to figure out what I did wrong.

So far I've been floored by the complexity of fixed income vs equities. I honestly don't know how people who do it professionally manage.

Does anybody else have an unconventional pastime right now?

five way intersection

Half assed construction

Tell me you're in Western PA without telling me you're in Western PA.

My 55% serious solution is to redistrict via random parameters every six years, without any human input.

That's an astounding number. I can do a set of 5 400lb deadlifts.

At multiple points in my life, I have been able to run a sub-seven mile.

There is absolutely no way in hell I'd ever be able to do both of those things in succession, much less multiple miles at that pace.

...what?

That makes sense. I've been strength training for about that long and it's a lot easier than running.

How long did it take you to train for a marathon? I was pretty proud of myself for even making it to the end of a 10k last year.

After all the medical expenses and surprise home repairs have settled, spending is $1,039.65 higher than it was at the same point last year.

It's a little frustrating. I've cut my discretionary spending pretty deeply, and I can't seem to get ahead compared to last year.

  • I've significantly cut down on eating out.
  • No major entertainment purchases (eg: I don't need a new speaker cabinet)
  • Groceries are basically just Aldi and Sam's club now.
  • I'm traveling less overall.

Unfortunately:

  • I've had a few big "block" expenses in the form of medical bills and home repairs.
  • The traveling I am doing tends to be long distance, and gas prices are high right now.

I guess all I can really do for now is keep it up and hope I don't get any more big surprises this year, so my attempts at frugality overtake those expenses.

I sometimes wonder if Franklin Roosevelt didn't do more damage to the US than any other president. In some ways he was a proto-Trump. He saw problems, and rather than try to fix the system, bulldozed through it and made up justifications, then the rest of the government had to rationalize things post-hoc to maintain the veneer of a Republic.

That's fascinating. Can you elaborate on how and why that's the case?

I'm not going to go into detail, but the basic gist of it is that exchange ticker symbols tend to be short strings with a lot of overlap. Despite the textual overlap, each one has a distinct identity and they are not interchangeable. While some LLMs deal with that kind of thing better than others, they all tend to have problems that get worse as the context window fills up, and compaction tends to cause problems as well.

As a toy example that isn't much of a problem anymore, BND and BNDW are not the same thing. They have different holdings, different rates of return, and different tax implications. That little W at the end means a lot, but next token prediction can have a hard time with it.

I was always fond of The Secret World.

Successful assassins use sniper rifles

Most successful ones seem to be in conversational distance.

  • Sirhan Sirhan: within arm's reach
  • Charles Guiteau: inside a subway stop
  • John Wilkes Booth: inside a theater box
  • Jack Ruby: about six feet away
  • Dan White: inside an office
  • Gavrilo Princip: inside a car
  • Mark Chapman: on the same sidewalk
  • Vance Boelter: inside a house
  • Carl Weiss: less than six feet away
  • Luigi Mangione: on the same sidewalk

It's been a while, but my recollection is that only really happens for characters who are disillusioned by the culture and want to leave it.

I have only read two novels by Atwood, but the thing that seems to run through both of them is that she combines the subtlety of a pulp writer with the pretentions of a literary fiction author. It's a bizarre combination.

Opus 4.7 seems to handle the stock ticker tests I do better than 4.6. I assume it's the new tokenizer. Otherwise the only difference I notice is that it's more expensive to run on the same thinking level

You mean other than changing the core premise?

If forced to work in the framework as it stands, I'd probably need to see a mix of human agency (possibly by resurrecting the early and largely abandoned concept of Referrers), and introducing more AI entities that are less... smarmily dickish? The only AI character who seems to genuinely care about biologicals on a personal and moral level is Falling Outside The Normal Moral Constraints, who other AI consider to be psychotic.