lagrangian
12hr ago
chubbyFIRE then I suppose, maybe flirting with FAT. HCOL. I think 3M would maintain my current life, 5M with kids, 7M with kids + buffer to pay for a divorce.
This still doesn't answer how to navigate the "how chubby" question with a partner. It's not like we haven't talked about it, it's just hard to really get more than vibes, hypotheticals, rough spreadsheet budgets.
lagrangian
12hr ago
all our money
my husband could have covered the art and I could have covered the music
These seem in contradiction to me. How can you consider separately whether to cover things if it's all "your" [combined] money?
Why are you thinking about income asymmetry?
It's hard not to at a FAANG vs good-for-not-being-FAANG ratio. Less glibly, because when I consider the tradeoff of being single, or at least not married, vs a marriage in which I have less control over spending, and much more to spend it on (square footage, children, travel), it is a hard sell. I feel a strong need to be in as much control as possible of how much/why the FIRE timeline slips. More control, easier sell. Five years easily, more likely ten, especially if the goal is to be so FI as to remain that way in the event of divorce.
Do you keep separate finances?
Largely a hypothetical consideration. There's a candidate in mind, but it's early. I'd certainly want to, and a strong prenup to boot.
lagrangian
12hr ago
wdym aim for 85%? Live on 85% of your income? and for that matter "leave it to my partner"?
lagrangian
17hr ago
How do you balance FIRE with not wanting to be a miserable tightwad? Most specifically interested in if anyone has thoughts on that in context of marriage + children + income gap (let's say 4 me:1 her). A particular potential (pun intended) Mrs. Lagrangian likes travel and activities more than I do and I have trouble thinking about that, especially in context of the income asymmetry.
lagrangian
3d ago
Right, that's the point of a basic sanity check. It's an easy check to rule out things that are insane, in this case "if this has a small TAM, it is not worth a large investment." AI does not have a small TAM, so this does not show that it is not worth a large investment.
But, as you said, it also doesn't show it is worth a large investment. It's a "fast rule-out" heuristic, not any kind of a "rule-in" one.
lagrangian
3d ago
The value of increasing white collar productivity by 10% permanently would be worth every penny spent on AI. It doesn't have to singularity to be worth it. There are nonobvious questions of how to capture the wealth as a company, but the basic "total addressable market" sanity check passes. TAM = 10% total white collar labor expense (well, value of marginal 10% of labor, but similar).
lagrangian
25d ago
Context: still seeing my high scoring secretary, and thinking through some things.
Married/serious-relationship'd Mottizens: not how did you find them, but what was the process like once you did?
E.g.:
- how quickly did things get serious?
- how obvious was it to you that you wanted them to get serious?
- and how quickly?
- was there something that at first you thought was maybe unacceptable that you got over?
- oppositely, something great you didn't notice/fully appreciate?
- or, was your gut just correct quickly?
it takes about one clock cycle for light to traverse a processor. this doesn't prove you wrong, quite, since there's still the possibility of a processor doing something much more clever with the distance it has than it does today.
i got nerd sniped here real hard, so here's a fundamental physics analysis (from Claude and I). Basically, three constraints (below) -> min latency of an operation is ~1e-13s, a 1e4 speedup from today.
That is far less than the "LLM cost/kernel syscall" ratio today, so current LLMs can never be fast enough. As to future algorithms that are magically better enough to close the gap, my best argument is "ehh I doubt it, definitely not soon."
- Margolus–Levitin: with a given energy, you can only switch between two states at a max frequency (min latency)
- Landauer: switching between states must dissipate a minimum amount of energy
- Thermodynamics 101: energy can only be dissipated so quickly
lagrangian
1mo ago
you are missing the point. it would add massive amounts of latency at the lowest level of the stack, and this ends up costing maybe a factor of 1000 even in the optimistic case. this is not "only gamers notice." this is "absolutely everything is uselessly slow"
latency is not ~ever picoseconds to start with - a clock cycle is 1/4GHz = 1/4 nanosecond = 250 picoseconds, and nothing is faster than that.
lagrangian
1mo ago
Realtime LLM code generation will absolutely never replace the core ("kernel") of an OS. The latency is unacceptable, even putting aside correctness and security.
we'll have microchips cheap enough for regular consumers to buy by the dozen from China that each make the entirety of Anthropic's current data centers look like a basic calculator in comparison.
Maybe. I doubt it, but it's not wildly unreasonable to think so. We could absolutely improve LLM throughput/efficiency with better hardware or algorithms.
When it's basically trivial for an entry-level PC to run the equivalent of 100 Mythoses at 100x the speed that we can today, I feel like it won't add enough overhead to the user experience to be noticeable.
No. You are conflating LLM throughput (/efficiency) with latency.
We can improve latency, to a degree. But, we will never have LLM + live-written OS code + compilation (whether via LLM or gcc etc) have latency close enough to pre-written OS code + gcc to not be noticeable, or even to be acceptable. This is a context where shaving off a single clock cycle matters.
A single LLM weight matrix multiplication takes ~100 million cycles, most spent on memory transfer of the weights. Even a radically more efficient algorithm has to have some amount of parametrization in it from an information theoretic standpoint - it's going to mean wayyy more cycles than highly tuned, handwritten in advance, code.
lagrangian
1mo ago
E.g. one pretty obvious thought I had was about LLM-based operating systems to replace Windows and Linux and iOS in the future, which won't need any software specifically written for it - just write any software in any language, including made-up language or pseudo-code, and the LLM would just "compile" that to the 1s and 0s required for whatever CPU to interpret to accomplish the logic of that code (this might last for a hot minute until it needs just some general list of specs
yeah that's not happening. an OS has to be extremely fast and secure. clock cycles matter. an LLM is a deeply terrible way to handle the lowest layer of hardware interaction.
the salvageable version of this idea is closer to an LLM writing whatever shitty electron app you need on the fly, running on a traditional OS and traditional app development frameworks (electron).
lagrangian
1mo ago
I had a similar project once, this makes sense to me. You can use heat shrink over the joint if you're having trouble getting it to seal.
lagrangian
2mo ago
I have a home gym and recurring calendar event for my in-home trainer/PT, on autopay. Claude pulls my calendar and generates receipts for insurance. Least action indeed.
I have some genuine/chronic physical therapy needs such that the insurance case is straightforward, but it's not impossible to find an excuse and/or morally flexible PT (conveniently, this can stand for either physical therapist or personal trainer).
lagrangian
2mo ago
This is a second order consideration. More total volume (reps * weight) matters more than how it's distributed. I don't even mean to over-index on that formula. Just in the broadest sense of "more work, more reward."
Do whatever makes you most likely to consistently lift. For me, that's medium (8) to high (15-20) reps at lower weight - I think it's more fun and less stressful re: form. I also think a babysitter trainer is worth paying for, even just occasionally, for accountability as much as anything. You may be able to convince your insurance to help.
lagrangian
3mo ago
I'm 5'9" and mid looking at best. Just two nerds on a date, and the whole "FAANG and wants kids" thing. I figure scare off anyone who isn't compatible, expediently.
lagrangian
3mo ago
Math yes, spreadsheet no. I think I even mentioned the sqrt n expected value optimization version. If that's not first date behavior, I'm not sure what is.
I'll manage to keep the existence of the spreadsheet to myself for a minute, probably. I brought it up once on a first date that I had already decided I was not going to request a second of, and it was actually a very interesting conversation.
lagrangian
3mo ago
I had to look up the actuaries reference - lowest divorce rate, neat. That kind of makes sense to me - if you're used to systematizing complex things and making mathematically optimal decisions, maybe you have good general judgement. It's a little surprising in that math != emotions, but maybe for (especially a pair of) such minds, it works.
Definitely have an actual spreadsheet. It's a fascinating artifact of my life. Basic ratings, number of dates, p(get to date i | got to date j), who ended it, that kind of thing.
lagrangian
3mo ago
My spreadsheet of first dates has a new high-scoring candidate. Third date scheduled. Vibe's similar to last time, but unlike last time, she's a) actually compatible on what we want out of life, and b) comparably interested (afaict). So that's delightfully terrifying.
The secretary problem says I'm done here unless I want to spend another ~5 years at the current (nontrivial) rate of meeting people, which I fucking don't.
lagrangian
4mo ago
At a closed-door meeting in Princeton, leading researchers said agentic AI tools now handle up to 90% of their intellectual workload—forcing a reckoning over who, or what, drives scientific discovery.
I find this wildly implausible.
lagrangian
4mo ago
High social status?
Yes. Not as high as a doctor, but absolutely. People - men and women - absolutely treat me differently (better) after learning where I work.
Money and status are too linked for it to not give status. There's some importance to it too - we (often) work on systems people know and care about. We're famously difficult to tell what to do professionally, which is itself a form of power/status.
lagrangian
4mo ago
Claim not supported by your source, according to which the teenager was killed in a drone strike against someone else - just bad luck.
Two U.S. officials speaking on condition of anonymity stated that the target of the October 14, 2011, airstrike was Ibrahim al-Banna [not the teenager], an Egyptian believed to be a senior operative in al-Qaeda in the Arabian Peninsula.[7][8] Another U.S. administration official speaking on condition of anonymity described Abdulrahman al-Awlaki as a bystander who was "in the wrong place at the wrong time", stating that "the U.S. government did not know that Mr. Awlaki's son [the teenager] was there" before the airstrike was ordered.[7]
lagrangian
4mo ago
Flash 3? Interesting (if so).
~Consensus at Google, for working on Google's codebase, is that Flash 3 is better than Pro for agentic work. They changed the default for Gemini CLI over and everything. My, uh, friend is heavily involved in beta testing stuff and definitely feels this way.
lagrangian
4mo ago
Use Flash, not Pro, for agentic tasks. Pro is smarter, but so much slower and more expensive that you will genuinely do better with Flash.
lagrangian
5mo ago
Using aistudio.google.com or the API may let you control how prudish it is.
Interesting, but not super useful for me specifically. HCOL + want excellent schools + FAANG puts me well outside those buckets.
More options
Context Copy link