@davidgobaud @atShruti Alright - I have a day job!! Nice meeting you here :)
English
Lingesh
710 posts
Lingesh
@lingeshgct
Open minded , realistic , want to make a difference ….
Katılım Nisan 2009
302 Takip Edilen35 Takipçiler
@lingeshgct @atShruti ok actually with that math the number should be $85M because $500k / year isn't enough need at least $2M. glad we figured this out!
$5M house is small but can get by maybe
English
I’m clearly hitting some nerve here for so many SF richies with $4m+ homes to comment on what it means to live a wholesome fulfilled life with $1.5m house and camping with their private school friends to create core memories for their children!
Shruti Gandhi / Array VC preseed rounds@atShruti
They will still be SF brokie - 50% in taxes (37% federal / 13% state) $3-4m cash on a home in SF Likely needs renovation $250k- $1m Leaves you with $1-2m Many with kids or on the way Nanny -$100k/yr Day care / School - $45k/year/1 kid Camps/Extra curricular - $30-100k Tesla - $50k They will still be at the office 996 to not really enjoy any of this and and will only have money to hike and camp.
English
I am going by the original tweet as the standard of living for being “not broke”.
I think 25 million after tax is the number.
5 million for housing. The remaining 20 million at 2.5% would net 500k a year - that should cover the living expense.
My definition of “broke” is different though !!
English
@lingeshgct @atShruti Curious - what is not broke in SF? Does it take $100M or $1B? x.com/davidgobaud/st…
David Gobaud@davidgobaud
Curious - what is not broke in SF? Does it take $100M or $1B?
English
Hahaha - the metric is wrong imo - I know centi millionaires with money problems (and it may even be true for billionaires) . But those problems are different from “broke”.
Living in sfo , having a private school , 2 million nest egg (plus another x million unsold) - this is stressful life but not broke life !!
English
@lingeshgct @atShruti the conclusion is wrong simply because the people that sold $10M likely have $40M+ gaining left 😂
unless $25M post tax is also broke?? 🤣
x.com/davidgobaud/st…
David Gobaud@davidgobaud
@atShruti yes without the below you are broke: Nanny -$100k/yr Day care / School - $45k/year/1 kid Camps/Extra curricular - $30-100k ps: dw though they sold 20% max? so $40M+ gaining left - easy to get loans, be more liquid, and they have another $20M+ post tax on the way!
English
The numbers seem right - but the conclusion seems off !!
Assuming it is a couple and they have jobs - the nanny and school fees should come from salaries. If they don’t work -they can cut on those.
Leaving that aside - they are still left with 2 million after tax and a 3.5 million home :) assuming a living expense of 10k a month , they can choose not to work for 20 years :)
English
They will still be SF brokie -
50% in taxes (37% federal / 13% state)
$3-4m cash on a home in SF
Likely needs renovation $250k- $1m
Leaves you with $1-2m
Many with kids or on the way
Nanny -$100k/yr
Day care / School - $45k/year/1 kid
Camps/Extra curricular - $30-100k
Tesla - $50k
They will still be at the office 996 to not really enjoy any of this and and will only have money to hike and camp.
The Kobeissi Letter@KobeissiLetter
BREAKING: OpenAI allowed more than 600 current and former employees to sell stock in October 2025, per WSJ. These employees collectively sold $6.6 billion worth of stock. That’s $11 million per person.
English
@chatgpt21 @pmddomingos The point he makes is the delta (of mythos over opus ) is not as high as say opus over one of the earlier models. And as far as I can see - this is a linear scale. So the conclusion of no singularity seems accurate. What’s your point ?
English
@pmddomingos You just really don’t get it do you? How are you allowed to teach?
English
@MillionInt Building a machine that builds industrial automation ? :)
English
Building the factory that builds factories that build trains and rails while the prototrain is already moving full steam ahead
English
At least several code reds in menlo park, seattle and mountain view?
English
@nikogrupen Interesting - looks like dspy on steroids..
Also did you provide the agent a set of heuristics/directions to explore ? I think that’s what the original auto research did (it was told to look for research papers)
English
@natashamalpani Leaving the disagreement aside - I do agree on the market potential - even if ai were used in a constrained manner - a verification system that adapts to usage and is on autopilot will be huge . Creating a system like that for existing agents will itself be a big business.
English
yes, think this is directionally right for the first wedge.
AI works best today when the action space is bounded ( coding, copilots,, narrow customer workflows).
but when the target itself isn’t fixed/ there isn’t a single ‘correct answer’ to verify against, we might have to solve for adaptation over time than immediate verification.
English
the verification problem cannot be closed by more engineering effort.
it has a structural property that makes it permanently hard.the moment you use an eval suite to improve your agent, you've contaminated it.
the agent adapts to the distribution. performance improves. the eval stops measuring what it was designed to measure.
the same dynamic hollowed out every foundation model benchmark. each started as a signal. each became a target. targeting them didn't improve the capability. it improved performance on the test.
verification isn't a state you achieve. it's a race you run permanently.
Natasha Malpani 👁@natashamalpani
English
@pmddomingos I think you meant it as a sarcasm :) (right?!) but this is a killer insight - I ask it to write a program , it creates a mathematical proof of the algorithm and then produces a code !!
English
@ankurnagpal Is carry going to automate this ?! This idea (and the others you post!) sounds sensible however every thing feels like it will be a 6 month undertaking that will need several hours of work every month :)
English
Who is this strategy for?
- You need a large taxable brokerage to make it worthwhile.. at least $1M+
- You pay taxes at a high rate to make the ~1% fee worth it
- You need a platform or investment advisor that can handle this for you
English
The most aggressive tax-loss harvesting solution for individual investors:
Long-short direct indexing
You simultaneously hold long and short positions in the market so you harvest losses on both sides
Here's exactly how it works (& who it's for):
English
By explicitly training on specific tasks, we ended up covering a very large area (in absolute terms) of the space of all possible tasks humans can do, but this large area only amounts to 0.00...01% of the total space. And that's why we still need general intelligence.
Teng Yan@tengyanAI
@fchollet what would be example of 'novel unfamiliar domains'? seems like we've touched most things
English
Interesting. The training data has lots of crud web apps and also the typical ai/ml code that frontier lab researchers write. Does your projects fall with in these categories?
I asked gpt 5.2 to compile a list from public data and it made several mistakes (but I didn’t ask it to write code to compile the list)
English
Does anyone on this planet ACTUALLY believe that Opus 4.6 could productively write code unattended for 14.5 human-dev hours (median)?
Even @METR_Evals doesn't, hence the disclaimer.
So do us a favor and stop the clickbait until you get better evals.
x.com/METR_Evals/sta…
METR@METR_Evals
We estimate that Claude Opus 4.6 has a 50%-time-horizon of around 14.5 hours (95% CI of 6 hrs to 98 hrs) on software tasks. While this is the highest point estimate we’ve reported, this measurement is extremely noisy because our current task suite is nearly saturated.
English
To the extend that it is true , to not know what is underneath the current abstraction today (like compilers) - current abstractions are deterministic whereas llm is non deterministic.
If you give a job to an expert - you may not need to know the details. But if you give it to an intern - you need to verify it.
Either llm abstractions become deterministic (formal proofs may be?) or expert human in the loop is essential.
English
I'm not sure you can be a good mathematician if you always delegate proofs and never learn to do them yourself, or a good software engineer if you don't learn to read and write code.
English
To be proficient at one level of abstraction, you also need to have a decent understanding of the layer below. Traditionally all apprenticeships start with menial tasks, which trains mechanical sympathy for everything the job builds upon.
English
Poking some holes -
1. Savings rate is too low - at this income level , people generally save much more as they are closer to retirement.
2. Lower exec level people in fang make this money (vp and above make more), but I believe in general they also save more , so life style might not match exactly. They tend to save most of the stock awards.
3. Housing costs seem high at least for seattle. There is generally no hoa . But manhattan estimates seem about right.
4. College savings seem low too.
English
How much do you need to make to live an "upper middle class" lifestyle in a major city?
India: ₹1.9-4.1cr/yr ($210k-$450k/yr)
US: $670k-$1.4M/yr
The bar has never been higher.
Assumes:
–Household income
–Own a 4bed house
–2 kids + private school + US college
–1 intl trip/yr
English
This makes sense. Another reason for legacy saas to stick around is distribution and the staying power of the companies (nobody wants to buy saas from a Company that could go bankrupt) .
If AI can equalize this (and it is a big if) - that could be a bear case for legacy saas. But in either case - saas itself is not dead.
English
The part of the SaaS bear thesis I actually agree with is "margins will have to come down, due to increased ease of migration & increased competition." SaaS margins were often way too high and that wasn't sustainable forever. It's a very different take than "SaaS is dead" though
English