ctchocula

As of now, Apple's version is uncompetitive for training. It is decent for inference, which is good but doesn't cover the training usecase which is the main usecase for TPUv4 and Nvidia GPUs like A100. A100 provides 312 TFlops, but M1 GPU only 5.1 TFlops, which is 2 orders of magnitude slower.

This blog [1] does make the argument that M1 Max has about 8x lower performance than a consumer grade Nvidia GPU 2080, but also uses 8x less watts, so it's possible Apple can make a product with similar performance/watt. However, I would say that they are farther away than TPUv4 for now, because the product doesn't exist.

FPS games are entertaining as well. The downside is you need to know: (1) the game (e.g. csgo), (2) roughly know the map (e.g. de_dust2), and (3) have a commentator that breaks down the mindgames both teams are using. If you have all 3 elements, then watching FPS is simply amazing. I've watched a former CS pro (friberg [1]) commentate some csgo matches on his stream. It's simply amazing how much he knows about the game, and some of the mindgames / strategies both teams use.

Agree with you that it's a bit easier to know what's going on with RTS games like Age of Empires, since you have a top-down isometric view instead of only first-person view like csgo.

[1] https://www.twitch.tv/fribergcs

If you've checked the author's previous post [1], you'll see that he admits to being 1200 Elo on chess.com, which is beginner / early-intermediate level. So him losing to GPT-4 may not mean much. Maybe that explains the supposed contradiction here.

[1] https://villekuosmanen.medium.com/i-played-chess-against-cha...

You're confusing cause and effect here. In 2021, even crap 1% 10-year treasury notes cost $100 to purchase, not $80. It's only after interest rates went up in 2022 due to faster than expected Fed rate hikes that 1% treasury notes were no longer as valuable. An intuitive way to think about this is a seller of 1% treasury notes can only ask for $80 when 3% treasury note new issues are available for $100, to compensate the buyer of the 1% treasury note for the lower, less attractive interest rate. The interesting quirk is that it's only the mark-to-market price that is underwater i.e. worth $80. If SVB had the privilege to hold these assets to maturity (i.e. waiting 10 years), they could get their principal $100 back + $10 in interest.

If they did do what you said i.e. "spend $80 to buy $100's worth of treasury notes", then they wouldn't be in the conundrum they're in today. Because their cost basis in that case would be $80 and they could simply sell that treasury note for $80 and they wouldn't have a capital loss at all (in fact, they would be up since they earned coupon payments). It's only because:

1. 2021: they purchased low interest treasury notes before interest rates jumped

2. 2022: interest rates jumped faster than expected

This caused SVB incurring capital losses, which meant their assets to be worth less than their deposits.

If you are curious and want to visualize some of the changes on a graph, you can check out VFITX [1], which is a mutual fund that holds a mix of treasury bills, bonds and notes for an average of around 7-year duration.

-2021 Jan 1: $11.63/share

-2022 Jan 1: $11.12/share = 4.4% cap loss (before factoring in coupon payments)

-2023 Jan 1: $10.15/share = 12.7% cap loss (before factoring in coupon payments)

[1] https://www.marketwatch.com/investing/fund/vfitx

Per Buy, Borrow, Die, they could take a margin loan using that $1M and use borrowed money (say $100K) to pay the rent. With enough accumulated shares, they wouldn't need to sell the shares in their lifetime. After they've passed away, they would pay off the loan and the stock gets a free step up to the appreciated cost basis.

Uber has been adjusted EBITDA profitable since 2021 and is close to GAAP profitability (might reach it this year or 2024).

> Waymo, by comparison, racked up its million miles driving around the SF Tenderloin, in the dark.

You might be confusing them with Cruise. Waymo is driving around SF driverless 24/7 so the million miles also come from during the day, while Cruise is limited to nighttime.

That's surprisingly affordable even for average Joe SWEs in the valley. I can see it being useful for when paperwork gets jammed in the USCIS pipeline for instance.

No, if you sell within 24 months of start of offering period it counts as Disqualifying Disposition and the discount (FMV on purchase date minus purchase price) is taxed as ordinary income and anything on top as cap gains. If you sell more than 24 months from start of offering period, you have Qualifying Disposition and get to compute discount as min(FMV at start of offering period minus price you would have gotten had you purchased on start of offering period, sell price minus purchase price) and use that for ordinary income with the rest of gains as cap gains.

Since stocks generally go up, waiting (2 years - offering period) for QD to use discount % x FMV at start of offering period can be more beneficial than discount % x FMV on purchase date, because you may end up paying less ordinary income and more cap gains. However, you do need to incur more risk to hold an individual stock for 2 years. Conversely, if the stock goes down then DQ may actually lead to you paying less in ordinary income, so you should pay attention to how the numbers look, how confident you are in your company's equity and what your IPS says.

You are correct and missedthecue is incorrect.

IRS laws say you can deduct $3000 ordinary income (i.e. income earned by working, so doesn't include capital gains) per year and deduct without limit against capital gains. Should your capital losses exceed your capital gains that year, you can carryforward the rest of the capital losses without limit onto the next year 2023, 2024, etc.

The whole point of the article is that "tax loss harvesting" beneficial to Steve Ballmer and the other billionaires mentioned in the article, because they have $Bs in capital gains to begin with (Steve Ballmer had highly appreciated MSFT stock as mentioned in the article). An average Joe like you and me even if we have that much capital losses would find it difficult to find the capital gains to deduct against, so we're typically limited to only $3000 per year deducting against ordinary income (compare against no limit for capital gains), which is much less useful tax shield compared to someone like Steve Ballmer who gets most of his income from capital gains, and who therefore can use his massive cap losses to shield against taxation much more effectively. This is the part that makes "Although this is also available to masses" somewhat moot. Even though it's "technically" possible, the masses don't have the massive capital gains to actually deduct against, so they don't benefit from this to nearly the same magnitude as someone who draws most of their income from capital appreciation.

Therefore, one could argue that this section of the tax code is "unfair" in year 2023. The article makes the case that wash sale rule was effective in the 1920s, but that in the 100 years that have passed, technology has made evading the wash sale rule trivial (especially for billionaires), so the rule as drafted in the 1900s is no longer effective at preventing what it intended.

Source: https://www.irs.gov/taxtopics/tc409

Higgs, who hypothesized the Higgs boson back in the 70s, has only published 2-3 papers in the decades since. Before it was announced that he won the Nobel prize, his University was deciding between forcing him out or gambling on him potentially winning the Nobel prize which would bring more reknown to the department. He has commented recently saying if he were a young, freshly minted PhD today he would not have gotten tenure with his publication record. That seems insane to me and a bitter critique of today's incentive structures created by the funding system we have that rewards incremental work over real breakthroughs.

Get your own ukulele, learn to play and inspire her to learn and get better! Ukuleles are great for jamming, so your second ukulele won't be a waste.

Learning to sing and strum chords at the same time was a transformative experience for me and incredible bang-for-buck compared to learning piano, because that immediately unlocks being able to perform most pop songs with chords. Someone above posted a great collection of pop songs written for ukulele that I can't wait to try playing tomorrow: https://sanjoseukeclub.org/song_book.html

Thank you for looking into this issue and the suggestion. Good luck on your project!

This is a good write-up that justifies dark matter using 7 independent pieces of evidence: https://medium.com/starts-with-a-bang/7-independent-pieces-o...

Not working for me on Firefox 107.0.1 / MacOS. I see the save button and the ChatGPT screen has a blue edge. However, when I hit "Save Conversation" the chatgpt_2022-12-10.json file that I get only shows [] and doesn't have the conversation.

Like most growth-phase startups from SV, Doordash are not profitable at the moment and have no profit to speak of with an EPS of _negative_ $2.41. Please correct your "$700m in gross profit a quarter" to "$1.7b in revenue a quarter" or "$296m in net loss a quarter" [1].

[1] https://www.sec.gov/edgar/browse/?CIK=1

Like most growth-phase startups from SV, Doordash are not profitable at the moment and have no profit to speak of with an EPS of _negative_ $2.41. Please correct your "$700m in gross profit a quarter" to "$1.7b in revenue a quarter" or "$296m in net loss a quarter" [1].

[1] https://www.sec.gov/edgar/browse/?CIK=1792789&owner=exclude

For Group C, potentially AI that improves critical thinking skills? This relies on the fact that more critical thinking leads to more democracy. It could be that this type of AI will improve education across the board (rather than critical thinking only), which will improve STEM skills in autocratic states as well, so I could see this falling into Group B as well.

I've had the opposite experience. Searching for soccer highlight videos on Google oftentimes returns a fake video (generally showing highlights of the wrong match or FIFA video game of the same match). However, searching on youtube makes it much easier, because you can see the channel and how many views it has had. Using this information, you can often minimize chances of clicking a fake highlight video.

Norwegian Wood is one of my favourites, so I look forward to reading it. Thanks for sharing!

It gets locked in for 6 months, so if you bought before or on Oct 28 you will get 9.62% for 6 months and the ~6% or so rate for the 6 months after that.

I knew American football was complex, but this article still surprised me in how sharp QBs (and football players in other positions) are. Love hearing experts explain their subject matter and talk about the sacrifices needed to become world-class (in his case, not having much spare time outside football). Well worth the read and he had some nice book suggestions as well that I'm gonna check out.

To clarify, there do exist closed form solutions for quintic functions; they just cannot be expressed using rationals and a limited set of operations of additions, subtractions, multiplications, divisions, and root extractions. However, if you do not limit yourself to those operations, there do exist closed form solutions to quintics. See [1] for a few examples.

[1] https://mathworld.wolfram.com/QuinticEquation.html

This is overblown, because there are cases where you do want bond funds and bond ETFs. Was just reading this article today [1], which presents some of their benefits.

- More convenient than individual bonds.

- Can't estimate future liabilities perfectly and therefore won't be able to perfectly duration match using individual bonds.

- If we hit a recession and interest rates drop, you'll get a nice price bump from your bond funds when your stocks are dropping.

That being said, I do agree that the dangers and risk of bond funds and ETFs don't get talked about often enough, and that you should know what you're doing before buying them. If you have a short-term need for the money (e.g. need 100k for downpayment in 12 months' time), you should not buy a bond fund with a duration longer than 12 months. That is when you will hit the danger parent brought up where you will get a NAV drop and won't have enough money at the end of 12 months to meet your liabilities. In that case, you should really duration match [2] with individual bond that matures in 12 months.

[1] https://occaminvesting.co.uk/against-duration-matching/

[2] https://occaminvesting.co.uk/duration-matching-an-introducti...

It's not possible, because monetary policy tends to be a lagging indicator of recessions [1] whereas the stock market is a leading indicator of recessions [2]. Also, there's the problem that dropping the interest rate doesn't mean the Fed will continue to drop it to zero. Similarly increasing the interest rate doesn't mean the Fed will continue to increase it until we hit a recession. The predictive power provided by such a lagging indicator combined with an imperfect correlation means that monetary policy alone isn't enough to time the market.

[1] https://fred.stlouisfed.org/series/INTDSRUSM193N

[2] https://fred.stlouisfed.org/series/NASDAQCOM

I agree with the spirit of this post, but I think you are going a bit too far in the other direction by stating that "the pensions would have been fine without the policy". From Matt Levine's article today, it was a more dangerous situation than you are making it out to be:

> "At some point this morning I was worried this was the beginning of the end," said a senior London-based banker, adding that at one point on Wednesday morning there were no buyers of long-dated UK gilts. "It was not quite a Lehman moment. But it got close." ...

> "If there was no intervention today, gilt yields could have gone up to 7-8 per cent from 4.5 per cent this morning and in that situation around 90 per cent of UK pension funds would have run out of collateral," said Kerrin Rosenberg, Cardano Investment chief executive. "They would have been wiped out."

It would be interesting to see whether it's possible to reconstruct the rules by comparing the negotiated prices to a baseline like Medicare price, and doing classification into discount buckets to recover the categories insurers negotiated.

Anecdata: I personally know 2 people in South Bay that lost power 2 hours ago.

Shorting your employer just in order to net out the overexposure seems putting the cart before the horse. Part of choosing who to work for is believing in the company's future financial outlook, because you are intentionally taking a long position in the company with your RSUs. Taking a short position both costs money and it would be simpler to take a long position in another company by choosing to work for them instead.

On the other hand, if you are 100% certain in your investment thesis that you don't want overexposure to any company including your employer, you could try direct indexing the rest of your portfolio e.g. buy S&P 500 except for your employer. Selling on vest is another simple alternative you could consider.

Agree with the first point. Either time-weighted return or IRR gives a more meaningful estimate of the return than simply dividing the final value with total amount invested ($5,200).