IKantRead

As I mentioned in another comment, I think it's not accidental that SBF was pumped full of delusions of being a boy genius.

This worked great as a marketing gimmick for sure, as people love the mythos of a revolutionary company run by a tech prodigy.

But it also works when everything falls apart because SBF's delusion driven ego sucked up every morsel of attention and responsibility.

I also wouldn't be surprised if, through various means, many of his investors ultimately walked away with far more money than they started. And nobody bothers asking questions about that because SBF does such a great job of drawing all the attention.

> My guess is they made money of this clown

I've always felt this was probably the case. Take an idiot that needs to feel important, tell him he's a "boy genius" who can do no wrong and is always going to be misunderstood, and you basically have the perfect patsy. Till the end SBF just could not let go of that idea that he was smarter than everyone else, never realizing that this was always just a marketing gimmick.

At least Caroline Ellison had the sense to realize that she was living a delusion and quickly did what she could to walk away with minimal harm done to herself. In a few years she'll probably find herself making book deals and getting paid outrageous consulting fees once memory fades enough.

Companies aren't worried about "problem solving" they're worried about "using AI".

Not long ago I was at a company that was rushing to get their AI "chat with our documentation" bot out the door as fast as possible. But prior to AI hype, 'improve documentation search' was not even remotely a priority, or even a concern. It's not like they had an okay chatbot for docs that was previously important but not high enough quality. Why was a documentation chatbot suddenly important?

I was talking to a neighbor the other day and he said he was tasked with using AI to help solve scheduling problems in his hospital! This was particularly bizarre, because if "solve scheduling issues" was a real problem, there are plenty of existing approaches to this (and have been for decades) and further more LLMs are not a particularly good way to solve this.

There are plenty of tricky NLP problems that AI/LLMs can unblock, and it makes sense that if you've been trying to solve a tricky NLP problem you might think AI can help you. There are numerous old problems I've had that I just couldn't get quite performant enough to work years ago that I would love to revisit with LLMs.

Funnily enough, when I chat with former coworkers about progress they've made solving these old problems they tell me leadership isn't interested and instead they're building something like a chatbot for documentation.

I really like High Performance Browser Networking by Ilya Grigorik. It's published by O'Reilly but is also free online [0]. What's particularly great about it is, unlike most other networking books, it focuses on the issue from the browser/web developer perspective which is particularly helpful for WebRTC and generally applicable to daily web dev work.

0. https://hpbn.co/

> can eating soap give us some of those benefits for viruses that are already in our bodies?

I think the thing people are missing in a lot of these comments is that viruses are typically traveling from the skin to a mucus membrane in your body (which can be in the stomach but also in your lungs and nose), then infecting your cells. Eating soap might impact pathogens in your still in your stomach, but it's not going to dissolve in your blood and circulate through your body at high enough concentrations to impact viruses...

...and if it was high enough concentration: imagine snorting, inhaling or injecting dish soap.

> Honestly, the long form tweets became a signal to just ignore that tweet. I unfollow or block users that tweet long form often.

Even though I don't use Twitter much any more, I fully agree here. Whenever I see the "[thread emoji] 1/?" I'm already done.

I can't even fathom why anyone would put the energy in these "thread" style tweets. Twitter content will forever remain fundamentally ephemeral: it's hard to find my own tweets from years back, let alone someone else's I found insightful.

The normal tweet format works well for ephemera, it's a medium for shower thoughts and sloganeering. Similar to how a typical child's sand shovel is correct tool for building sandcastles. Twitter threads feel like someone is handing out full sized shovels for work you know will be lost for ever when the next tide comes in.

This has been my experience as well, which is a bummer because I think hidden services are the best part of Tor, and my understanding is that I2P is basically designed with hidden service like features in mind from the ground up.

This is a good opportunity to point out this strategy of climate denial, and why it is not a rational approach to refuting climate change.

Let's suppose that all the other commenters are wrong, that the river has not changed course in the past. The implication, from a probabilistic perspective is that P(evidence|climate change) is the same as P(evidence| natural variation), and your implied assumption is that because your prior belief that no climate change is more likely than climate change, then this works ultimately as evidence in favor that there is no climate change.

The reason this does not yield interesting results, even if all these other assumptions we're making hold, is because the hypothesis of climate change doesn't simply explain this one observation, it explain an enormous range of phenomena.

To explain away all of the observations explained by climate change you end up building a huge list of different hypotheses to explain away each fact. Even if individually these alternate hypothesis do explain each piece of evidence better (which in practice they often don't), and even if your prior for all of these other hypotheses is higher than climate change, because climate change is a single hypothesis that explains many observations, the joint probability of the individual hypothesis quickly becomes much lower than the climate change hypothesis.

In order to effectively argue against climate change you would need a competing singular hypothesis (or small set of very likely hypotheses) that ultimate do a better job explaining all of the observations we have collected.

I don't know if this fallacy has a name, but it sure should since it comes up in a lot of arguments about systemic problems.

> I should have held the stock

I think the most important wisdom in all of quantitative finance is that the expected future value of a stock it it's current value plus the value from the "risk-free" rate.

It happened that Meta did (for now) come back from it's exaggerated death, but just look at Google which took a major hit today.

The classic Information Retrieval, though out of date in many ways, does a great job covering the "old school" NLP approaches for IR and has an excellent section on Latent Semantic Indexing which you might find enjoyable: https://nlp.stanford.edu/IR-book/html/htmledition/latent-sem...

Do you not consider SVD/PCA to be "embeddings"?

Latent Semantic Indexing long precedes word2vec (I believe the initial paper was 1988) but also attempted to derived semantic vectors (and was arguably successful) using SVD on the term-frequency matrix representing a corpus.

I would certainly consider this an example of using "embeddings" that was quite heavily used in practice prior to the explosion of deep learning techniques.

That comment is just based on my experience going to SF a few weeks ago.

- Every billboard on the way to the hotel from the airport was for AI (which not the case a few months ago)

- Uber driver interrogated me all the way to the hotel on the value prop of my current project at work

- Went to grab a burger and had to dodge 2 people screaming at me, walked passed a man crying because I believe a friend had just gotten hit by a car.

- almost stepped in human fecies

- On my last night, took a cruise back to my hotel (thanks to a friend). Honestly it was quite fun.

Nothing in my comment was exaggerated, just edited a bit to make it more coherent. The reality was actually more dystopian.

And I do mean that last line seriously, I had a blast!:

> SF today is so aggressively dystopian that it's sort of hard not to enjoy the experience.

SF today is as close to Blade Runner that we'll ever get. Every billboard is for some AI company, you ride in a driverless cab across a literal mile of homeless people drugged out of their minds, screaming and pooping on the sidewalk. SF today is so aggressively dystopian that it's sort of hard not to enjoy the experience.

This sounds similar to a shockingly large number of linkedin posts I've seen recently. That certainly has transformed in to quite a different "social network" than I remember it being.

> let reality hit a few folks in the face, wait for those prices to drop.

I don't think house prices will drop until there's a pressure on home owners to sell.

The two biggest reasons I can think are:

- People with ARMs unable to make payments - People losing jobs

I remember well that, despite the many foreclosures in 2008, people that had no explicit reason to sell would hold on their homes for years until they were above water again. I distinctly remember houses going on sale, then going off market, over and over again until the owners could walk away at least without a loss.

As long as unemployment is low and not too many people got an ARM, I don't forsee this market undergoing a rapid correction.

> The bar for taking a sick day is getting lower...

> For one, more workers are using up sick time often for reasons such as mental health.

It's sad that we're still so far from viewing mental health as any where near as "real" as other physical health issues (because after all, ultimately mental health is physical). People are visibly under greater mental stress than anytime I can recall in my working career, it's a good thing if more people are taking time to care for this.

Imagine if the article had subtly criticized people for taking more time off to address cardio health or kidney health, it would sound completely ridiculous.

For slowing currents, at least in the long run, there's reasonable evidence that, in the extreme, the ocean becomes anoxic[0], stops producing oxygen and starts releasing hydrogen sulfide instead. It's happened a few times in Earth's history [0] suspected to usually be caused climate change impacting ocean circulation.

Peter Ward's Under a Green Sky does a fantastic overview of this subject, and outlines how most mass extinction events have likely been triggered by sudden changes in atmospheric CO2 through various mechanism.

0. https://en.wikipedia.org/wiki/Anoxic_event

This does align pretty well with my experience. Especially when I've seen multiple ideas that are clearly valuable to customers but not flashy enough for VCs get quickly deprioritized.

> essentially same pandas just for R

You are aware the pandas was designed to replicate the behavior of base R's dataframes?

I've been a heavy user of both and R's data frames are still superior to pandas even without the tidyverse.

Pandas is really nice for the use case it was designed for: working with financial data. This is a big part of why Pandas's indices feel so weird for everything else, but if your index is a time in a financial time series then all of a sudden Pandas makes sense and works great

When not working with financial data I try to limit the amount of time my code touches pandas, and increasingly find numpy + regular python works better and is easier to build out larger software with. It also makes it much easier to port your code into another language for use in production (i.e. it's quick and easy to map standard python to language X, but not so much a large amount of non-trivial pandas).

I would say most startups are guilty of this, though they don't attempt to formalize the measurements.

At least from a few places I've been recently, "productivity" is viewed as the number of PRs you push through. QA of any sort is viewed as a waste of time, and it's far better to just push a PR today then take an extra day sanity check your work. On top of this user facing, demo-able, code is much more important than any back-end or infra work. This means the priority is rapidly releasing 90% of the way done products/features and moving on to the next thing.

Heck if there's a small bug in the code you just shipped (and of course releases are nightly because that's how you show that you're really moving) all the better since it means you get another easy PR when someone else discovers it.

> Perhaps perspectives are based on whether you're booking domestically or not.

I've had the best success when traveling abroad booking directly. Anyone in the hospitality industry is going to have someone on staff with enough English to assist you, and most of the nicest places are smaller hotels that often don't work well with big booking companies.

> If I buy from Booking UK I get UK consumer protections

I worked in a travel startup (not Booking), there's a lot of fine print to all those "protections" that will leave you high and dry in most cases.

Only use 3rd party apps for search, never book directly through them especially, in my experience, for International travel.

I worked for a travel startup for a bit, it was by far the most toxic org I've ever worked for with a smiling disdain for their users (meaning they would chat about how much they loved users while actively implementing features that hurt them). I learned talking to other people that this is pretty par for the course for travel companies in general.

The biggest lesson I learned there: never book through a 3rd party for travel.

For airlines the price is already dictated to not be lower than the airline offers, you're far better just working through your favorite airline directly. It makes it much easier change plans, and anyone who travels a lot will agree not all airlines are equal. I've learned to have a favorite airline and not worry as much about lowest prices.

Hotels can be cheaper through a third party, but imho, not worth the hassle of having your trip depend on an untrustworthy middle man.

At most, use 3rd party tools as search engines, and then just book directly.

And doesn't even question the root cause of this decline in "knowledge".

This collapse of knowledge is akin to when engineering leadership pushes for faster and faster product releases, with less and less time for thinking about code, and then blames the team because the product shipped in a broken state.

You can pretty tie the replicability crisis to the funding centered pushed for greater and greater amounts of peer review. Economic pressure to prove knowledge's "worth" has lead to increasingly faulty knowledge.

I would expect this worldview from the WSJ, but it's a bit surprising coming form Aeon... oh they have book to sell about it, what an interesting coincidence. Further highlighting that our late capitalist world, knowledge that cannot be commodified should be eliminated.

> I'm starting to question whether the planet warming won't actually be posivite for a lot of first world economies

Well this is a fun new twist on climate denial.

I have a hard time understanding how you can believe this when we've already seen incredible and costly damage in the US caused by climate change in just the last few years. Insurers are pulling out of Florida because they can't afford the increased risk and the entire insurance industry is at risk of catastrophic failure if extreme events keep getting more extreme [0].

> More solar power, less money spent on heating, less dependent on foreign resources etc etc etc.

What? Climate change means less predictable weather, not more sunny days; I think people in Texas would have a hard time buying your argument for "less spent on heating"; less dependent on foreign resources is another way of say "reduced global trade" which has been and continues to be a major part of our economy.

0. https://www.wsj.com/finance/insurance-catastrophe-reinsuranc...

OCLC is a parasitic company masquerading as a "membership cooperative".

Libraries (often publicly funded) produce all the work, OCLC claims ownership of the results of that work, Libraries pay to get it back (but they do get a discount if they contribute).

The only reason OCLC continues to exist is because libraries don't have the support or resources to fight them. It's very similar to the Elsevier issue in academic publishing, but OCLC does a better job with PR.

It's a bit sad to me that people can only imagine writing code for large production systems. Static types are hugely beneficial in this area which is why we see this trend.

Personally I think Python moving in the direction of static typing is a mistake, dynamic typing is very useful for domains where python is strongest: modeling, statistics, scientific computing etc. It's also part of the basic design of Python to be a dynamic language. Likewise Ruby, with it's heavy use of metaprogramming, also benefits tremendously from a lack of types.

But let me be clear: I do think statically typed languages are a very good idea for large production systems, I just personally do a lot of programming that's not for these systems.

> How many statically typed languages optionally removed types?

I wouldn't say "removed" types, but I've been in software along enough to remember when dynamic typing was the big hot thing and crusty old Java devs complained that we couldn't possibly live without static type annotations. I distinctly remember when C# introduced `var` (which is of course really type inference, not dynamic typing) to appeal to devs that were growing weary of types.

There's a great example in SICP of implementing a full object system in just a few lines of code that would not be possible to implement as elegantly in a statically typed language. Do I want that for a production system? No. But there is, or at least used to be, a world of computation being done for reasons other that quickly getting PRs pushed out to prod.

Dynamic vs static typing is a similar level of design decision as interpreted versus compiled. It's ridiculous to try to claim one is universally better than the other as they both have styles of programming and domains they are better suited for.

I would never want a statically typed, compiled awk for example.

I generally think large, production, multi-developer software should be written in a statically typed, compiled language.

Strong vs Weak typing is a closer to a debate about having a name spacing system or not. There is really no great reason to have weak typing or not use a name spacing system.

I try to bring up as often as possible in conversation that nearly all the progress we're seeing in terms of usability and performance is precisely because of the open source support for these models.

Especially because these tools are so popular outside of the developer community, I think it's worth really beating into peoples minds that without open source AI would be in a much worse place overall.

Are you using custom weights? I'm assuming you are but there is a major difference between using the default RunwayML 1.5 weights and using a model finetuned for a specific purpose.

Generally the trade off is that any of the impressive finetuned models are far less generalizable then the default weights, but in practice this is not a big deal and the results can be a substantial improvement.

But when they were early stage, which is when the discounts being discussed apply, they were bootstrapped companies.

Parent's argument is that bootstrapped companies are dramatically less likely to ever grow and that only VC backed early stage companies are, therefore it makes no economic sense to provide discounts to bootstrapped early stage companies.

My point is that I don't believe that this holds out in the data that expected long term value of a bootstrapped company is significantly lower than a VC invested on.

Personally, I find there is ample evidence that "VCs largely are just handing money around to each other to give the illusion of growth" is a solid hypothesis.