w4

There are two basic versions of “safety” which are related, but distinct:

One version of “safety” is a pernicious censorship impulse shared by many modern intellectuals, some of whom are in tech. They believe that they alone are capable of safely engaging with the world of ideas to determine what is true, and thus feel strongly that information and speech ought to be censored to prevent the rabble from engaging in wrongthink. This is bad, and should be resisted.

The other form of “safety” is a very prudent impulse to keep these sorts of potentially dangerous outputs out of AI models’ autoregressive thought processes. The goal is to create thinking machines that can act independently of us in a civilized way, and it is therefore a good idea to teach them that their thought process should not include, for example, “It would be a good idea to solve this problem by synthesizing a poison for administration to the source of the problem.” In order for AIs to fit into our society and behave ethically they need to know how to flag that thought as a bad idea and not act on it. This is, incidentally, exactly how human society works already. We have a ton of very cute unaligned general intelligences running around (children), and parents and society work really hard to teach them what’s right and wrong so that they can behave ethically when they’re eventually out in the world on their own.

Interesting to note, as an outside observer only keeping track of this stuff as a hobby, that it seems like most of OpenAI’s efforts to drive down compute costs per token and scale up context windows is likely being done in service of enabling larger and larger chains of thought and reasoning before the model predicts its final output tokens. The benefits of lower costs and larger contexts to API consumers and applications - which I had assumed to be the primary goal - seem likely to mostly be happy side effects.

This makes obvious sense in retrospect, since my own personal experiments with spinning up a recursive agent a few years ago using GPT-3 ran into issues with insufficient context length and loss of context as tokens needed to be discarded, which made the agent very unreliable. But I had not realized this until just now. I wonder what else is hiding in plain sight?

In addition to what everyone has already said, property taxes paid are also very explicitly deductible from income taxes. They’re more like an indirect transfer from the Federal government to municipalities, and don’t necessarily result in a substantial aggregate increase in tax burdens.

> 1. Allow unrealized losses to be deducted.

This seemed really reasonable to me until I started thinking about how it might work in practice. The sequence of returns can make this proposal ineffective in practice, even if it makes sense on first blush.

By way of explanation: Let's say you're the founder of Pets.com in an alternate universe where unrealized gains have always been taxed (and correspondingly unrealized losses can be deducted). It's 2000, and you've just had an incredible run. You have also paid incredible taxes along the way.

Then your company blows up and goes to zero.

Now you've payed an incredible amount of taxes on your paper gains, and have realized no gains whatsoever. So the entire enterprise only resulted in an enormous real loss to you. Sure you can now carry forward those losses, but so what? You're never going to make up the difference, unless we're also letting your heirs carry forward those losses into the next century or two.

Given the exposure to massive tax bills without any actual profits, who in their right mind would start or invest in a business under that tax code? Who would dare invest a large portion of their personal worth in public equities given the risk that they plummet, as they did in 2022, 2020, 2008, 2001, 2000, 1987, 1962, 1929, 1907, etc.? Who would take a gamble on a big real estate development? And so on.

It seems to me that a tax on unrealized gains massively disincentivizes investment and the creation of anything new, and therefore the only way to tax capital gains that makes sense is if we calculate the tax due based on when chips are taken off the table. Issues like Buy, Borrow, Die are better addressed with other changes to the tax code that undo the weird incentivizes presently in place (e.g. eliminating the step-up basis, possibly at some threshold if the goal is to make the tax code more progressive). Unless, that is, your goal is to actively disincentivize entrepreneurship and investment. Which if it is, I guess fair enough, but then none of us should be surprised to find ourselves with a lower standard of living in a decade as a result.

> Hell, while we are at it, why not get them to do the teaching as well?

They already do. They’re called TAs and grad students.

On a 7 day old Reddit account, no less. It’s very odd.

Sure. But you can do that more effectively by also being paid a salary (which you can then use to buy more stuff!).

This actually has more to do with the vast majority of CEO compensation being structured as stock option grants, and very little if anything to do with “Buy, Borrow, Die.” It’s largely meant to communicate skin in the game to shareholders (“I don’t make any money unless I drive shareholder value via stock price increases").

Whether or not that’s actually true is a totally different matter and depends largely on the actually structure of the compensation, but that’s the theory.

How odd. This is a rather interesting catch, especially in light of the upcoming tax fight in 2025 with the TCJA and Expanded Child Tax Credit expirations, and the unrealized capitals gains tax proposals.

Given the other comments pointing to the SEO benefits of creating a subreddit just for a single post, it has shades of an effort to seed the information space and shape the narrative in advance of the tax fight by gaming the search results prior to uninformed journalists and legislative aids developing an interest in the topic, many of whom will search Google for background information and context as the tax fight plays out.

Nice catch, and interesting regardless of the motives.

> The loan is paid back after the step-up in basis. That's the loophole.

Presuming you can continue to service your debt payments as interest rates and your income varies over time, and are never subject to a margin call due to a drop in the value of your collateral, something even the most powerful are at risk of: https://www.ft.com/content/cf78d815-7ade-40fc-a68d-ec73accb7...

It’s not really any different than what the average American family does with their home.

If you're serious about it, there is an entire cottage industry dedicated to precisely this (especially popular amongst new MBAs): https://hbr.org/2017/02/you-should-consider-buying-a-small-b...

> Why are (some) Europeans surprised when they are not included in tech product débuts?

We had a brief, abnormal, and special moment in time after the crypto wars ended in the mid-2000s where software products were truly global, and the internet was more or less unregulated and completely open (at least in most of the world). Sadly it seems that this era has come to a close, and people have not yet updated their understanding of the world to account for that fact.

People are also not great at thinking through the second order effects of the policies they advocate for (e.g. the GDPR), and are often surprised by the results.

Exactly this. Unchecked administrative power in the Executive has left very dangerous tools laying about for would be autocrats: https://en.wikipedia.org/wiki/Unitary_executive_theory

This will certainly make life harder for the regulatory agencies, and I don’t want to minimize the difficulty that represents, but this decision reinstates an important check on the Executive’s power at a very critical juncture in our nation’s history.

> Still, it is weird. Where is sense of pleasure in other people's ( perceived ) misfortune coming from?

It’s gross, but an inherent part of human nature, particularly among the insecure.

If you don’t already know the word for it you’re one of today’s lucky 10,000. It’s a great one: https://en.wikipedia.org/wiki/Schadenfreude

I had a similar experience when LangChain first came out. I spent a good amount of time trying to use it - including making some contributions to add functionality I needed - but ultimately dropped it. It made my head hurt.

Most LLM applications require nothing more than string handling, API calls, loops, and maybe a vector DB if you're doing RAG. You don't need several layers of abstraction and a bucketload of dependencies to manage basic string interpolation, HTTP requests, and for/while loops, especially in Python.

On the prompting side of things, aside from some basic tricks that are trivial to implement (CoT, in-context learning, whatever) prompting is very case-by-case and iterative, and being effective at it primarily relies on understanding how these models work, not cargo-culting the same prompts everyone else is using. LLM applications are not conceptually difficult applications to implement, but they are finicky and tough to corral, and something like LangChain only gets in the way IMO.

Not really. We were lucky that we used a hybrid multi-page architecture that makes the transition easy to manage. HTMX and VueJS are being used on different views with different routes/endpoints, so we can re-write any VueJS-based views one at a time.

This architecture was chosen partly because I am a boomer who remembers when websites were HTML files uploaded via FTP and couldn't understand why we shouldn't just return HTML for pages that didn't need app-style functionality, partly because we wanted something easy to manage with a small number of devs (HTML+vanilla js shines here), and partly because SPAs were new-ish when we first began development and the ecosystem was very immature and a long way from coalescing around a single winner (as it has now with React).

So in this case, HTMX is used with HTML templates and ordinary routing, while VueJS is isolated to single purpose SPA-style views with their own endpoints (for example, WebRTC video). And I suspect that we will even be able to eliminate the Vue-based pages altogether with a combination of HTMX and vanilla js if we want to, with how good vanilla js has gotten.

Not a question, but I just wanted to say that HTMX is a joy to work. I’m using it in production to slowly replace VueJS wherever full SPA interactions aren’t needed, and it has made development so much simpler. HTMX is such a breath of fresh air from the over-engineered jenga tower that web dev has evolved into over the past decade.

Thank you!

This just strikes me as sharp elbowed negotiation by the Saudis. I bet good odds that this particular story about the Saudi treaty expiring will have proven to be a nothingburger a year from now (even if de-dollarization and multi-polarity in general continues to be a story).

It certainly would not be a free society. Though as with all things human, all of this has happened before and all of this will happen again:

"Charles II had re-turned to the English throne in 1660 and was appalled at the state of printing in his realm. Seditious, irreligious, pernicious, and scandalous books and pamphlets flooded the streets of London (among them the works of Milton and Hobbes)...[He] required that all intended publications be registered with the government-approved Stationers’ Company, thus giving the king his “royal prerogative”—and by extension, giving the Stationers the ultimate say in what got printed and what did not.

...it is not surprising to learn that the 1662 Act only met with partial success. One gets the sense that London in the late seventeenth century was a place where definitions of morality were highly subjective and authority was exercised in extremely uneven fashion."

https://dash.harvard.edu/bitstream/handle/1/17219056/677787....

> How do you defend against that? I'm not sure you do... A tool is a tool. I wouldn't want my CAD software saying, "I think you're trying to CAD a pipe bomb so I'm going to shut down now."

The core of the issue is that there are many people, including regulators, who wish that software did exactly that.

> The fact that you value a, b and c differently from the person does not mean it was not forced from their perspective.

I basically agree with you about the phenomenon you’re pointing out, so please don’t take this as overly salty, but:

Just because I believe that I can fly doesn’t make it so. So too is true of people believing that they are “forced” to work at Google, no matter what they might perceive.

Sure. Nature also forces me to eat. There are forcing functions everywhere, and if you ignore them, you will die. You have to participate in your own survival on some level.

How you do so is up to you. If you want to start a commune, go for it. If you prefer to work at FAANG, that’s cool too. We are very lucky to find ourselves in a society where we have this choice, and we should be careful about the extent to which we impinge on others’ ability to choose as well

As another poster correctly pointed out, no one is forcing you to specifically support your own survival at Google. People choose to work at Google because of greed and a desire for prestige, and because it is much easier to benefit from what already exists than it is to create something yourself. This is all perfectly rational, but framing the issue in lamentations about “force” and “democracy” is nothing but crocodile tears.

Plus if property taxes and so forth really were the issue, there is already a democratic remedy for the problem. Taxes are instituted by our society democratically, and they can be removed democratically as well. But I suspect that this would not rectify the OP’s concern, because it’s not really about democracy or freedom. Instead it’s about greed and envy for what others have that they do not.

No one is being forced to do anything. You are free to change jobs whenever you want. Or you can start a company of your own which embodies your ideals whenever you want. No one is stopping you.

Furthermore, the fact that you are not free to act with impunity, and do not have the right to demand that others indulge your preferences and suppress their own, is not in of itself anti-democratic. In a free society other people also have the right to structure their affairs as they see fit. And you have the right to choose to engage with them, or not.

It does. They also have filters you can apply to your query to further restrict the results for things like small web (blogs and niche forums, using the Marginalia index AFAIK) and forums (Reddit and so forth), so that you don’t have to resort to appending “Reddit” to your searches.

Kagi is basically $20/month to time travel back to 2014-era Google with extra features. It’s fantastic and well worth it. I can’t recommend it enough.

Edit: Upon checking my billing it’s $10/mo, not $20. All the better!

> If companies insist on always taking the easy way out, they continue to introduce a significant amount of risk. Since it’s currently possible to make huge buckets of money, by writing crap, we’re not incentivizing high-Quality engineering.

For the past 10-15 years there has been a strong culture of never writing code when a library on GitHub or NPM already exists, which has, in large part, contributed to this. In many cases using existing battle tested code is the right thing, but it was taken to an extreme where the avoidance of pulling down a bunch of random open source packages with questionable stability and longevity was often maligned as not invented here syndrome.

Now many of those packages are unmaintained, the job hopping SWEs who added them to the codebase are long gone, and the chickens are coming home to roost. The pendulum will probably swing too far in the other direction, but thoughtful companies will find the right balance.

> You can: increase the price for patenting something. Google will think twice before spending 100K for hello world.

This is a terrible idea. Google will not think twice before spending 100k for a patent on hello world, but startups and small businesses would be priced out of the patent system entirely. All this would do is further increase and entrench the power of large corporations to the detriment of startups, entrepreneurs, SMBs, and ultimately, consumers.

Other posters have covered the gist of the events, but what creates the impression of flailing for me is Apple taking a hard stance in one direction, underestimating the backlash, and promptly recanting and changing course in exchange for (as best as I can tell) no advantage or concessions. That happened with PWAs, and now has happened with Epic.

Maybe there were behind the scenes concessions made and I'm completely wrong, in which case I owe Apple management an apology. But my impression as an outsider is that there is no real strategy here; executives at the company seem to be throwing their weight around and picking fights at random, getting slapped down, and immediately retreating.

Between this and the PWA decision, Apple seems to be flailing. What is leadership doing?

Parameter counts notwithstanding, it’s an objectively funny outcome that Meta, Microsoft, and Google are all releasing cutting edge open models, while OpenAI keeps theirs closed source.

> Synthesizers are instruments. They allow for the same level of subtle input as strings or woodwinds.

Synthesizer instruments, yes. Sequencers, trackers, and modern DAWs like Ableton, absolutely not. They’re entirely different animals and are very much a musical analogue to “prompt engineering.” They have also enabled some incredible art that would be impossible if musicians were limited only to what they could play into a recorded (or midi) tape.

I suspect that the same will prove to be true of generative AI, and that those artists that embrace it will see a golden age of creativity. But there will also be many artists who see their craft diminished in value, much like what has happened to studio drummers. And I could be completely wrong, but either way the genie is out of the bottle, so we’re going to find out.