_vertigo

I lost all of my photos when I was a college student too. I was way too irresponsible to actually back anything up. Kind of a bitter lesson.

I don’t think that is a good example. No one is debating whether LLMs can generate completely new sequences of tokens that have never appeared in any training dataset. We are interested not only in novel output, we are also interested in that output being correct, useful, insightful, etc. Copying a sequence from the user’s prompt is not really a good demonstration of that, especially given how autoregression/attention basically gives you that for free.

The blog post discusses this point directly. Did you read it?

You couldn’t design a better system for incentivizing leaks if you were trying. Hell, the CEO literally said as much. Not sure how you can conclude the markets aren’t the problem.

Rhetorical question: why do non-insiders still bet in these markets? Surely, after all of the focus on insiders, people will begin to realize that betting without insider knowledge is a fool’s gambit..

> No schema changes. No new indexes. No query rewrites. Just telling Dapper the correct parameter type.

This was my exact thought process reading this. The business side of my company does not care or want to wait for complex solutions that sound cool to engineers. If anything, we have the opposite problem: convincing business stakeholders when complexity is in fact warranted.

We’re beyond parody or fiction at this point. This wouldn’t be out of place in Snow Crash

This is just a rant with zero actual content or insight. Maybe using an LLM would have made it more legible.

He produces value by refining a raw resource. I don’t want to go to the flea market on my Sunday morning, maybe I’ll pay his markup.

> Code is not an asset it's a liability, and code that no one has reviewed is even more of a liability.

Code that solves problems and makes you money is by definition an asset. Whether or not the code in question does those things remains to be seen, but code is not strictly a liability or else no one would write it.

I’m American and I don’t like Trump. If Italy did bomb San Francisco and you asked me what I thought of that, I’d say I disapproved.

If China invaded overnight and absconded with Trump, I’d say I disapproved even though I don’t like him.

If you invent a person and assume their identity, that's still a false identity.

> I always grew up with the assumption that everything on the internet is most likely fake.

How you interact with the internet is not really relevant to the discussion. The average person does not interact with the internet in the same way that people on this forum do, so that should not be the yardstick by which we judge this.

Every action is self-interested if you squint enough

There is a difference between hiding your identity for the purpose of privacy and for the purpose of deception.

There’s a difference between writing anonymously and assuming a false identity.

That’s stupid. We can have a law that protects users from bullshit like this. Lacking technical savvy should not mean you forfeit the right to your data.

Lies, damned lies, and statistics

Some guy writes a whimsical blog post about wanting to work at a company that takes pride in their work and the top comment criticizes the blog post because they didn’t specify explicitly in their 300 word post that the company has to not be evil too.

Jesus, we’re fucked aren’t we?

Yes, obviously, everyone knows that. When all you have to file is a 1040, reading one of the instructions documents is fine. When you have to use several forms it start to add up.

Please. Are you implying we need AI to the same degree we need clean water?

Your chemicals in river analogy only works if there were also a giant company straight out of “The Lorax” siphoning off all of the water in the river.. and further, the chemicals would have to be harmless to humans but would cause the company’s machines to break down so they couldn’t make any more thneeds.

Honestly, AI could have written this.

Well, that’s helpful to know so that other people can know to ignore what you write on this

So your take is that if they are therapists, it’s a conflict of interest, and if they aren’t therapists, they’re not qualified to make the assessment?

I think that depends on whether your definition of “doomerism” is the same as theirs.

All of your examples are tools that incrementally improve upon what came before and, importantly, still require the user to have an understanding of and experience with the underlying skill.

The risk being posed here is that AI may not land as an incremental improvement that still requires the user to maintain some understanding of the underlying skill.

We aren’t quite there yet with the current LLMs. You still need to have a base level of knowledge to use them effectively. But if they were just a little bit better, hallucinated just a little bit less, the value of actually knowing things goes way down.

What would the incentive be to learn the underlying skill or area if the LLM can handle things just fine on its own? Why not just let the LLM figure it out and do it? And at that point, it ceases to be a tool and starts to be something you are completely dependent on. That is a risk.

I am a much better "engineer" for having taken the advanced classes, but more to the point, I was a much better programmer for having taken the advanced classes.

The total amount of code I had written in my first two years of university was not very much; by the time I had taken some of the advanced courses I had written a lot more code and had thus become a better programmer (still not a good one, but better!). I had written probably 10x more code by the end of year 4 than the end of year 2 thanks to practicum courses, etc.

Yep, that's all true. Like I said, bungled at the level of fine details, pretty questionable if it was a good decision, but largely effective at actually getting people to come in.

People do work noticeably shorter days, a handful of people come in for less than 30 minutes (huge waste of resources and time), and yeah, there is a morale problem.

One thing I have noticed is that the more co-located a team is, the more people behave like pre-Covid — in at 9:30, out at 5. Satellite workers who are coming in are the most unhappy (for obvious reasons) and are the most likely to badge in, grab a coffee and check their emails, and then leave and finish the day at home.

Amazon's approach to RTO seemed broadly effective at getting workers into the office even though it was largely bungled at level of fine details.

1. Track badging 2. Enforce RTO via chain-of-command. SVPs, VPs, and directors have aggregate RTO compliance metrics. SVPs crack the whip on VPs to get their numbers up, VPs crack the whip on their directors, and directors crack the whip on their reports, and so on, all the way down.

The actual implementation of the policy was a mess and there was poor messaging especially at the beginning. I'm pretty sure it still doesn't take PTO or illnesses into account — if there's an issue with that, you're supposed to work it out with your manager — and from what I heard they simply handled the holidays by excluding the work weeks before and after Christmas from the metrics. But, ultimately, the chain of command approach appears to have worked very well in terms of actually getting people to come in.

edit: Just noticed from your profile that you also work at Amazon, so I guess there's nothing new in my comment for you :)

A toaster in 1950 would probably be closer to $250. Also, I don't like this counterargument because it holds that we should expect to be no better at creating high quality appliances today than in 1950 despite 73 years of advances in engineering and technology.