vitaelabitur

thanks a lot, much appreciated

Yeah, pretty accurate description of how depression looks.

I'm reminded of Lars von Trier's Melancholia. I don't recommend it to anyone but it's the best, most visceral portrayal of the condition I have found in any medium. Watching it was strangely cathartic.

Shameless plug for my structured LLM outputs handbook which is written in a similar spirit: https://nanonets.com/cookbooks/structured-llm-outputs/

Thanks for sharing, appreciate it.

Would you say it is convenient to use it for writing, say, 500 words at a go?

> And really, really boring and slow a lot of the time

If you only watch the story-driven scenes in Lawrence of Arabia, and skip the prolonged shots of the desert, you would miss out feeling the same vastness and heat Lawrence is feeling.

There is a limit to how much a film can make you think or feel. Films that reach the highest limits need "boring" voids in-between the primary scenes. These voids are not to ingest more, but to help digest what has been ingested in previous scenes, with subliminal scenes and silence that let the right thoughts and feelings grow.

The word you are looking for, if you are looking for one, is declinism.

We traded books for films, and now films for short videos, always moving towards what is easier to enjoy.

Quite a while ago, books became a taste that needs to be patiently acquired. Someone starting to read today is more likely to develop the taste by gradually easing into books that demand more and more. Say maybe Huxley -> Camus -> Wilde -> Dostoevsky.

Now that short clips are here, the same has happened to films. The uninitiated need to sit through Scorsese, Hitchcock, Wilder, Kubrick, Altman before attempting Fellini, Antonioni, Tarkovsky, Ozu, Resnais.

And by the way, someone who is naturally inclined to love films (or books) won't be affected, even today. Am I wrong? The way they are described here, I would crush these film students.

This is my all-time favorite mathematics book. It approaches calculus fundamentals through a series of fictional conversations between a teacher and a student.

It is easy to read, yet I remember it gave me an odd visceral sense of what calculus is (or maybe I just thought it did).

I am reading it again.

You are absolutely right — Let me know if you want to read my personal anecdote on "Dead Internet Theory"...

Yeah, I especially hate how paranoid everyone is (but rightly so). I am constantly suspicious of others' perfectly original work being AI, and others are constantly suspicious of my work being AI.

There's no advantage wrt output quality, but it can be more economical in some high-error regimes, with less LLM calls used in resampling (max 2 for most errors).

We used Docusaurus.

One of the authors here. I've read the paper. Brilliant work, especially the slicing implementation for denser token masks.

> Increasing context length by complaining about schema errors is almost always worse from an end quality perspective than just retrying till the schema passes.

Another way to do this is to use a hybrid approach. You perform unconstrained generation first, and then constrained generation on the failures.

Aren't LLMs just super-powerful pattern matchers? And guessing "taps" a pattern recognition task? I am struggling to understand how your experiment relates to intelligence in any way.

Also, commercial LLMs generally have system instructions baked on top of the core models, which intrinsically prompt them to look for purpose even in random user prompts.

I went home for holidays last month. One day, my mom had a complaint about her food delivery and raised a ticket in the app. She was assigned "someone" on chat, and she carefully typed her issue. Then, she got a call from the same "person" who asked her to explain her issue in detail. After the call, she came to me confused and frustrated. She said the "person" on the other end kept giving unrelated solutions, and signed off saying they were happy to have resolved her issue.

Of course, you know this "person" on the other end was an LLM, which I figured once she handed over her phone. I was livid, and despite having better things to do, wasted the next few hours sending a notice to the legal team. They paid a small change to shut down the issue.

Looking back, if the app had at least stated she was talking to a machine and given her an option to escalate to human support, the situation would not have deteriorated.

I feel LLMs can never be used for negative interactions like complaints, or transactional interactions like placing orders. Scope should be limited to answering factual, generic questions, like "What's my order's ETA?", etc.

I might try this if you link or briefly explain the supporting research on your landing page.

Yeah, but I tried switching to minified JSON on a semantic labelling task and saw a ~5% accuracy drop.

I suspect this happened because most of the pre-training corpus was pretty-printed JSON, and the LLM was forced to derail from likely path and also lost all "visual cues" of nesting depth.

This might happen here too, but maybe to a lesser extent. Anyways, I'll stop building castles in the air now and try it sometime.

I tokenized these and they seem to use around 20% less tokens than the original JSONs. Which makes me think a schema like this might optimize latency and costs in constrained LLM decoding.

I know that LLMs are very familiar with JSON, and choosing uncommon schemas just to reduce tokens hurts semantic performance. But a schema that is sufficiently JSON-like probably won't disrupt model path/patterns that much and prevent unintended bias.

Guilty. I was clearly influenced by the prose. Can we simply let bygones be bygones?

It's ironic I've been waiting for smart glasses with displays ever since I found them in books and films as a kid, but now I see them as artifacts signaling a dystopian future. That they are tied to Meta does not help.

I feel a lot of us are starting to reject new tech that seeks to aggressively deepen our digital immersion.

I beg to disagree.

I believe in the Orwellian rules religiously, but I like indulging in maximalism from time to time. It felt apt that the article uses a dense, historical style to help it evoke a part of cultural history bygone.

That said, it might be objectively bad writing or AI, and perhaps I just don't know any better.

I wouldn't be surprised to see all of us deeply fetishizing the analog experience in the coming years.

Activities like letter writing might get gentrified, with private businesses charging a premium for delivery.

What does your tool/service do? Write a blog on "how to do XYZ". Use your personal blog, website, or guest post on Medium, etc.

Be helpful. Share an easy and free way to do it without using your tool/service. Then introduce your tool/service. Share its advantages and some proof that it works. Be genuine and honest, don't oversell it. Acknowledge alternative tools/solutions. Distribute it on reddit, twitter, etc.

If your blog is extremely good, Google Search will pick it up in a couple of weeks.

That might be the case. Maybe I am writing from a place of nostalgia and subconscious declinism.

You are right, parents can still steer their children.

But in some cases like mine, parents might not have the time or inclination to do so, for whatever reasons.

I was able to discover Pickwick Papers, Hardy Boys, Harry Potter, Lord of the Flies, Spirited Away, Almost Famous, etc. on my own. If I were young today, I don't think I would discover the right things.

https://www.youtube.com/watch?v=ba5HllbvLf4, The Many Adventures of Winnie the Pooh (1977) Ending

I was a quiet kid. Books, shows, and films shaped my sensibilities and moral tastes.

I wonder if that remains possible today. Content designed solely for the dopamine hit seems to crowd out everything else.

Makes sense. The default assumption is indeed shifting. Today, you are guilty of using AI until proven innocent. Students and some artists might certainly benefit from what you demoed.

This is interesting, I'll play the devil's advocate a bit more.

1. What is stopping me from opening AI outputs on another screen and typing them myself?

2. Have you figured out how to extend this concept to music and photography?

I am struggling to understand the need for this. If AI output is indistinguishable, that could mean -

1. It is as good as human output. For this to happen with the current state of AI, a human still needs to be an expert to guide and review. In this case, I don't think it hurts anyone to use AI.

2. It is worse than human output but the humans consuming it don't know any better. The AI reproduces successful patterns to produce books, paintings, music, films, instead of producing novel ideas/concepts that add to society. This is a deeper problem, and humans in pop culture have been doing it for years without AI. And besides art, bad outputs like vulnerable code or inaccurate text get detected naturally sooner or later.

Here's what has worked for me on day 0:

What problem does your tool solve? Write a blog on "how to solve/do XYZ".

Share an easy and free way to solve the problem without using your tool. Then introduce your tool and share its advantages and drawbacks. Be genuine and honest, don't oversell it. Acknowledge alternative tools and solutions.

Quick tips on how to write well - https://codecrafters.io/blog/writing-for-developers