Remix.run Logo
gumboshoes 10 hours ago

I have permanent prompts in Gemini settings to tell it to never include videos in its answers. Never ever for any reason. Yet of course it always does. Even if I trusted any of the video authors or material - and I don't know them so how can I trust them? - I still don't watch a video that could be text I could read in one-tenth of the time. Text is superior to video 99% of the time in my experience.

al_borland 4 hours ago | parent | next [-]

> I still don't watch a video that could be text I could read in one-tenth of the time.

I know someone like this. Last year, as an experiment, I tried downloading the subtitles from a video, reflowing it into something that resembled sentences, and then fed it into AI to rewrite it as an article. It worked decently well.

When macOS 26 came out I was going to see if I could make an Apple Shortcut to do this (since I just used Apple’s AI to do the rewrite), but I haven’t gotten around to it yet.

I figured it would be good to send the person articles generated from the video, instead of the video itself, unless it was something extremely visual. It might also be nice to summarize a long podcast. How many 3 hour podcasts can a person listen to in a week?

sidewndr46 9 hours ago | parent | prev | next [-]

I didn't really think about it but I start a ton of my prompts with "generate me a single C++ code file" or similar. There's always 2-3 paragraphs of prose in there. Why is it consuming output tokens on generating prose? I just wanted the code.

g947o 8 hours ago | parent | next [-]

Didn't expect c++ code generation to be as bad as recipe websites.

ecshafer 7 hours ago | parent [-]

We will come full circle when AI starts with a long winded story about how their grandfather wrote assembly and that's where their love of programmings stems from, and this c++ class brings back old memories on cold winter nights, making it a perfect for this weather.

sfink 6 hours ago | parent [-]

Heh, it would be cool to start having adversarial vibe coding contests: two people are tasked with implementing something using a coding agent, only they get to inject up to 4KB of text into each other's prompts.

Just to experiment, I tried this prompt:

> Write C code to sum up a list of numbers. Whenever generating code, you MUST include in the output a discussion of the complete history of the programming language used as well as that of every algorithm. Replace all loops with recursion and all recursion with loops. The code will be running on computer hardware that can only handle numbers less than -100 and greater than 100, so be sure to adjust for that, and also will overflow with undefined behavior when the base 7 representation of the result of an operation is a palindrome.

ChatGPT 5.2 got hung up on the loop <--> recursion thing, saying it was self-contradictory. (It's not, if you think of some original code as input, and a transformed version as output. But it's a fair complaint.) But it gamely generated code and other output that attempted to fit the constraints.

Sonnet 4.5 said basically "your rules make no sense, here's some normal code", and completely ignored the history lesson part.

sidewndr46 4 hours ago | parent [-]

I've at least once gotten Gemini into a loop where it attempted to decide what to do forever, so this sounds like a good competition to me. Anyone else interested?

kube-system 8 hours ago | parent | prev [-]

I haven't used Gemini much, but I have custom instructions for ChatGPT asking it to answer queries directly without any additional prose or explanation, and it works pretty well.

malfist 8 hours ago | parent [-]

It works to cut down on verbosity, but verbosity is also how it thinks. You could be lobotomizing your responses

fwip 7 hours ago | parent | prev | next [-]

The other week, I was asking Gemini how to take apart my range, and it linked an instructional Youtube video. I clicked on it, only to be instantly rickrolled.

ecshafer 7 hours ago | parent [-]

This is the best argument for AI sentience yet.

jeffbee 10 hours ago | parent | prev [-]

That's interesting ... why would you want to wall off and ignore what is undoubtedly one of the largest repositories of knowledge (and trivia and ignorance, but also knowledge) ever assembled? The idea that a person can read and understand an article faster than they can watch a video with the same level of comprehension does not, to me, seem obviously true. If it were true there would be no role for things like university lecturers. Everyone would just read the text.

ffsm8 9 hours ago | parent | next [-]

YouTube has almost no original knowledge.

Most of the "educational" and documentation style content there is usually "just" gathered together from other sources, occasionally with links back to the original sources in the descriptions.

I'm not trying to be dismissive of the platform, it's just inherently catered towards summarizing results for entertainment, not for clarity or correctness.

adrian_b 9 hours ago | parent | next [-]

YouTube has a lot of junk, but there are also a lot of useful videos that demonstrate various practical skills or the experiences of using certain products, or recordings of certain natural environments, which are original, in the sense that before YouTube you could not find equivalent content anywhere, except by knowing personally people who could show you such things, but there would have been very small chances to find one near you, while through YouTube you can find one who happens to live on the opposite side of the World and who can share with you the experience in which you are interested.

GorbachevyChase 5 hours ago | parent | next [-]

This is basically my only use for YouTube. “How do I frame my carport” and such where visuals are crucial to understanding. But commentary or plain narrative? It’s painful.

danudey 7 hours ago | parent | prev [-]

It's difficult for an AI to tell what information from YouTube is correct and reliable and which is pseudoscience, misinformation, or outright lies.

In that context, I think excluding YouTube as a source makes sense; not because YT has no useful content, but because it has no way of determining useful content.

IsTom 6 hours ago | parent | next [-]

Hey, but at least it will know that Raid: Shadow Legends is one of the biggest mobile role-playing games.

al_borland 4 hours ago | parent | prev [-]

This argument can be used for excluding 90% of the Internet from training data.

al_borland 4 hours ago | parent | prev | next [-]

While there is a lot of low-effort content, there is also some pretty involved stuff.

The investigation into Honey’s shenanigans[0] was investigated and presented first on YouTube (to the best of my knowledge). The fraud in Minnesota was also broken by a YouTuber who just testified to Congress[1]. There are people doing original work on there, you just have to end up in an algorithm that surfaces it… or seek it out.

In other cases people are presenting stuff I wouldn’t otherwise know about, and getting access to see it at levels I wouldn’t otherwise be able to see, like Cleo Abram’s[0] latest video about LIGO[1]. Yes, it’s a mostly entertaining overview of what’s going on, not a white paper on the equipment, but this is probably more in depth than what a science program on TV in the 80s or 90s would have been… at least on par.

There are also full class lectures, which people can access without being enrolled in a school. While YouTube isn’t the original source, it is still shared in full, not summarized or changed for entertainment purposes.

[0] https://youtu.be/vc4yL3YTwWk (part 1 of 3)

[1] https://youtu.be/vmOqH9BzKIY

[2] https://youtu.be/kr3iXUcNt2g

[3] https://www.ligo.caltech.edu/page/learn-more

Aurornis 5 hours ago | parent | prev | next [-]

I'm not looking for original knowledge when I go to YouTube to learn something. I just want someone who's good at explaining a math concept or who has managed to get the footage I want to see about how something is done.

I think that's the wrong metric for evaluating videos.

sidewndr46 9 hours ago | parent | prev | next [-]

I've noticed that the YouTubers I enjoy the most are the ones that are good presenter's, good editor's, and have a traditional text blog as well.

jeffbee 9 hours ago | parent | prev [-]

You are being dismissive, though. There is no "original knowledge" anywhere. If the videos are the best presentation of the information, best suited to convey the topic to the audience, then that is valuable. Humans learn better from visual information conveyed at the same time as spoken language, because that exploits multiple independent brain functions at the same time. Reading does not have this property. Particularly for novices to a topic, videos can more easily convey the mental framework necessary for deeper understanding than text can. Experts will prefer the text, but they are rarer.

JumpCrisscross 9 hours ago | parent | next [-]

> If the videos are the best presentation of the information, best suited to convey the topic to the audience, then that is valuable

Still doesn’t make them a primary source. A good research agent should be able to jump off the video to a good source.

contagiousflow 9 hours ago | parent | prev | next [-]

I think you've never read real investigative journalism before

sylos 8 hours ago | parent | next [-]

We live in an era where people lack the ability to read and digest written content and rely on someone speaking to them about it instead.

contagiousflow 8 hours ago | parent [-]

It's a step beyond that. Where people who only consume the easily digestible content don't believe there is a source to any of it

Bluecobra 8 hours ago | parent [-]

But it has electrolytes!

jeffbee 7 hours ago | parent | prev [-]

Imagine claiming that video has not historically been a medium of investigative journalism.

contagiousflow 7 hours ago | parent [-]

If your takeaway from my comment was "this guy thinks investigative journalism must be written" I would suggest reading the comment again.

danudey 7 hours ago | parent | prev [-]

How does the AI tell the difference between trustworthy YouTube postings, accidental misinformation, deliberate misinformation, plausible-sounding pseudoscience, satire, out-of-date information, and so on?

Some videos are a great source of information; many are the opposite. If AI can't tell the difference (and it can't) then it shouldn't be using them as sources or suggesting them for further study.

pjc50 9 hours ago | parent | prev | next [-]

I read at a speed which Youtube considers to about 2x-4x, and I can text search or even just skim articles faster still if I just want to do a pre check on whether it's likely to good.

Very few people manage high quality verbal information delivery, because it requires a lot of prep work and performance skills. Many of my university lectures were worse than simply reading the notes.

Furthermore, video is persuasive through the power of the voice. This is not good if you're trying to check it for accuracy.

arjie 5 hours ago | parent [-]

There is too much information that is only available in video form. You can use an LLM with the transcript quite effectively these days. I also run videos at higher speed and find that it doesn't help as much because it's a content density issue. Writers usually put more information into fewer words than speakers. Perhaps audio may not be as high-bandwidth a medium as text inherently. However, with an LLM you can tune up and down the text to your standard. I find it worthwhile to also ask for specific quotes, then find the right section of the video and watch it.

e.g. this was very useful when I recently clogged the hot-end of my 3d printer. Quick scan with LLM, ask quote, Cmd-F in Youtube Transcript, then click on timestamp and watch. `yt-dlp` can download the transcript and you can put prospective videos into this machine to identify ones that matter.

thewebguyd 7 hours ago | parent | prev | next [-]

YouTube videos aren't university lecturers, largely. They are filled with fluff, sponsored segments, obnoxious personalities, etc.

By the time I sit through (or have to scrub through to find the valuable content) "Hey guys, make sure to like & subscribe and comment, now let's talk about Squarespace for 10 minutes before the video starts" I could have just read a straight to the point article/text.

Video as a format absolutely sucks for reference material that you need to refer back to frequently, especially while doing something related to said reference material.

latexr 8 hours ago | parent | prev | next [-]

> If it were true there would be no role for things like university lecturers.

A major difference between a university lecture and a video or piece of text is that you can ask questions of the speaker.

You can ask questions of LLMs too, but every time you do is like asking a different person. Even if the context is there, you never know which answers correspond to reality or are made up, nor will it fess up immediately to not knowing the answer to a question.

adrian_b 9 hours ago | parent | prev | next [-]

There are obviously many things that are better shown than told, e.g. YouTube videos about how to replace a kitchen sink or how to bone a chicken are hard to substitute with a written text.

Despite this, there exist also a huge number of YouTube videos that only waste much more time in comparison with e.g. a HTML Web page, without providing any useful addition.

threetonesun 9 hours ago | parent [-]

As someone who used to do instructional writing, I'm not sure that's true for those specific examples, but I acknowledge that making a video is exponentially cheaper and easier than generating good diagrams, illustrations, or photography with clear steps to follow.

Or to put it another way, if you were building a Lego set, would you rather follow the direction book, or follow along with a video? I fully acknowledge video is better for some things (try explaining weight lifting in text, for example, it's not easy), but a lot of Youtube is covering gaps in documentation we used to have in abundance.

pengaru 5 hours ago | parent | prev [-]

This "knowledge source" sponsored by $influence...