Hacker News Reader

Comment by rcarr | Hacker News Reader

The AI/ChatGPT hype is starting to piss me off, it seems like about 50% of HN is now articles about AI. Meanwhile, stuff like the genuinely incredible Unreal Engine 5.2 demo that hit Youtube six days ago only got 14 upvotes on here. What the fuck is going on?

I've been using ChatGPT for the last week or two and it's not got a single coding question I've asked it right. Seems alright as a 'rubber duck' for generating ideas and seems okish for creative writing but for not a hell of a lot else at the minute.

The visual AI art stuff does seem worth the hype though but yeah, I'm feeling burnt out on this shit too. Based on everyone I speak to, I think the majority of people are. The pandemic probably didn't help.

View on HN

We do have a lot of hype and buzz. There's a giant AI hype bubble right now and it's going to pop and bring us back to reality.

But at the same time, people are flipping out because it is, well, machine learning. I was following AI, mostly as a hobby interest, though I did some undergrad courses, before it got all buzz-wordy. Just ten, even five years ago, some of the stuff being demonstrated, were the sort of thing I thought would take many decades to develop, and that I may not see in my lifetime. Speech recognition, machine translation, machine vision, for example.

For more than half a century, good machine translation and speech recognition were the holy grail of AI. An incredible amount of work went into them. Specialized systems, grammars and parsers, programmed general knowledge databases. Collecting and correlating enormous aligned bilingual text corpuses with phonetic annotation. After several decades of such work, it got us to mediocre-quality Google Translate style services that were the norm until recently.

Now, without any of that specialized work, with a much more general learning algorithm, there are translation systems that are superior to anything humans designed to specifically be machine translators. It just falls out accidentally from the large language models -- they can also, incidentally, translate, and do better than anything specifically designed to translate ever did.

> Speech recognition, machine translation, machine vision, for example. All basically solved problems in the last few years.

Is this true or is there just a massive gulf in application at the moment?

For one trivial example take something like indexing historical newspapers that you would really expect to be in the class of “solved problems” because it’s largely typeset, but the commercial offerings [1,2,3] are just chock full of errors and there is absolutely nothing I can apply off the shelf that gets better results without a ton of extra effort and compute.

1. Newspapers.com 2. Newspaperarchive.com 3. Genealogybank.com

> Is this true or is there just a massive gulf in application at the moment?

It is only true if you drink the cool aid.

Speech recognition - Siri still have major issues undestanding me. Youtube text2speech routinely mistranscribes because it simply have no understanding of the language.

Machine translation is hilarious at best and dangerously wrong at worst.

Machine vision still could not spot the difference between pastry and furry animals last time I checked.

All of these are examples of non-working over hyped tech. It is not a list of "basically" solved problems.

My company has replaced hundreds of human transcribers with a speech to text model.

DeepL is actually really good a lot of the time.

When was the last time you checked the status of machine vision, because the problem you suggested is not hard for it anymore.

Your company can only pull the rug like this because the public is not very picky it seems.

A real human transcriber still outperforms any automated system in existence.

> Your company can only pull the rug like this because the public is not very picky it seems.

My company performed extensive accuracy testing and lets people choose to use a human transcriber if they want. Most people are perfectly happy to use the machine transcription. You are just seem bitter. Did an AI steal your wife?

I must have drank the cool aid because when I talk to my AI assistant, it bloody well understands me a lot better than my wife :D.

Machine translation still makes a few mistakes but hardly more than a human.

Machine vision: I worked on a factory where a machine would approve/reject products passing through at a ridiculously high speed, and it never got one wrong. This was 15 years ago.

Your experience is completely the opposite of mine.

> Youtube text2speech routinely mistranscribes

Isn’t text2speech the opposite of transcribing?

Speech2text would be transcribing, and text2speech would be speech synthesis.

Anyway, assuming you meant speech2text: I found YouTube‘s transcribing quite good. It even understands stuff that is inaudible to me (especially in movie snippets). Of course it’s not perfect, but neither am I.

TTS is where I started when I was working on this sort of thing, I commonly just say text to speech to mean either direction.

Thanks, I mean speech2text, the youtube auto-caption feature specifically. Perhaps you have enjoyed it. I regularry use it due to bad hearing, and it is hilariously often mistranslating stuff that a human never would, simply because it does not understand context. It is a dumb system.

> Machine translation is hilarious at best and dangerously wrong at worst.

I picked a random passage from a novel in French I am currently reading. ChatGPT translated the three paragraphs I ran it on correctly; there are no major quibbles to be had. It is good, coherent English, a correct translation, which closely follows the French original, even capturing some of the poetic imagery effectively.

I'm sure after another paragraph or two there will be a weird screw-up. And there's no consistency in a running translation of any length. Etc. Yes, it's not perfect. Not fully human-equivalent.

Still. I remember when machine translation like I just did was the realm of science fiction. And I thought it would remain science fiction for a long time. The fact that such a thing isn't mind-blowing shows how far things have come, hasn't it?

> Speech recognition - Siri still have major issues undestanding me.

I am using speech-to-text AI transcription every day. It's been revolutionary for me. I am hard of hearing. The cutting edge is Whisper, and it is leaps and bounds over the state-of-the-art just a year ago: https://github.com/openai/whisper

> Speech recognition - Siri still have major issues undestanding me.

Have you tried OpenAI Whisper, especially with its "large" model? Siri and Youtube shouldn't be the yard sticks to judge the entire field, they both have unique hardware constraints and they're far from the state of the art.

You have some odd issue here... You're thinking that any technology is going to be perfect, and it's not, humans are not either. Don't put your base line as perfect, but at the rate of human failure.

Siri isn't even the latest models that have a much lower error rate.

Whisper is better at speech recognition than humans. Learn about the SOTA instead of mentioning bad mainstream products made years ago.

Google's is the best business the world and chatgpt opens up round 2 over who gets the hold the rains of that beast.

The hype originates in the business comminity, and it's about all the money that will be made, not technology.

The technology part is over as far as business is concerned. It's good enough to get everyone in the world to type their every need into this textbox rather than another. Everything else is downstream from there.

Transformers, rotators, discombulators, those things get nerds excited, but what's whipping people into a frenzy is the race to GOOG 2.0

>Google's is the best business the world

I think at the same time you're ignoring the deep problems that are affecting google right now. Google makes their money on ads, and AI could deeply effect that market causing them drastic profitability drops.

Doesn't matter what you can make, companies don't cut their own throat in the short term, even if it will kill them in the long term.

I've not seen the Unreal Engine demo, but I think GPT-4 is a big leap in a much more general purpose innovation than an improvement in a video game engine. The improvements to the GPT system have created quite a lot of buzz, as we are all still trying to figure out what it means.

One thing to note is that ChatGPT is still GPT-3.5 sometimes. Their page says it is GPT-4 if capacity is available, but I suspect GPT-4 might be pretty busy right now. You can pay $20/mo for GPT-4, which I am close to doing just to try it more.

But from what I have heard, GPT-4 is going to be better at answering those coding questions, and plenty of people here are paying for that, so this might explain why you are underwhelmed and other people are enthusiastic. I was pretty dismissive of this stuff, recently tweeting that ChatGPT made for a poor rubber duck, but my friend has convinced me to pay a little more attention, and now I am seeing some of the value people are talking about.

I'll bare that in mind, I'm not convinced by what I've seen so far of it though.

The two main Unreal demos for anyone interested:

https://www.youtube.com/watch?v=Dj60HHy-Kqk

https://www.youtube.com/watch?v=pnaKyc3mQVk

It looks good but doesn't really blow me away. Ten seconds into the demo I keep thinking, I've seen stuff like this before and the jungle immediately seems to be dead, not a single leaf moving etc.

I mean the level of detail is amazing but there's still the question of how much value does it add to the game in the end. And: how good does MetaHuman, impressive as it is, need to get to escape the Uncanny Valley? Or does it ever?

Most jungles are very still unless you go to the top of the canopy. All the plants block the wind.

Not surprised this didn't do well. YouTube videos are a time commitment, plus you need sound. I'm almost never somewhere where I can watch a video on the spot.

I suspect most on this site prefer skimable text content. I personally hate videos

Farcry achieved this quality 14 years ago, and the environment is richer and more alive:

https://youtube.com/watch?v=s6j-2vrGkc0

I paid for GPT-4 last night. I ended up having a long conversation with it giving me really good advice on how to proceed with my non profit organization. There are a handful of writers and philosophers, both historical and contemporary, who I have really appreciated over the years. Last night I asked it to help me understand how my work relates to these philosophies, what kind of people might be receptive to my ideas and why, which countries have the right mix of people to warrant more direct outreach, how to craft my message for the specific values of those individual countries, and more. When asked, it even gave me encouragement as I often feel worn out from shouldering the burden of an atypical non-profit partly alone.

It feels like a narrow general intelligence with a lot of knowledge. Narrow in the sense that it is only text based, but general in the sense that it did very well handling these complex questions. It felt like an expert in all the random little things I have personally been focused on in my philosophical journey.

I've just watched the first Unreal demo. What's funny is I grew up in the redwoods, before the drought at a time when it was considered a temperate rainforest. As I have gotten older I've had to move a little farther from the redwoods, and I feel like I have left a piece of me there. I have always been interested in video game experiences which reproduce the feeling of being in the forest. Recently I got a VR headset and have been having conversations with my friends about how much I'd like to create more accurate forest environments so I can feel like I am there even when I'm not able to get out there.

So I am an ideal audience to appreciate these video game engine demos. And they look nice! But GPT-4 is much more significant to me. It is providing me detailed feedback and advice on how I can manage my career and my life in ways which are really energizing. When I write scripts for my youtube videos, I know I will be able to ask it for specific advice on topics to consider, or parts of it which I might be able to improve. I am not going to do what some high school students do and have it do all of my writing for me, as my personal voice and expression matters a lot to me. But what I will do is share with it my writing and ask for feedback based on my ideals and goals. It can serve as a teacher providing feedback on my work.

That is much, much more valuable to me than an improved video game engine. This is going to be an ally in my career and help me feel as though I don't have to do all the thinking alone.

Heh bear in mind they limit how much you can use gpt4 before telling you to wait for 2 hours. Right when my plot for a 4th LOTR book called “Redemption of Sauron” was getting good. Honestly the best use of gpt that I have come up with.

just use the API which has no limit

> The AI/ChatGPT hype is starting to piss me off, it seems like about 50% of HN is now articles about AI.

That's a bit like complaining in the late 90s that a lot of the articles were about the internet taking off.

I think maybe it's a little overhyped. I suspect that some of the incredible performance of GPT4 is down to good engineering and is going to be less impressive when used for bigger real world problems, and it won't improve as fast as some people on here think it will.

However, it's clearly a quantum leap* in technical progress. It's likely going to change the world as much or more than the widespread adoption of the internet, and in a shorter timeframe. And it is fascinating to many of us.

*in the colloquial sense of the term. A gigantic and sudden advance.

> seems okish for creative writing

It may seem like that, but it's neither ok nor even okish in reality. Whatever time you save on not writing, you'll lose double that editing, and the end result will be average at best. Of course, average is still better than a sizeable chunk of human population is capable of producing, and sometimes it's enough. For anything of higher quality, it's hard to take it seriously. You can use it as a thesaurus or a dictionary in a bind, or for estimating how your text will be understood by less well-read readers. For creative, but factual writing it's even worse - you never know if it started hallucinating already or will in a moment. I'm not saying it's not useful - in specific circumstances it can be, but the utility for creative writing is, IMO, marginal.

Disclaimer: I'm not a professional writer, but an avid reader and a bit of a blogger. I'd love to be wrong on this and corrected by a real writer - especially if they can share in what way they use ChatGPT that's worth it :)

> The AI/ChatGPT hype is starting to piss me off, it seems like about 50% of HN is now articles about AI.

It wouldn’t even bother me if it was just normal hype, but it’s the “omz this is going by to usher in a utopia that just happens to be like my favorite sci fi book” crap

I went and watched that demo, and unless I found the wrong one, it looked like a 6 minute Rivian advert. The funny bit is that made it feel… scammy.

I am in agreeemEnt with this. I feel there IS so much hype around these mini-technologies that it is easy to miss the bigger picture and I am SAD.

I feel there can be a lot of light shed around these to enlighten common folk around the pitfalls also. But such is the nature of this tech at this point. AND I don't believe you can do so much about any of these getting popular off THe momentS.

> Meanwhile, stuff like the genuinely incredible Unreal Engine 5.2 demo that hit Youtube six days ago only got 14 upvotes on here.

Epic shows incredible looking demos periodically, but you don't get to see games that actually look like that until a decade+ later.

I am tired as well. It has evident limitations and is not the crystal ball people believe.

Text is a ubiquitous part of our lives, whereas Unreal Engine is utilized by a relatively smaller group of people. It seems the hype surrounding it is proportionate.

The unreal engine demo still had that uncanny valley around the mouth when talking. If it didn't have that it would be amazing.

> it's not got a single coding question I've asked it right

Have you used common prompting techniques like CoT and self-reflection?

additionally it gets a lot better via iteration, eg. feeding it back any error messages and asking it to try again.

Reply on news.ycombinator.com