Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

First Impressions with GPT-4V(ision)

blog.roboflow.com

Post reply on HN
Print view View original

61–70 of 345 posts

Re: First Impressions with GPT-4V(ision)

#61

Post by troupo »

wokwokwok wrote: post #26

I’m impressed, technically, but this seems niche. Who holds their phone up and takes a photo then wants to know it was a photo of? That’s weird. If you don’t know what it is, wtf did you take photo? The obvious use here is natural language improvement / photo editing for photos, but this is just a stepping stone to that, and bluntly, as it stands… the examples really don’t shine… Great for the vision impaired. …not s…

> Who holds their phone up and takes a photo then wants to know it was a photo of?

I do. For plants, and occasionally for birds.

Re: First Impressions with GPT-4V(ision)

#62

Post by kossTKR »

troupo wrote: post #60
dottjt wrote: post #18

Earlier quoted context omitted.

Oh wow, I'm completely fucked as a front end developer.

Has AI been able to figure out conflicting and quickly changing product requirements and implement them on top of existing technical debt? You'll be fine.

But if this development continues AI will surely be able to just parse the entire frontend, then look at the repo, then look at the specifications, then when you ask for a specific feature it will instantly implement it gracefully.

I don't see why it wouldn't understand piles of hotfixes on top of each other, or even refactor technical debt in tight coupling with existing or historical specification.

Or is there a reason this is not going to happen in a few years?

Re: First Impressions with GPT-4V(ision)

#63

Post by zwily »

layer8 wrote: post #47

The discrepancy between the two answers regarding the set of coins is jarring. From the answer to the first question, one would assume that it can’t tell the currency. The answer to the second question shows that it actually can. The fact that LLMs don’t reflect a consistent inner model in that way, and hence the users’ inability to adequately reason about their AI interlocutor, is currently a severe usability issue.

I’ve gotten in the habit of asking chatgpt “are you sure?” So many times it will (correctly) correct itself, state that items are hallucinations, etc. It always makes me laugh.

Re: First Impressions with GPT-4V(ision)

#64

Post by steveads »

I use ChatGPT fairly frequently, but am frequently annoyed at how hesitant it is to answer any question that is somewhat subjective.

From it's Pulp Fiction response: "...However, whether you personally consider "Pulp Fiction" a good movie depends on your taste in film."

To avoid this noise I often add something like this to my queries "skip any preamble or qualifications about how x is subjective" and the results are much better.

Re: First Impressions with GPT-4V(ision)

#65

Post by op00to »

hypertele-Xii wrote: post #39
cs702 wrote: post #10

Sure, there are a few edge-case failures and mistakes here and there, but I can't help but be in awe . AWE. Let me state the obvious, in case anyone here isn't clear about the implications: If the rate of improvement of these AI models continues at the current pace, they will become a superior user interface to almost every thing you want to do on your mobile phone, your tablet, your desktop computer, your car, your…

So you won't be able to do anything without Internet connection to the AI mainframe? No thanks.

Until the AI mainframe runs on your $device

Re: First Impressions with GPT-4V(ision)

#66

Post by dottjt »

troupo wrote: post #60
dottjt wrote: post #18

Earlier quoted context omitted.

Oh wow, I'm completely fucked as a front end developer.

Has AI been able to figure out conflicting and quickly changing product requirements and implement them on top of existing technical debt? You'll be fine.

Not yet, but give it time. The concept of self-driving vehicles even a decade ago seemed absurd (or even AI for that matter), but now it all seems like a reality.

And that's not even taking into account all the advances we'll have with AI within the next decade that we haven't even thought about.

But yeah, you may be right.

Re: First Impressions with GPT-4V(ision)

#67

Post by extr »

One thing I found incredible recently: The advanced data analysis version of GPT-4 can apparently do OCR by writing code to run pytesseract and then interpreting the result. I was seeing what would happen if I gave it an image and found it surprising that it actually worked.

Re: First Impressions with GPT-4V(ision)

#68

Post by altcognito »

Andoryuuta wrote: post #48
yeldarb wrote: post #19

Earlier quoted context omitted.

If true, this is wild. I suppose a human could spend 10 seconds per Captcha, so they could do 360 per hour. Add some overhead for not being operating at peak performance every minute of every hour & call it 250. Let's say you can hire someone for $2, that works out to a bit over a penny per Captcha. I don't think OpenAI has published pricing for GPT-4 Vision yet, but if we assume it's on par with GPT-4, and uses only…

You'd be surprised, or perhaps horrified, by how cheap (self-proclaimed) human-based captcha solving services are. If you just search for "captcha solving service" the first few results that come up offer 1000 solves of text-based captchas for Whether these are actually human based, or just impressive OCR services, it seems like they are still much more cost effective than GPT-4 is for now.

I imagine they are a mix.

Re: First Impressions with GPT-4V(ision)

#70

Post by bastawhiz »

wokwokwok wrote: post #36
otoburb wrote: post #29

Earlier quoted context omitted.

>> Who holds their phone up and takes a photo then wants to know it was a photo of? >> Great for the vision impaired. Yes, this is great for the estimated 285 million vision impaired people around the world[1]. [1] https://www.bemyeyes.com/about

Did you read my comment? I literally said that it’s for vision impaired. That’s great. …but it’s niche . I’m sitting on my couch right now and I can think of like 20 things I could chat to chatgpt about. I can see literally nothing in my visual range want to take a photo of and run image analysis over. It’s like Shazam. Yes, it’s useful, but, most of the time, I don’t need it. I would argue this is true for this, for…

There's enough vision-impaired people in the world to equal the population of Japan, Korea, and Vietnam combined. And beyond those people who would get obvious utility, this is essentially Google Lens on steroids—I simply can't figure how you could call this "niche". Maybe you won't use it multiple times per day, but plenty of people will. Hell, just now I was wondering why the leaves on one of my plants are starting to brown and could have used this.
Post reply on HN