ChatGPT-4 is here with multi-modal functionality

So ChatGPT-4 was launched recently with some impressive features. Not only does it seem to do many of the things that ChatGPT-3.5 did (but better), but now it has what they’re calling “multi-modal” functionality, meaning that it doesn’t just accept text as an input.

OpenAI had a developer demo recently where they showcased some of the model’s new abilities. I’ve linked the part below where they show an impressive example of ChatGPT being able to read a photo of a hastily scrawled idea and produce some functioning code based on it :point_down:
ChatGPT-4 demo showing multi-modal functionality

People are already thinking of some mind-blowing applications. :exploding_head:

I’m sure we can put our heads together and think of some more…

I’ll start with something simple: auto-suggest alt-tags for images?

What about "find me the contact info for these people! (send images) < creepy lol

BTW, now that I think about it - you can already use ChatGTP / REWPert to gather contact info zoikes (NOT GDPR Compliant!)

Agreed that it’s creepy but in theory definitely a time saver! :grinning:

1 Like

A whole new world of spam! (If abused)

ChatGPT-4’s multi-modal functionality can be applied to the real estate industry in various ways, enhancing the experience for both buyers and sellers. Here are some potential applications:

  1. Property description generation: ChatGPT-4 could analyze images of a property and generate detailed, accurate, and engaging property descriptions for listings.
  2. Virtual staging: ChatGPT-4 could use property images to generate virtually staged rooms, allowing potential buyers to visualize how the space might look with furniture and decor.
  3. Automated property valuation: By analyzing images of a property and comparing them to similar properties in the area, ChatGPT-4 could estimate the property’s value.
  4. Neighborhood analysis: ChatGPT-4 could analyze images of a neighborhood and provide insights into its amenities, demographics, and overall feel, helping potential buyers make informed decisions.
  5. Image-based property search: Users could provide images of their dream home, and ChatGPT-4 could find similar properties on the market based on the visual features of the input images.
  6. Property maintenance suggestions: ChatGPT-4 could analyze property images and identify areas that need maintenance or improvements, helping property owners prioritize tasks and maintain their investments.
  7. Architectural style identification: ChatGPT-4 could recognize and describe the architectural style of properties, helping buyers and sellers better understand the unique features of a home.
  8. Construction progress monitoring: ChatGPT-4 could analyze images or video feeds from construction sites to track progress, identify potential issues, and ensure that work is being carried out according to plans and regulations.
  9. Smart city planning: ChatGPT-4 could analyze satellite images, urban maps, and other visual data to help city planners optimize infrastructure, green spaces, and transportation systems.
  10. Interactive virtual tours: ChatGPT-4 could guide users through a virtual tour of a property by providing information about specific features, answering questions, and offering insights based on the visuals.
1 Like

It would be interesting to see how ChatGPT-4 handles aerial/satellite imagery.

Could you, say, give it a satellite picture of an area and ask it to generate a neighborhood description along the lines of “Quiet tree-shaded cul-de-sacs with double side-walks. Most of the properties are large single family homes with double-garages facing the street.” ?

Did you check to see if these correspond to active (or previously active) agents?

There’s also a chance it could’ve interpreted “available” as “these aren’t in use”.

In its current state, GPT-4 is unable to analyze or generate descriptions based on images. I don’t know if you are already playing with it (if you have access via a 3/3.5+ account) but we’re still dealing with a language model AI, and ChatGPT-4 is currently unable to process or analyze images directly. Its primary function is to understand and generate human-like text based on the input it receives. Analyzing and generating descriptions from aerial or satellite imagery will require integration with an image recognition AI, which is separate from ChatGPT-4’s capabilities as is available, even within the beta, as of today.

As it stands, if one provides a text description of an area or a set of features they’d like ChatGPT-4 to include in a neighborhood description, ChatGPT-4 would be more than happy to generate a text description for them. However, it cannot do this based on a direct input of an image at present.

Saying AI will bring about a whole new world of spam is not inaccurate at all, but when I first read it I figured you meant a whole lot more spam, and I truly believe we will see the opposite!

We have learned from repeated studies on SMS marketing that fewer targeted messages that are important to the user, delivered at the right moment result in higher overall conversion, subscription rates, sales, volume, and, therefore, profit and ROI.

It seems reasonable that as we move from pure automation and the use of the “shotgun approach” we will see fewer legitimate companies’ text messages being delivered to our phones, because it’s better for their bottom lines, because we respond to it better, and spammers would be silly to not recognize the same thing. Scammers/hackers/bad actors as well.

We are used to talking about scam calls because they happen every day. They are just randomly calling 1,000 people to talk to 100, to convince 10 that they are real to have 1/2 of them should actually follow through by sending them money/gift cards, etc.

With AI - assuming it were “really good” would allow them to just call the 200 who are most likely to answer, get on the phone with 75, to convince 9 that they are real and have 1/2 of them actually follow through.

The fewer spam/scam calls/emails/texts we get, the less we will talk about it as a feature of everyday life, and the higher the effect of each message will be, just as legitimate companies have seen with SMS marketing.

Non-AI based computer programs have been used for years to automatically scrape and spam email addresses and other contact information. These programs, known as web crawlers or bots, search the internet for email addresses and other personal information by analyzing web pages, forums, and social media platforms. Once collected, this information is often used for spamming or sold to third parties.

A few examples of large-scale attacks, leaks, and usages are:

  1. River City Media Spam List (2017): River City Media, an email marketing firm, accidentally leaked a database containing 1.4 billion email addresses and additional personal information. The leak occurred due to a misconfigured backup system, which allowed spammers and cybercriminals to access the information.
  2. LinkedIn Data Breach (2012, 2016): LinkedIn, a professional networking site, experienced two significant data breaches. In 2012, around 6.5 million hashed passwords were leaked, while in 2016, a hacker sold a database containing 117 million LinkedIn users’ email addresses and passwords.
  3. Yahoo Data Breach (2013-2014): Yahoo suffered multiple data breaches, with the largest one occurring in 2013. In this breach, 3 billion user accounts were compromised, including email addresses, names, telephone numbers, and encrypted passwords.

With the increasing capabilities of AI, it is expected that spamming and targeting efforts will become more sophisticated and harder to detect. AI-powered tools can potentially:

  1. Identify potential targets more efficiently, by scraping not just email addresses but also social media activity and behavioral patterns.
  2. Craft highly personalized phishing emails and messages, based on the targets’ online behavior, preferences, and habits, increasing the likelihood of a successful attack.
  3. Bypass spam filters and security systems by continuously evolving and adapting to new detection techniques.

In conclusion, the use of AI tools by spammers will likely make their attacks more potent and challenging to prevent. Therefore, it is crucial for individuals and organizations to adopt robust cybersecurity measures and stay updated on the latest threats and defense strategies.