Study Reveals ChatGPT-4's Remarkable 'Theory of Mind' Abilities, Outperforming Previous Models

A new study published in Proceedings of the National Academy of Sciences reveals that many large language models (LLMs) like ChatGPT are showing “theory of mind” abilities which are seen in humans. While testing ChatGPT-4, the researchers found that it can perform 75% of the tasks that a six year old can too. This shows that LLMs are showing improvement in their reasoning abilities. Theory of mind refers to the ability of humans to understand beliefs, emotions and mental states of other people, and then they interact with them on the basis of that. In humans, this ability is developed in their early childhood and continues to develop throughout their lives.

The researcher, Michal Kosinski, said that LLMs can predict preferences of users based on what websites they visits, what products they purchases, their music choices and other behavioral data. While predicting the behaviors, it is also important to know the psychological processes of the individuals. For the study on LLMs, the researcher used a false-belief task, a psychological test, to understand the ability of LLMs to predict responses.

Two types of tasks, the Unexpected Contents task and the Unexpected Transfer task, were used for the false-belief test. In the Unexpected Contents task, a subject sees an object with a misleading title and assumes the misleading title to be accurate. In an Unexpected Transfer task, an object gets moved without the subject knowing and the subject searches for the object in the same place. The LLMs tested had to predict and conclude what a human would do if he encountered these two situations. Kosinski evaluated 11 LLMs and created 40 false beliefs to test them. Each false-belief scenario targeted the model's comprehension and understanding of the real world.

The results of the tests showed that GPT-1 and GPT-2 weren't able to solve false-belief tasks, concluding that earlier models of ChatGPT don't have ability to do so. On the other hand, 20% of the tasks were performed accurately by ChatGPT-3 which is equivalent to tasks performed by a three year old. The LLM with the best performance was ChatGPT-4 which was able to complete 75% of the tasks accurately. It predicted 90% of Unexpected Contents tasks while 60% of Unexpected Transfer tasks. The results also showed that ChatGPT-4 was able to adjust its predictions based on context and reasoning instead of simple patterns.

Image: DIW-Aigen

Read next: AI Chatbots Provide Non-Judgmental Mental Health Support but Struggle with Memory and Complex Issues
by Arooj Ahmed via Digital Information World

‘Nudify’ Apps Continue Advertising on Meta’s Platforms Despite Company’s Attempt of Banning Adult Content

Meta seems to have its hands full with a lot of problems after entering 2025.

The tech giant has clear set guidelines about adult content sharing on its popular social media apps including Facebook and Instagram. However, that does not seem to be enough to stop AI-based nudity apps from spreading, as per 404Media.

These apps use AI technology to produce naked or explicit images. The shocking part is how they continue to be marketed across Instagram and Facebook despite Meta’s efforts to crack down on adult content.

The nudity apps create images of popular celebs and influencers without consent. Most of them are explicit in nature and seem very real at first glance. If that’s not all, their ads are doing great on Meta’s apps as they. Users fall for them and engage with them, allowing them to flourish and expand. One prominent example is Crush AI.

This image modifier platform gets most of its users from Facebook and Instagram, both under the ownership of Meta. Furthermore, data taken from Similarweb last month shed light on it getting a quarter of a million visits for the service. That’s nearly 90% of the traffic coming from Meta’s apps.

The nudify app makes it very clear from the start what users can expect from its service. The ads feature real people such as models and celebrities, not to mention famous OnlyFans stars. You can upload any image or simply erase a popular star’s clothing with the service.

Meta is trying its best to crack down on the ads but it seems like its efforts might be in vain. Apps like Crush AI produce fake profile pictures through AI and run new ads through those pages. Interestingly, every ad markets unique domain names which redirect the app when you click on it.

Reports from Mashable shared more on the matter and how AI apps and deepfakes keep using the advertising platform to give rise to more traffic and money from AI-based adult material. The ads go against the company’s policies and this firm removes them altogether after becoming aware of certain ads.

Meta, on the other hand, uses another standard for moderating ads. Content shared by users on Facebook and Instagram gets detected by Meta automatically and therefore removed as a result. However, a lot of the content goes unnoticed when published on the company’s ad platform.

Image: DIW-Aigen

Read next: The Truth About X’s User Stats: Are the Numbers Adding Up?
by Dr. Hura Anwar via Digital Information World

The Truth About X’s User Stats: Are the Numbers Adding Up?

X is trying hard to share its performance figures which do not look super cool, because all of the original figures can be tracked in seconds and they cannot deceive people. X has gotten a track record of sharing exaggerated data and they are still doing it in 2025. The CEO of X, Linda Yaccarino, shared a post on X and claimed that users have spent 364 billion seconds on the platform in 2024, which makes about 11,500 years combined. She didn't say whether this stats was for per day or per year, but it is presumed that it was for per day. Now if we equate average time spent by one user on X per day, it will only equate to 0.07 minutes per day which isn't that much.

X has 250 million daily active users and if we divide 364 billion seconds per active user, it is only equal to 24 minutes per day. Even though it is enough, it is still not equal to what X had claimed in March 2024, saying that users spend an average of 30 minutes on the app per day. It was also reported by X that 8 billion cumulative active user minutes per day are spent on X, which means 480 billion daily user seconds.

This suggests that either X is experiencing a decline in usage, or there may be inaccuracies in its reported stats. Before Elon Musk, it was also reported that users were spending even 38 minutes per day on the app and then it also decreased to 24 minutes per day. After Elon Musk took over X, its daily usage decreased drastically. X is known for not being transparent with their data and figures, so we cannot know what the CEO is reporting without proper context.

What we know for sure is that X isn't achieving anything because it's seeing a drop in its daily active users. This isn't a criticism to X because it's a private company and doesn't need to provide any of the official data. But it is just a little observation that X is really chaotic with its data reporting and has a tendency to give inaccurate data without any explanation.

Image: DIW-Aigen

Read next:

• Can Backlinks Boost Your Brand’s Presence on ChatGPT? New Research Questions Their Impact

• AI Chatbots Provide Non-Judgmental Mental Health Support but Struggle with Memory and Complex Issues
by Arooj Ahmed via Digital Information World

Can Backlinks Boost Your Brand’s Presence on ChatGPT? New Research Questions Their Impact

Seer Interactive, a digital marketing agency, published a new study which found that website mentions in ChatGPT answers are affected with brand mentions on Google rankings. Many brands aim to get discovered on LLMs like ChatGPT and Copilot but optimization for website mentions on LLMs is a bit different. Generative engine optimization is somewhat different from traditional search engine optimization and many brands don't seem to know about it. However, by focusing on producing content that is both high-quality and relevant, brands can significantly improve their chances of achieving visibility on search engines and generative AI platforms like ChatGPT..

To get features on ChatGPT, your website rankings on Google matter but the variety of the content and backlinks don't. The study found that LLMs mentioned websites the most which were ranked on Google’s page one. Some of the Bing rankings also seemed to matter to get mentioned on LLMs. There was some expectation that backlinks might play a big role in getting mentioned on LLMs but the study found that they don't have much correlation with that. If your website has diverse content formats, it also makes it hard for websites to get mentioned on LLMs.

An analysis of 10,000 questions about SaaS industries and finance were asked on ChatGPT 4o API for the study. After running the questions, it was found that backlinks do not have any correlation in triggering brand mentions. There are also some factors other than search rankings that affect LLM mention, like website's on-page optimizations, PR partnerships and many more. There was also a study by DemandSphere and Botify about AI Overviews, mentioning that 75% of websites featured on AI Overviews are from top 12 organic search result rankings. So this shows that if you want to have mentions of your website or blog in AI generated answers on LLMs, make sure your website is ranking well on search results.

Read next: Job Trends In The Spotlight: AI Is Transforming Industries And That’s Causing Skills to Become Obsolete
by Arooj Ahmed via Digital Information World

ChatGPT Gives Paying Users The Chance To Schedule Reminders Or Recurring Requests

OpenAI wants more users to come forward and pay for its premium subscriptions. This might be one main reason why the world-famous AI tool is offering great features to those who have signed up for a little extra.

The latest on this front is an offering where users can ask the AI assistant to schedule reminders. The same is the case for any cases that reoccur. This arrives in the form of a new beta request dubbed Tasks which will launch to the ChatGPT Plus, Pro, and Team located around the globe.

The new offering means users get the chance to organize reminders with the AI tool including those related to simple everyday tasks. They can even enable the chatbot to follow up through push alerts, no matter which app was used to perform the tasks.

The AI giant says anyone tired of repeating themselves with recurring requests on the tool can do so by scheduling repeated tasks. For instance, you can ask ChatGPT to provide you with weekend plans depending on where you’re currently located.

The latest feature seems to be the company’s plan to enable AI agents to work independently. As per the CEO, 2025 will be all about breakthroughs in AI. If that was not enough, he even claimed that this might be also when he would be adding them to the company’s own workforce to get things done.

This feature might be quite similar to what Alexa and Siri seem to be giving its users and it appears that OpenAI is heading in that same direction. These scheduled requests for more information are very independent in nature and unique in identity.

They portray some very exciting new capabilities that former digital assistants weren’t capable of. Users get the chance to access them by clicking on 4o scheduled tasks through dropdown menus on the tool. This is where they can roll out messages informing the AI assistant which actions they’d like to create.

You might even see some suggestions related to specific tasks depending on former chats. Users get the opportunity to manage commands by chatting with their assistant on any app or even by using the tasks manager that’s found on the web version of the AI tool.

Through this latest feature, the tool could browse the internet on a fixed schedule. It won’t run any extended searches inside the background. However, you can’t ask it to make purchases on your behalf.

The feature may not seem groundbreaking, considering similar capabilities offered by services like IFTTT and Make.com for some time. However, this addition is a welcome enhancement for ChatGPT users. It highlights OpenAI’s commitment to pushing the boundaries of automated systems, marking an exciting step forward in expanding the practical capabilities of AI tools.

Read next: Can Facebook’s Community Moderation Strategy Reverse Declining Downloads and Attract Users Amid TikTok’s Potential US Ban?
by Dr. Hura Anwar via Digital Information World

Can Facebook’s Community Moderation Strategy Reverse Declining Downloads and Attract Users Amid TikTok’s Potential US Ban?

Mark Zuckerberg, CEO of Meta, recently announced that Facebook is introducing community moderation of posts, just like X’s community notes. Previously, Facebook used to do in-house handling for moderating posts on the platform. Many people are assuming that the reason for this is political, and it may be somewhat correct, but Facebook is doing all of this because of decreasing demand and low user engagement. We all know that demand for Facebook has decreased and it has lost millions of users in previous years.

According to the estimates by Appfigures, Facebook saw 601 million new downloads in 2017 which was an impressive number. Then in 2018, the downloads increased and touched 630 million. After 2018, the number of downloads started decreasing a bit but 2021 was the year when Facebook was never the same. In the year 2021, Facebook’s downloads dropped to 488 million. Facebook lost what most apps cannot even dream about achieving in their lifetime. Downloads dropped even lower in 2023 at 400 million.

So, this just shows that Facebook has lost millions of new users in just a few years and now there are a lot of people who aren't going to see ads on the platform. Even though Facebook’s downloads were dropping in the previous years, 2024 saw some rise in new downloads as the platform gets 513 million installs. So Facebook adding community-powered moderation practices may seem like a political move to some, but it is also a move to protect the app from any more growth decline. Now that TikTok is on the verge of getting banned in the US, some users might come to Facebook too.

Read next: 

• After Two Years of Decline, Smartphone Market Recovers with Strong Growth, AI and Premium Shifts

• Google's Search Dominance Erodes Amidst Rising Competition and Market Shifts
by Arooj Ahmed via Digital Information World

Google's Search Dominance Erodes Amidst Rising Competition and Market Shifts

A new report by Statcounter is shedding light on Google’s share of the worldwide search engine market which fell below 90%. This was a record as it never occurred in the past decade.

The last time something as significant as this was seen was in 2015. The data shows how the search market share produced consistent values of under 90% during the last three months of 2024.

The figures outlined in the report showed how they fell to 89.34% in October and then similar findings were seen in November and December of 2024. It’s a clear trend that the Android maker is seeing market share loss consistently.

The question on many people’s minds had to do with where the drop arose. As per the stats, the market share was very consistent in most parts of the world except for Asia. This could be a possible explanation for the overall drop seen by the tech giant.

Google’s American search market share rose to 90.37% in November but fell to 87.39% in the month after. In most other months of 2024, the search market share was mostly the same, varying a little between 86 to 88%.

When you look at the bigger picture, Google was under attack for nearly two whole years over how unhelpful the search results were becoming despite it attaining dominating figures. This might have a lot to do with the firm’s unlawful monopoly status. It had a commanding consistency of 90 to 92% for nearly 10 years.

Are we finally beginning to witness people moving far away from search engines? Now that only time can tell experts are on alert to see how stats vary in the next few months. But where did searchers disappear?

Statcounter shared how Bing, Yahoo, and even Yandex got the benefit linked to Google’s loss in search market share. Runner-up Microsoft Bing stood under 4% for the last five months in 2024. You might see this as a celebration for some in the world of search marketing domain and even those linked to SEO. These are both the areas where Google lost a huge search market share. But the loss was not huge which is why Statcounter was left with no choice but to revise the information it published.

With all of that having been said, it is important to note that Statcounter has faced data reporting issues in the past, so these statistics could potentially be affected by such bugs. Further more third-party and first-party analysis will be needed to confirm the trends.

Read next: What’s OpenAI’s Economic Blueprint? A Bold Call to Reshape AI, Policy, and Global Innovation
by Dr. Hura Anwar via Digital Information World