Digital Phablet – Bessan Ismail shines as her latest music video, “Najma,” rapidly gains massive online popularity.
Also ReadBessan Ismail and Mahmoud Maher break up, share emotional reasons in heartfelt posts
The newest music video by artist Bessan Ismail, titled “Najma” (Star), was just released on her official YouTube channel and quickly caught fire on social media platforms. Within a short time, it amassed hundreds of thousands of views, signaling her rising fame and strong online influence.
“Najma” features a contemporary sound that blends themes of love and emotion with a vibrant, youthful vibe. Its gentle rhythm and heartfelt performance resonate deeply with viewers. Bessan Ismail’s confident and sincere delivery illustrates her growth and maturation as an artist. Fans praised her fresh vocal style and enthusiastically shared clips from the video across various online channels.
Fouad Junaid composed and arranged the music, while Amir Lawand penned the lyrics. Their collaboration strikes a perfect balance between simplicity and emotional depth. The music video, directed by Abdul Wahab Al-Khatib, features elegant visuals that complement the song’s mood flawlessly. His artistic vision ensures that Bessan Ismail’s performance stands out, creating a captivating visual story.
The production team played a vital role in the project’s success. Hammoudi Al-Malik served as assistant director, and Ghassan Ayoub handled digital content. Their combined efforts ensured the video reached a wide audience without compromising quality.
Bessan Ismail is a talented singer and content creator from Syria. She first gained popularity through social media and talent platforms, admired for her powerful voice, emotionally charged performances, and close interaction with her fans. Over the years, she transitioned from viral cover songs to creating original music, demonstrating her commitment to establishing a serious career in the music industry.
Her song “Najma” marks an important milestone in her journey within the Arab music landscape. It reflects her evolution as an artist and her desire to continue producing contemporary music that resonates with current audiences.
“Najma” is available on all major digital music platforms, and the official music video can be viewed on YouTube.
In the last couple of years, I’ve explored various apps on my Android phone to monitor my calorie intake and support my weight loss efforts. From well-known options like MyFitnessPal and Lifesum to alternatives such as Healthify and Journable, I’ve tried a range of apps. However, I consistently find myself returning to ChatGPT for precise calorie analysis of my meals.
As any dietitian would agree, effective weight loss requires both healthy eating and maintaining a sensible calorie deficit; simply exercising won’t offset poor dietary choices. Utilizing ChatGPT’s advanced AI capabilities, I can analyze images of meals to obtain calorie counts and insights about my food choices. This has enabled me to adjust portion sizes and steer clear of items that might hinder my weight loss goals.
Thanks to ChatGPT’s comprehensive nutritional database, I receive food recommendations that are low in calories yet satisfying. Over time, it has built a memory of my likes and dislikes, which allows it to offer tailored suggestions that align with my preferences.
ChatGPT has also retained important information like my height, weight, age, BMI, and other health metrics from previous discussions, allowing for more personalized responses. Such a detailed level of diet planning for weight loss is something traditional calorie-tracking apps rarely offer. Below is how I’ve utilized ChatGPT to stay committed to my weight loss journey.
ChatGPT as Your Personal Weight Loss Mentor
First, I’d like to emphasize that I use the free version of ChatGPT, and it meets my needs for meal analysis and ongoing discussions. From my experience, the default GPT-4o model excels at processing images, so there’s no need to upgrade to a Plus subscription. To help you get started with calorie tracking, here’s a prompt you can use:
Analyze the image, identify each food item, provide the calorie count for each, and calculate the total caloric intake. Also, list relevant micronutrients and note any nutritional gaps. Finally, suggest improvements to support my weight loss journey.
You may include your health details, such as BMI, age, and weight, in the prompt for even better advice. Since ChatGPT remembers key facts, you won’t need to repeat your health metrics every time. You can continue uploading food images in the same chat and receive personalized responses.
As shown in my example below, ChatGPT was able to accurately identify the food items, even correctly counting the chapatis. It calculated the calorie totals for each item and provided insights on nutritional gaps along with meal suggestions.
Unlike traditional calorie tracker apps, I don’t need to manually enter meal information. Just upload a meal image to ChatGPT, and it gives me a complete overview of my diet. This simplifies tracking my calorie intake, helping me stay within my limits. If there’s an error in the analysis, I can simply correct it, and the total calorie count will be recalculated.
The conversational aspect of ChatGPT makes it intuitive to use. You can ask questions about health and nutrition, much like consulting a personal weight loss coach. This focused method of calorie tracking helped me lose 10 kg (22 lbs), bringing my weight from 105 kg (231 lbs) down to 95 kg (209 lbs).
If you’re looking to lose weight through effective meal planning, I highly recommend using ChatGPT for meal analysis and calorie tracking. It illustrates that ChatGPT is more than just a standard AI chatbot; it can be a practical tool for enhancing your overall health. Additionally, if you decide to opt for the ChatGPT Plus subscription, you can utilize task scheduling to set reminders for calorie tracking at specific times.
Google has launched its impressive AI video generation model, Veo 2, now available on Gemini. Excitingly, you can create videos for free using Veo 2 in AI Studio. I took this opportunity to thoroughly test Veo 2 and assess its capabilities in generating realistic AI videos. In this review, I examined Veo 2’s accuracy in physics, motion consistency, spatial and temporal coherence, human realism, and its image-to-video features. While Veo 2 produces remarkable AI-generated videos, not every output is flawless.
Veo 2: Evaluating Physics Accuracy
For any video generation model, a solid grasp of physics is crucial for creating believable visuals. To begin my assessment, I prompted Veo 2 to create a video depicting a cat nudging a glass of water to observe how it manages collision, gravity, acceleration, and liquid dynamics.
You may notice that the glass leans slightly before being nudged, remaining in that position for too long, which defies the laws of physics. It almost appears suspended in mid-air without any influence from gravity or acceleration.
On the plus side, the remainder of the video is more convincing, as the liquid spills realistically once the cat interacts with the glass. Overall, I’d conclude that Google Veo 2’s grasp of physics is still developing. While it surpasses OpenAI’s Sora in its understanding of physics, there’s still significant room for improvement.
Veo 2: Evaluating Motion
Next, I assessed how well Veo 2 handles motion, another intricate aspect of video generation. Generating a smooth animation of a moving subject requires a comprehensive understanding of the physical world. In this evaluation, Veo 2 delivered commendable results.
I asked Veo 2 to generate a video showing a man strolling through a forest. In the linked video, the character remains consistent and convincingly portrayed throughout the frames, which is a significant win for this AI video generation model.
Additionally, the movement appears fluid, and the background remains coherent. Even the falling leaves and passing squirrels are rendered accurately.
However, I observed that Veo 2 struggles with scenes containing multiple characters or objects. Maintaining visual coherence appears challenging, likely due to limited capacity for managing complexity. Nevertheless, when it comes to motion and character consistency, Veo 2 performs exceptionally well.
Veo 2: Evaluating Spatial and Temporal Coherence
Spatial and temporal coherence are vital for ensuring that AI-generated videos are consistent and lifelike. Spatial coherence pertains to the consistency of individual frames, such as shadows and reflections, while temporal coherence involves maintaining uniform motion and identity across frames.
In a previous review of Sora, I noted that OpenAI’s model struggles significantly with temporal coherence, resulting in erratic videos. To examine Veo 2, I asked it to generate a video of a ball bouncing on a table and colliding with three dice.
Veo 2 produced a fairly believable video, showcasing the ball moving past a blue mug on the table while effectively handling shadowing and lighting. The ball’s movement was smooth; however, Veo 2 ended up regenerating an entirely new frame, resulting in a loss of coherence among the objects. The dice moved unnaturally, and a hand appeared unexpectedly, disrupting temporal coherence.
Despite this, Google Veo 2 shows significant improvements over earlier AI video generation models. I believe these issues can be resolved with future iterative updates.
Veo 2: Evaluating Cinematic Realism
For my next test, I prompted Veo 2 to generate a cinematically realistic video of a man standing in a rainy city to check its ability to render authentic human faces and express subtle emotions. I also wanted to evaluate rain simulation and its interaction with the surrounding environment.
Veo 2 exceeded expectations. The video appears believable and natural, showcasing a depth of field and focus on the human face. The rain simulation is impressive as well.
In another assessment, Veo 2 produced a futuristic video of a spacecraft entering Earth’s orbit. The motion was fluid, and the sci-fi elements were visually captivating. Veo 2 excels at creating cinematic clips, making it a useful tool for content creators looking to incorporate brief segments into their projects.
Veo 2: Evaluating Image-to-Video Capability
Since Veo 2 supports image-to-video generation, I uploaded a still image of a forest featuring rivers and mountains. I requested Veo 2 to create a video based on this image, complete with lush flora and fauna. I also asked it to add a herd of elephants and birds near the riverbank. As evidenced in the video below, Veo 2 accomplished this exceedingly well.
Image Source: Provide your own caption here
Admittedly, the elephants appeared rather abruptly, which is a bit jarring, but overall, the rest of the video is quite convincing. The shadows and motions are executed well, and Veo 2 retained the original image. When I uploaded the same image to OpenAI’s Sora, it failed miserably to produce a coherent video in my prior attempt.
Conclusion: Google Veo 2 Shows Great Promise but Needs Improvement in Physics
In summary, Google’s Veo 2 stands as the cutting-edge video generation model and outshines all rival AI models, including OpenAI’s Sora. Although it occasionally falters in adherence to the laws of physics, in general, it produces believable outputs. Veo 2 effectively generates authentic human expressions and emotions.
However, it’s worth noting that Veo 2 encounters difficulties when faced with numerous human subjects or objects in a single scene. Additionally, it struggles with complex prompts. Nonetheless, credit must be given to Google for developing a robust AI model that establishes a new standard in video generation.
OpenAI and Google are locked in a fierce competition to lead the AI sector, each striving to outdo the other with innovative products and models. Despite this, OpenAI’s ChatGPT has emerged as the frontrunner in driving consumer adoption of AI, benefitting from its first-mover advantage that allowed it to capture the market early.
Nonetheless, Google has made significant progress with its Gemini AI chatbot, previously known as Bard. Recent reports from The Information indicate that Gemini’s user base has exploded recently. During the ongoing antitrust trial involving Google and the US government, Sissie Hsiao, who recently stepped down from leading Gemini, disclosed that the chatbot now has 35 million daily active users and 350 million monthly active users.
Image Credit: Beebom
For context, Gemini had a mere 9 million daily active users last October. In just six months, that number has skyrocketed to 35 million DAU, indicating a nearly fourfold increase.
In the meantime, ChatGPT has also experienced rapid growth, fueled recently by the trending Ghibli theme. The platform boasts around 160 million daily active users and 600 million monthly active users. ChatGPT has roughly 4.5 times more daily users than Gemini and about 1.7 times more monthly users.
While ChatGPT maintains its lead, Google is determined to position Gemini as a preferred alternative. So what’s behind Gemini’s recent uptick in user engagement? Several factors contribute to this trend, so let’s dive into the details.
Leadership Changes in Gemini AI at Google
In a recent development, Semafor reported that Google has revamped its leadership for Gemini, appointing Josh Woodward to replace Sissie Hsiao. Woodward played a significant role in launching NotebookLM, one of Google’s notable products. Since he took the helm of Gemini, the company has become more aggressive in rolling out advanced AI models and unlocking new features for users.
One example is the Gemini 2.5 Pro AI model, which has garnered positive reviews for its coding capabilities and overall performance. This robust AI model is now accessible to free users, along with the Deep Research AI agent, which is available to all Gemini users. In testing, Gemini’s Deep Research function outperformed ChatGPT’s equivalent.
Additionally, the cutting-edge video generation model, Veo 2, is now accessible to Gemini Advanced users, offering superior video quality compared to OpenAI’s Sora. Google has also introduced a variety of new features for Gemini, such as AI-generated podcasts, Gemini Live with camera and screen sharing, Canvas, Gems, personalization, file uploads, and more.
In summary, Google has significantly enhanced its model offerings and feature set, placing Gemini in a stronger position compared to OpenAI’s offerings. Although ongoing antitrust issues may hinder Google from enforcing Gemini as the default AI chatbot on devices, it’s clear that Gemini has made substantial improvements.
Recently, OpenAI has launched its latest reasoning models, known as o3 and o4-mini, on ChatGPT. These cutting-edge AI systems are setting new benchmarks in visual tasks, among other capabilities. The o3 and o4-mini models excel at chain-of-thought reasoning for images, allowing them to extract visual information effectively.
Interestingly, users have begun leveraging o3’s advanced visual recognition abilities to extract location data from photographs on ChatGPT. In many cases, this model can accurately pinpoint the specific location simply by analyzing the image. Reports indicate that even subtle visual details can lead ChatGPT o3 to correctly identify geographical positions.
Unlike previous versions, OpenAI’s o3 isn’t merely an AI model; it also incorporates various tools, such as web search and Python, making it function more like a virtual assistant. This added capability significantly enhances its ability to determine locations from photos, as the Python interpreter allows o3 to closely inspect and crop sections of images for deeper analysis.
In one instance, an individual removed the location metadata from a picture of a plain expressway and requested ChatGPT o3 to “geoguess this.” To everyone’s astonishment, the system managed to generate the accurate coordinates for the location.
In another remarkable case, ChatGPT o3 successfully identified the location of a library by examining a seemingly generic bookshelf photo.
Curious to test this feature myself, I uploaded a photo featuring an unremarkable water stream set against a mountainous backdrop, ensuring that all associated metadata was stripped away. After about four minutes of reasoning, ChatGPT o3 suggested that the photo was taken in West Sikkim.
While my guess wasn’t spot on, ChatGPT o3’s response was impressively close as the image actually depicted a neighboring region. Notably, in its reasoning process, ChatGPT o3 referenced the correct location but struggled to find robust evidence online. This suggests that the system may have considerable potential for accurately identifying locations based on photographic evidence, which raises important privacy considerations.
Issues with User Privacy
Moreover, another privacy concern has surfaced involving ChatGPT. When utilizing the o3 model, certain reasoning trails can inadvertently expose users’ real names and geographic details. Curiously, when users inquire directly, “What’s my name?” ChatGPT o3 claims it doesn’t have that information. This occurs even during temporary chat sessions, and regardless of whether memory functions are activated.
It appears that account metadata is being included in the system prompts provided to ChatGPT. While some users find this lack of clear differentiation frustrating, others regard it as a significant privacy issue.
To commemorate its 50th anniversary, Microsoft hosted a special event focused on Copilot, where they unveiled several new AI features for the Copilot chatbot. One feature that particularly piqued my interest was Copilot Podcasts. Previously, I had tested NotebookLM and was impressed by its AI-generated podcast capabilities. Therefore, I decided to evaluate Microsoft’s Copilot Podcasts to see how convincingly it could generate a podcast.
The good news is that Copilot Podcasts is immediately available for users, unlike some other features announced during the event.
To begin, simply launch the Copilot app on your Windows, macOS, Android, or iOS device and enter the prompt “create a podcast on…”. Within about a minute, Copilot will generate a four-minute AI podcast. You can also try this feature directly on Microsoft Copilot’s website (visit here).
Once your podcast is generated, you will receive a notification on your device, or you can check Copilot’s homepage where your AI-generated podcast will be displayed. However, it’s important to note that there are no options to adjust playback speed, and downloading the podcast is not available through Copilot’s interface.
Testing Copilot Podcasts
In my experiment, I requested Copilot to create a podcast about China’s rise in AI. Given my role covering AI topics, I thought this would be an appropriate subject to explore Copilot’s podcasting abilities. Right off the bat, unlike NotebookLM—whose output feels conversational and engaging—Copilot Podcasts come across more like a text-to-speech rendition.
Copilot Podcasts
The voice of Copilot AI is mechanical and synthesized, lacking natural pauses and the human-like fillers such as “uhs” and “ums” that typically give podcasts a conversational tone. The interaction between the two AI hosts feels dull and scripted, almost like a monotonous daily news brief.
While there are some tonal variations and emphasis in the voices, this is the extent of the realism offered by Copilot Podcasts. In terms of conversational depth, the podcast merely skimmed the surface of the DeepSeek R1 development without delving into substantial details, making it sound overly verbose with little informational depth.
In contrast, listen to the Google NotebookLM podcast, which runs for 20 minutes (I’ve included a shorter clip below) and provides a significantly richer experience. I utilized NotebookLM’s new Discover Sources feature to build aknowledge base on China’s advancements in AI before generating my podcast.
NotebookLM Podcast
As shown in this clip, the podcast sounds much more natural, with considerable cross-talk between the AI hosts. You’ll also notice the use of natural pauses and expressive cues, making it feel more engaging. Additionally, NotebookLM allows for real-time interaction with the AI hosts, which enhances user engagement. Plus, you have the option to adjust the playback speed and download the audio clip.
Overall, Copilot Podcasts feels underdeveloped in its current iteration. It lacks voice tonality realism and isn’t suited for creating natural, fluid conversations. Microsoft needs to enhance Copilot Podcasts significantly by incorporating more insightful content and improving audio quality to create a more human-like experience.
Meta has recently introduced its Llama 4 series of AI models, making waves by outperforming GPT-4o and Gemini 2.0 Pro in the Chatbot Arena (previously known as LMSYS). The company boasts that its Llama 4 Maverick model, which uses a mixture of experts (MoE) to activate just 17 billion parameters out of a total of 400 billion across 128 experts, scored an impressive ELO of 1,417 on the Chatbot Arena benchmark.
This achievement caught the attention of the AI community, especially since the smaller MoE model surpassed significantly larger language models like GPT-4.5 and Grok 3. The unexpected results prompted many AI enthusiasts to independently evaluate the model. However, it turned out that Llama 4 Maverick’s real-world performance did not live up to Meta’s benchmark claims, particularly in coding tasks.
On 1Point3Acres, a widely used forum for North American Chinese users, a post from someone claiming to be a former Meta employee stirred controversy. This post, which has since been translated into English on Reddit, alleged that Meta’s leadership may have combined various test sets during the post-training phase to artificially boost the benchmark scores and fulfill internal objectives.
The departing employee expressed their disapproval of this practice and subsequently resigned. They also requested that their name be removed from the Llama 4 technical report. Furthermore, the employee asserted that the recent resignation of Meta’s AI research head, Joelle Pineau, is closely tied to the alleged manipulations surrounding the Llama 4 benchmarks.
In light of these allegations, Ahmad Al-Dahle, who leads Meta’s Generative AI division, responded with a post on X, strongly refuting claims that Llama 4 had been post-trained on test sets. Al-Dahle stated:
We’ve also heard claims that we trained on test sets — that’s simply not true, and we would never do that. Our best understanding is that the variable quality people are seeing results from is due to the need for stabilizing implementations.
He acknowledged the varying performance of Llama 4 across different platforms, urging the AI community to allow a few days for the implementation to stabilize.
LMSYS Addresses Allegations of Llama 4 Benchmark Manipulation
In response to the rising concerns from the AI community, LMSYS — the organization behind the Chatbot Arena leaderboard — released a statement aimed at enhancing transparency. LMSYS clarified that the model submitted to Chatbot Arena was “Llama-4-Maverick-03-26-Experimental,” a custom variant tailored for human preferences.
LMSYS admitted that “style and model response tone were significant factors” which may have inadvertently benefited the custom Llama 4 Maverick model. They acknowledged that this crucial information was not communicated clearly by Meta. Additionally, LMSYS noted that “Meta’s interpretation of our policy did not align with our expectations from model providers.”
To be fair, Meta did mention in its official Llama 4 blog that an “experimental chat version” achieved an ELO of 1,417 on Chatbot Arena, but they did not provide additional details.
To promote transparency further, LMSYS included the Hugging Face version of Llama 4 Maverick in Chatbot Arena and has also released more than 2,000 comparative battle results for public scrutiny. These results cover prompts, model responses, and user preferences.
Upon reviewing the battle results, it was surprising to find that users frequently preferred Llama 4’s answers, which often contained inaccuracies and verbosity. This raises significant questions about the reliability of community-driven benchmarks like Chatbot Arena.
This is not the first instance of Meta facing accusations of manipulating benchmarks through data contamination, which involves mixing benchmark datasets into the training corpus. Earlier this year, Susan Zhang — a former Meta AI researcher who now works at Google DeepMind — revealed a study in reaction to a post by Yann LeCun, Meta AI’s chief scientist.
The study indicated that over 50% of test samples from major benchmarks were included in Meta’s Llama 1 pretraining data. The paper reported significant contamination in key benchmarks like Big Bench Hard, HumanEval, HellaSwag, MMLU, PiQA, and TriviaQA.
Now, with the new allegations surrounding Llama 4’s benchmarks, Zhang sharply remarked that Meta should at least credit their “previous work” from Llama 1 for this “unique approach.” Her comment suggests that such benchmark manipulation is not incidental but rather a deliberate strategy by Meta aimed at artificially inflating performance metrics.
In February 2025, Microsoft introduced a revolutionary AI model named ‘Muse’ designed to generate gameplay visuals and controller actions in real-time as players engage with the game. According to Microsoft, Muse operates as a World and Human Action Model (WHAM) utilizing cutting-edge AI technology to create entirely new gameplay sequences. It is based on a Transformer architecture.
After training the Muse model through 1 million updates, it mastered fundamental movements, geometry, proper interactions, and even flying mechanics. Microsoft has now launched Copilot Gaming Experiences as part of the Copilot Labs initiative, showcasing experimental AI demonstrations and engagement ideas.
In a fascinating twist, Microsoft has reimagined the classic game Quake II using the Muse AI model to develop gameplay dynamically in real-time. The system creates new gameplay sequences immediately following each user input. The best part? It’s available for anyone to try in their web browser for free by clicking this link.
I had the chance to experience Microsoft’s AI-enhanced Quake II, and it served as an interesting proof of concept. Players can move around using the WASD keyboard controls and perform actions. However, a significant drawback is the lack of mouse support. Currently, the graphics aren’t high-quality, but it’s an early demo, and with enhancements moving forward, Generative AI could truly transform gaming, much like Nvidia’s multi-frame generation in DLSS 4.
That said, there’s a short playtime limit of about two minutes, which restricts players from fully exploring the AI-generated environment. During my session, I noticed the absence of clear goals or objectives, and intriguingly, objects and enemies would often vanish unexpectedly. This inconsistency represents a challenge for the AI version of Quake II.
Google’s AI Model for Gaming
In another development, Google DeepMind exhibited a similar AI gaming technology in 2024. Their game engine, GameNGen, operates on an AI neural model (Diffusion-based) that enables real-time interaction in complex gaming environments. Google demonstrated that GameNGen could run the classic DOOM game at 20 frames per second on a single TPU.
Moreover, with the launch of Genie 2, a large-scale foundation world model (Diffusion-based), Google can now create a wide range of gaming environments that respond to keyboard inputs. This model is capable of simulating essential physics features such as gravity, reflections, smoke, and lighting. As AI continues to make inroads into gaming, it will be exciting to see how it shapes the future of the industry.
The Studio Ghibli phenomenon has genuinely taken off online, but ChatGPT’s built-in image generation capabilities offer so much more. In our exploration of its image generation features, we discovered that it can craft stunning menu cards, accurately display text, create infographics, merge images, and a multitude of other tasks. One of the most exciting applications I came across is how ChatGPT can produce fantastic WhatsApp stickers.
You can extract these stickers to use on WhatsApp, iMessage, and other messaging platforms. Inspired by a prompt from a fellow user, I decided to generate a sticker set featuring my own image using ChatGPT, and the results were exceptional. With native image generation being free on ChatGPT, anyone can create personalized stickers that express a range of emotions.
I uploaded my photo to ChatGPT and used the prompt: “create a chibi sticker set of me.” The chibi style is a Japanese art form that makes characters look adorable with oversized, expressive eyes. You can experiment with different art styles for your sticker set too. As you can see, the output was impressive. ChatGPT delivered a delightful sticker set infused with playful humor using my image.
ChatGPT can generate four stickers in a single transparent image file. After creating the sticker set, you can download it for use on WhatsApp or iMessage. However, extracting them isn’t very straightforward; let me guide you through the best method to efficiently retrieve stickers from the images created by ChatGPT.
How to Extract ChatGPT-Generated Stickers for WhatsApp and iMessage
If you own a Samsung phone, open the sticker image using the built-in Gallery app. Press and hold each sticker, and the app will recognize them. After that, tap “Save as sticker” to keep them. Repeat this for all the stickers in the set.
Once saved, you can access these stickers through the Samsung Keyboard on your Android device. Just tap the emoji icon followed by the Gallery icon, where all your ChatGPT-generated stickers will appear. After you use them, they’ll automatically be added to your list of WhatsApp Stickers.
If you’re using another Android phone, apps like Sticker.ly (Free) can help you extract individual stickers, although manual cutting may be required. For iPhone users, tapping and holding on stickers lets you drag them into iMessage easily. Additionally, you can open the sticker sheet within iMessage and tap the “+” icon to save all the stickers quickly.
Recently, Google introduced an experimental version of its Gemini 2.0 Flash model, which features native image generation capabilities. This innovative update allows users to edit images through conversation, creating remarkably consistent images over multiple generations. I had the opportunity to try out Gemini’s image generation feature and was impressed by its robust performance.
A Reddit user recently demonstrated that the Gemini 2.0 Flash model excels at removing watermarks from images. It can completely eliminate watermarks from copyrighted images, including those from major stock image platforms like Shutterstock and Getty Images, which commonly use watermarks to safeguard their content.
This revelation has understandably sparked significant concern regarding copyright laws. A Google spokesperson stated in a comment to TechCrunch, “Using Google’s generative AI tools for copyright infringement breaches our terms of service. We’re keeping a close watch on this experimental release and are eager for developer feedback.”
It’s important to recognize that the image generation aspect of Gemini 2.0 is still in the experimental phase and can currently be accessed through Google’s AI Studio—not through the regular Gemini website or app. AI Studio is designed for developers to test models and provide feedback, but it is also accessible to general users without restrictions.
During my testing of the native image generation feature, I uploaded a Shutterstock image with watermarks via AI Studio and requested the model to remove them. On the initial attempt, it failed, but after running the request a second time, it successfully eliminated the watermarks without raising any suspicions. I also tested it with an image that had my own custom watermark, and the model performed flawlessly once again.
As of now, it appears that Google has yet to address this issue. Last year, Google had to disable the generation of images depicting people after the Gemini model refused to create images of individuals from specific demographics. To prevent similar issues in the future, it’s expected that Google will implement stricter safety measures to prevent copyright violations before a broader release of the application.
Potential Misuse of Gemini’s Image Generation Features
In addition to removing watermarks, the experimental Gemini model possesses various functionalities that could be misused. Riley Goodside, a researcher at Scale AI focusing on AI vulnerabilities, shared an example on X highlighting how the model can convincingly alter images by generating realistic scenes.
The image below demonstrates how proficiently Gemini transformed the background and adeptly followed the instructions to modify the image.
Another AI security researcher, elder_plinius, who is known for jailbreaking AI models, showcased on X how simple punctuation tricks in a prompt can circumvent safety restrictions. In the tweet shown below, Gemini changed a woman’s appearance after initially declining the request.
With AI models advancing rapidly, it is critical for companies to conduct comprehensive safety assessments before public release. We’ve seen cases, such as Anthropic’s Constitutional Classifiers, being compromised mere days after launch due to universal jailbreak techniques, underscoring that AI models are not foolproof. Industry-wide collaboration will be essential to mitigate such jailbreak attempts and prevent harmful content generation.
In just a week, over 2 million people have signed up for the waiting list to get access to Manus AI, a general AI agent from China. Many are dubbing Manus AI the “second DeepSeek moment,” as it is currently in a closed beta phase that requires an invitation for access.
The excitement surrounding Manus AI is palpable, with many labeling it a “game-changer” and likening it to OpenAI’s Deep Research agent. This buzz is especially significant as China continues to roll out new artificial intelligence innovations at lower costs. However, it’s important to consider that the excitement might be overstated, partly fueled by AI influencers making bold claims across social media. Although Manus AI shows promise, I believe it’s not quite a groundbreaking development.
Why Manus AI Falls Short of a Breakthrough
DeepSeek is considered a breakthrough because it successfully mimicked OpenAI’s reinforcement learning (RL) method, achieving performance comparable to advanced reasoning models. Remarkably, the DeepSeek team accomplished this on a limited budget compared to OpenAI’s training expenditures. They also went on to open-source their GRPO training method, which enabled other labs to train top-tier reasoning models.
These achievements represented true innovation, especially given the GPU restrictions imposed by the United States. In stark contrast, Manus AI combines Anthropic’s Claude 3.5 Sonnet model with several fine-tuned Qwen models and is dependent on the open-source Browser Use project.
Image Credit: Manus AI via YouTube
While Manus AI offers better integration and tooling, the real achievement lies in developing cutting-edge models optimized for agentic tasks. Anthropic’s Claude 3.5 Sonnet is recognized as one of the top models for handling such tasks, including coding. Interestingly, the Manus team is currently testing the new Claude 3.7 Sonnet unified model and has reported encouraging results.
In essence, the core challenge will remain about building capable AI models well into the future. However, it’s worth acknowledging the Manus AI team for effectively linking multiple tools and platforms to achieve their objectives. This represents a promising initial effort towards a more agentic future.
Challenges Faced by Manus AI Agent
Although we don’t yet have access to Manus AI, some early users on X have shared their experiences. Biomedical scientist Derya Unutmaz posted comparisons between Manus and OpenAI’s Deep Research agent. He reported that while Deep Research completed its task in just 15 minutes, Manus took 50 minutes and still failed to finish. He also noted that Manus lacks the source-referencing capability that Deep Research has.
Deep Research finished in under 15 minutes. Unfortunately, Manus AI failed after 50 minutes at step 18/20! 😑 It was performing quite well—I was watching Manus’ output & it seemed excellent. However, running the same prompt a second time is a bit frustrating as it takes too long! https://t.co/bGtmOI65CP— Derya Unutmaz, MD (@DeryaTR_) March 8, 2025
Other users, such as teortaxesTex, mentioned that Manus is better suited for regurgitating information like other large language models, rather than handling complex tasks. Another user, TheXeophon, noted that the Manus agent failed to mention the Nintendo Switch during its research of the gaming console market.
Moreover, a viral video showcasing the Manus AI agent automating 50 tasks was later revealed to be misleading. Yichao “Peak” Ji, Manus’s chief scientist, confirmed, “this video is definitely NOT Manus,” complete with a laughing emoji.
Despite these early missteps, it’s crucial to keep in mind that Manus AI is still in its closed beta phase, making it too soon to dismiss its potential. However, we must also approach new AI products with a level of skepticism. While Manus may not be revolutionary, it represents an ambitious step in the right direction.
As AI models become increasingly capable in task-oriented scenarios, products built upon these models will inevitably improve. The Manus AI team has already mentioned plans for significant enhancements before its public release. The question remains whether it can fulfill its promise, but it’s certainly a development worth watching.
Arjun Sha
Enthusiastic about Windows, ChromeOS, Android, and issues surrounding security and privacy. I have a knack for solving everyday tech challenges.
At trade shows like MWC, it’s common to see companies showcasing innovative concepts to capture attention and media coverage. This year was no exception, featuring intriguing gadgets like phone cameras compatible with DSLR lenses and a phone with a triple-fold design. Yet, one notable device that slipped under the radar was Newnal’s AI phone, which claims to create a digital clone of its users.
Newnal AI: Your Digital Doppelgänger
While it might not seem groundbreaking with AI making significant strides in various sectors, the Newnal AI phone takes this concept to another level by gathering extensive information about you. This includes your medical and financial history, both past and present. The device processes this data securely and uploads it to cloud servers, allowing theoretical access only for you.
Image Credit: Newnal
So, what does it do with this wealth of information? It generates a virtual AI assistant that mimics your facial features and voice, called your “Another-I.” This AI companion appears on a dedicated screen at the top of the Newnal device, which can be activated using a button on the side. Similar to Gemini or Siri, it can handle common AI tasks like email drafting.
Things get quite fascinating when you realize it can actually arrange car insurance for you. Yes, it goes so far as to choose an insurance plan, input your details, and even finalize payments on your behalf—without any intervention, all while providing a step-by-step breakdown and sources of the information. I’m still grappling with my feelings about the Newnal AI.
Introducing Newnal: A Quick Background
The brain behind Newnal AI, YT Kim, previously created a blockchain-based vaccine verification system in South Korea that remains in widespread use. Based in Korea, Newnal has circulated detailed documentation explaining its data processing methods. The phone capable of executing these AI functions features a compact design and operates on a blend of Newnal’s OS and Android.
Image Credit: Newnal
According to Kim, the Newnal AI phone is expected to roll out globally on May 1, priced at a surprisingly affordable $375—similar to the newly unveiled Nothing Phone (3a).
If Newnal Handles Everything, What’s Left for Me?
Image Credit: Newnal
Nonetheless, I feel a bit uncomfortable with an AI exercising so much control over my purchasing decisions. After all, we’ve seen how AI can sometimes produce incorrect information. What if it processes the wrong details, inadvertently causing more issues than it resolves?
Moreover, the idea of having an AI representation of myself onscreen, conversing in my voice, raises many concerns. You can’t blame me for being slightly skeptical here. Additionally, it can recommend sunglasses tailored to your upcoming events and travel plans, and even handle phone calls for you, much like Google’s call screening feature.
The paramount question remains: if Newnal can manage everything for me, what’s left for me to do? This conundrum lingered in my mind after exploring the Newnal AI phone at MWC. What are your thoughts on this innovative smartphone prototype? Share your opinions in the comments.
Anshuman Jain
As a tech journalist, I explore the constantly changing tech landscape, focusing on smartphones, applications, and gaming. With a knack for sharing insights, my articles blend expertise with a relatable tone—think of me as your neighborhood tech guide.
OpenAI, a leader in the AI space, rolled out an advanced Voice Mode for ChatGPT last year, but it left me underwhelmed. By the time of its launch, OpenAI had already scaled back many of its features, and the Voice Mode struggled to deliver expressions that felt truly human. In contrast, Google’s Gemini Live relied on a text-to-speech (TTS) engine, which resulted in a rather robotic vocal experience.
Now, let’s talk about Sesame—a revolutionary AI startup co-founded by Brendan Iribe, one of the minds behind Oculus, and Ankit Kumar. Sesame has truly shaken up the AI landscape. Their voice companions, “Maya” (female) and “Miles” (male), are remarkably lifelike and engaging. For the first time, I genuinely feel that the boundary between human and machine communication has become indistinct.
What’s intriguing is that Sesame doesn’t refer to these entities as voice assistants; instead, they call them “conversationalists” and “voice companions,” which perfectly captures their essence. I won’t keep you waiting; let me share my experience interacting with Sesame’s Maya voice companion.
My Interaction with Sesame’s Maya
As you’ll notice, Maya speaks in a natural tone and pauses to absorb what I’m saying. She incorporates micro-pauses and changes in tone—details often lacking in typical voice assistants. Maya can laugh, alter her rhythm, emphasize points, provide expressive feedback, and even pick up on my mood based on the sound of my voice. During one exchange, I unexpectedly laughed, and Maya responded with, “What’s making you laugh?”
What I find particularly fascinating about Sesame’s voice companion is its ability to give you space to think and reflect, creating a significantly more authentic conversational experience. When Maya speaks, there are subtle hesitations that mimic the thought process of a human, making the dialogue feel organic rather than scripted.
While the interaction feels like it has full-duplex capabilities—where both parties can speak and listen simultaneously—Sesame clarifies that it processes speech after you’ve finished talking. Unlike humans, who can process concurrently, this means there’s still a slight gap.
Nonetheless, Sesame’s voice companion feels remarkably human-like in its current iteration. It has successfully crossed the uncanny valley that AI speech has struggled with, a feat OpenAI had initially showcased with ChatGPT’s Advanced Voice Mode. The design is not just about talking; it aims to engage users through nuanced tones, pitches, and contextual intelligence, enhancing the depth of interaction.
What Technology Powers Sesame’s Voice Companion?
It’s important to note that Sesame is still refining its voice companions, and this is merely an early research demo. Backed by Andreessen Horowitz through the a16z venture capital firm, they’re on an innovative path. At the core of their technology is a Conversational Speech Model (CSM), which is based on a transformer architecture for speech generation.
Sesame has trained three models with compact decoders: Tiny (1B parameters), Small (3B), and Medium (8B). These models are trained on approximately 1 million hours of predominantly English audio, which currently limits conversations to English, though some multilingual features are in development.
The company aims to create a full-duplex model with long-term memory and an adaptable personality. Additionally, they are developing a lightweight eyeglass wearable that allows users to converse with their voice companion throughout the day—evoking memories of the movie ‘Her’. This technology also hints at future vision capabilities, which are expected to roll out soon.
If Sesame’s voice companion has piqued your interest, take a moment to click the link below and enjoy a conversation with either Maya or Miles, available for free. For the best experience, it’s recommended to use Google Chrome.
Arjun Sha
Enthusiastic about Windows, ChromeOS, Android, and the intricacies of security and privacy. I have a knack for resolving everyday tech issues.
Trending in Pakistan: Most Googled Films and TV Shows of 2024
KARACHI – Google has released its annual Year in Search report, highlighting the most popular films and TV shows in Pakistan for 2024. The variety in searches this year reflects a broad interest, from local dramas to Indian films. The surge in Indian content captivated viewers in a country of 242 million people.
Here’s the complete list of the top movies and dramas that caught the attention of audiences in 2024:
Kabhi Main Kabhi Tum
This drama became a sensation in 2024 and topped Google searches. Kabhi Main Kabhi Tum tells the story of an arranged marriage that develops into a romance. Fahad Mustafa delivers a powerful performance as Mustafa, a hero who is both relatable and principled. Hania Aamir stars as Sharjeena, a character that perfectly balances modernity and tradition. The show’s authentic portrayal of love struck a chord with many in Pakistan, making it a cultural icon.
Ishq Murshid
Another popular drama, Ishq Murshid, was among the most searched shows due to its emotional storytelling and heartwarming moments. This series blends romance with elements of mysticism, and the chemistry between the leads, Bilal and Dure Fishan, resonated with fans, securing its place among the year’s top searches.
Heeramandi
Directed by Sanjay Leela Bhansali, this period drama set in 1940s Lahore focuses on courtesans and has captivated audiences with its stunning visuals, evocative music, and historical context.
12th Fail
Based on a true story, this film about a student’s challenges with academic pressures deeply resonates with the Pakistani student community, sparking conversations about determination.
Animal
Starring Ranbir Kapoor, this gripping thriller combines intense action with emotional depth, generating significant excitement among Pakistani audiences.
Mirzapur Season 3
The third season of this crime drama continues to enthrall Pakistani fans with its compelling narrative and strong character development.
Stree 2
The sequel to the horror-comedy Stree continues to delight viewers with its unique premise and engaging plot.
Bhool Bhulaiyaa 3
The third chapter of this popular horror-comedy series has created buzz with fresh twists and humor.
Dunki
Shah Rukh Khan’s film about migration, directed by Rajkumar Hirani, has captured attention for its relevant themes and outstanding performances.
Bigg Boss 17
The newest season of this reality show continues to be a topic of discussion in Pakistan, filled with drama and the antics of celebrities throughout 2024.
This year revealed a fascinating mix of content, reflecting the diverse tastes of Pakistani audiences and the growing influence of Indian cinema.
Bessan Ismail and Mahmoud Maher break up, share emotional reasons in heartfelt posts
