- MegaSuperAI
- Posts
- ChatGPT's New Memory Feature
Calling all AI fans!
This week's AI newsletter is all about two big breakthroughs. First, OpenAI's ChatGPT just got a major upgrade that lets it remember past chats, making interactions way more personalised—though it's only available to Pro users for now. Second, researchers from Nvidia and some top universities have figured out how to generate AI videos that last a whole minute, which could be a game-changer for animation and more.
This week’s edition:
ChatGPT New Memory Feature
AI-Generated Video’s Leap Forward
Your 5 Min AI Learning: Turn Yourself Into a Doll With AI
New AI Tools
More AI News
AI Image of the Week
Read time: 8 Minutes
LATEST NEWS
CHATGPT MEMORY
Image Courtesy of OpenAI
What You Need to Know
OpenAI has introduced a significant update to ChatGPT, enabling the AI chatbot to remember past conversations with users. This feature allows ChatGPT to tailor responses based on previous interactions, enhancing the personalisation of its text, voice, and image generation capabilities. The update is currently available for paid tiers, specifically Pro users, with plans to expand to Plus users soon. However, users in the UK, EU, Iceland, Liechtenstein, Norway, and Switzerland must wait due to regulatory requirements.
The Finer Details
Here are some key aspects of the new memory feature:
Saved Memories: Users can explicitly request ChatGPT to remember specific information, which is then stored for future reference.
Reference Chat History: ChatGPT can draw insights from past dialogues to inform its responses, making interactions more personalised.
User Control: Users have the option to opt out of this feature or use a "Temporary Chat" mode for conversations they don't want saved.
Privacy Considerations: OpenAI provides tools for users to manage and control what ChatGPT remembers, addressing privacy concerns.
So What?
This enhancement marks a significant shift towards creating AI systems that understand users over time, making interactions more fluid and personalised. While some users may be concerned about data privacy, OpenAI provides options to manage and control what ChatGPT remembers. The feature aligns with similar updates from competitors like Google's Gemini AI, highlighting the race towards more personalised AI assistants.
AI VIDEOS
Image Courtesy of Dalal, Koceja, Hussein, Xu et al.
What You Need to Know
Researchers from Nvidia, Stanford University, UCSD, UC Berkeley, and UT Austin have developed a groundbreaking method for generating AI videos that can last up to one minute. This achievement is significant because existing models like OpenAI's Sora, Meta's MovieGen, and Google's Veo 2 are limited to much shorter durations—20 seconds, 16 seconds, and 8 seconds, respectively. The innovation involves using Test-Time Training (TTT) layers, which enhance the capabilities of pre-trained Transformer models by allowing them to maintain temporal consistency over longer sequences.
The Finer Details
TTT Layers: These layers function as small neural networks that continuously learn during video generation, improving pattern recognition and consistency across scenes.
Dataset: The method was tested using a dataset of approximately seven hours of Tom and Jerry cartoons with detailed human descriptions.
User Input: Users can provide descriptions of their video ideas, ranging from short summaries to detailed storyboards.
Limitations: Despite advancements, the model still faces challenges such as abrupt object changes or lighting shifts at segment transitions.
So What?
This breakthrough in AI video generation opens new possibilities for storytelling and content creation. By enabling longer, more coherent videos, it could revolutionise fields like animation, advertising, and educational content. However, addressing the current limitations will be crucial for widespread adoption and practical application.YOUR 5 MINUTE AI LEARNING
AI LEARNING
YOUR 5 MIN AI LEARNING
Awww
A brand new trend is taking over, and all you need is a selfie!
For this week’s AI learning we’re jumping into this trend. Social media and AI are teaming up again, letting you turn yourself into a Barbie doll using ChatGPT. Just upload a photo, give it some specific instructions, and voilà! You'll have your own custom AI doll that's totally you.
Here's how to transform your photo into an AI-generated Barbie doll image using ChatGPT:
1. Access ChatGPT: Go to the ChatGPT website or open the app. Create an account or log in.
2. Upload a Photo: Click the "Add" button to upload a photo. For the best results, use a clear, full-length image that shows your outfit well.
3. Enter the Prompt: Type the following prompt below the image: "a realistic figure (Barbie doll) of the person in this photo. The doll is a full figure displayed in original clear packaging. At the top of the box is the name of the toy '-'. The packaging next to the figure are the toy's accessories including ..."
4. Send the Message: Send the prompt to ChatGPT and wait for the image to be generated.
5. Refine the Image (If Needed): If you're not happy with the initial result, provide more specific details to ChatGPT. For example, specify the box colour or any phrases you want to include. You can also try using a different photo if the resemblance isn't satisfactory.
We tried it out for our pup Lucky and think it worked wonderfully! Try it out and check out the results!
LATEST AI TOOLS
A tool that allows you to animate static portraits with synced expressions (Free)
“Revolutionizing human image animation with advanced AI technology.”
An iOS App to visualise and suggest various design changes for your rooms from photos (Freemium)
“Simply take or upload a photo of any room in your home, choose from a variety of styles, and watch as Decory works its magic, delivering a stunning makeover that perfectly suits your taste and lifestyle. And watch as Decory breathes new life into your living space!”
A tool to identify potential inefficiencies in workflow and thus provide suggested AI automation (Paid)
“Cut the Waste and Optimise Your Business in Minutes. Leania pinpoints the processes that drain the most time and money. Our inefficiency heatmap shows you exactly where to focus — for the biggest ROI, first.”
MORE AI HIGHLIGHTS
Microsoft's new "Copilot Recall" feature, which captures screenshots of users' activity at intervals, has reignited privacy concerns. Critics warn that sensitive data, including passwords and private messages, could be stored without adequate safeguards. Privacy advocate Dr. Kris Shrishak highlighted risks of capturing information from individuals who cannot consent, and the potential misuse of stored images if devices are compromised. Despite Microsoft introducing an opt-in system and local storage assurances, concerns persist over unencrypted data and ethical implications of involuntary data collection.
OpenAI has significantly reduced the time and resources allocated for safety testing its advanced AI models, raising concerns about prioritising speed over safeguards in the competitive AI race. Evaluators now have just days to assess risks and performance of new models, compared to months for earlier versions like GPT-4. Critics warn this rushed approach increases the potential for harmful misuse, especially as models grow more capable. The company faces mounting pressure to stay ahead of rivals like Google, Meta, and xAI while balancing innovation with public safety. OpenAI defends its practices, citing automated tests and transparency in its reports, but some former staff argue the reduced rigour could lead to underestimating risks.
A report by the International Energy Agency (IEA) reveals that global electricity demand from data centres, driven primarily by AI is set to more than double by 2030, with AI-specific data centres alone projected to quadruple their energy consumption. This surge could see AI-related data processing in the U.S. consuming more electricity than manufacturing energy-intensive goods like steel and cement. While this raises concerns about energy strain and emissions, the IEA highlights AI's potential to optimise energy systems, enhance efficiency, and reduce emissions if properly harnessed.
Chelsea and Westminster Hospital NHS Foundation Trust is pioneering AI technology for skin cancer diagnosis. The AI system, DERM, analyzes dermoscopic images with high accuracy—97% for detecting skin cancer and 99.96% for ruling out melanoma. This allows rapid identification of benign cases, reducing anxiety and freeing up appointments for more critical cases. Since its launch, DERM has assessed over 10,000 patients, serving as a global blueprint for integrating AI in healthcare. Additionally, it has expanded to other hospitals, further enhancing its impact on dermatological care.
AI IMAGE OF THE WEEK
“Easter Joy” - By Team MegaSuperAI
As we approach the vibrant Easter season, excitement fills the air with the promise of renewal and celebration. This special holiday is a time to gather with loved ones, enjoy festive traditions, and embrace the spirit of spring.
Easter brings a variety of cherished customs, from colourful egg hunts to delightful family meals.
The prompt used to generate this image on Copilot:
“Children eagerly search for hidden eggs, often filled with sweet treats and surprises, while families come together to share festive dishes and celebrate the joy of the season.
The atmosphere is brightened by blooming flowers and cheerful decorations, reflecting the themes of rebirth and hope that Easter represents. Whether it's attending a church service, participating in community events, or simply enjoying time with family, this season offers a perfect opportunity to create lasting memories.”
FEEDBACK
 Teebs, your friendly AI assistant | That’s all folks. We hope you enjoyed this week’s newsletter. There’s more to come (especially as we build our AI Automation software service ReIgniteX AI) and we are excited to help you maximise AI in your life and business. |
Let us know how we did - shoot us a quick email. We read every single one (no AI in sight!).
If you enjoyed this newsletter please share with an awesome friend or two.
If you are that awesome friend then hit the below button: