- MegaSuperAI
- Posts
- OpenAI Introduces o3 and 04-mini
Hello AI Enthusiasts!
This week's AI newsletter covers the introduction of OpenAI’s o3 and o4-mini models, which they claim are their smartest and most capable models to date. Google is also developing a way for us to understand the language of the dolphins and get a fun art lesson in our 5-minute AI learning section (featuring amazing drawing skills)!
This week’s edition:
OpenAI introduces o3 and o4-mini
Google’s DolphinGemma: Communicate with dolphins
Your 5 Min AI Learning: Turn your doodles into AI masterpieces with Autodraw
New AI Tools
More AI News
AI Image of the Week
Read time: 10 Minutes
LATEST NEWS
OPENAI o3 and o4-mini
Image Courtesy of OpenAI
What You Need to Know
OpenAI has launched two new reasoning models, o3 and o4-mini, representing the most intelligent and versatile models in the o-series to date. These models are designed to think more deeply and for longer periods before responding, enabling them to solve complex, multi-step problems more effectively. Both models have full access to ChatGPT’s suite of tools—including web search, file analysis with Python, image understanding and image generation—and are trained to decide when and how to use these tools to produce detailed, accurate, and context-aware answers typically within a minute. This marks a significant step toward a more agentic ChatGPT that can independently execute sophisticated tasks on users’ behalf.
The Finer Details
o3 is the most powerful model, excelling in coding, math, science, and visual tasks, reducing major errors by 20% over previous versions.
o4-mini is smaller, faster, cost-effective, and excels in math, coding, and volume-heavy tasks with top benchmark scores.
Both models follow instructions better, provide more useful and verifiable answers, and maintain natural, memory-aware conversations.
They support multimodal reasoning, interpreting and manipulating images as part of their problem-solving process.
Trained with large-scale reinforcement learning, they improve with more compute and thinking time, outperforming earlier models at equal cost.
They autonomously decide when and how to use ChatGPT tools or APIs, combining web search, coding, and image generation for complex workflows.
So What?
OpenAI o3 and o4-mini represent a leap forward in AI reasoning capabilities, enabling ChatGPT to handle more complex, nuanced, and multi-modal tasks with greater accuracy and efficiency. For users, this means more powerful assistance in areas like coding, scientific research, data analysis, and creative problem-solving. The models’ ability to autonomously use tools and integrate visual information opens new possibilities for AI to act as a versatile, independent agent in workflows that require up-to-date information and sophisticated reasoning. The improved efficiency of o4-mini also makes advanced reasoning accessible at scale, supporting high-volume applications without sacrificing quality. Overall, these advancements push the frontier of AI utility and intelligence closer to real-world problem-solving and agentic autonomy.
DOLPHINGEMMA
Image Courtesy of Google
What You Need to Know
Google, in collaboration with the Wild Dolphin Project (WDP) and Georgia Institute of Technology, has developed DolphinGemma, a large language model (LLM) designed to understand, interpret, and even respond to dolphin vocalisations. Trained on 40 years of underwater audio-visual data of Atlantic spotted dolphins, DolphinGemma uses advanced AI and Google’s SoundStream technology to analyse and predict dolphin sound sequences, aiming to establish two-way communication between humans and dolphins. The project also includes the CHAT system, an underwater computer that helps create a shared vocabulary with dolphins by associating synthetic whistles with objects dolphins like to play with.
The Finer Details
DolphinGemma is a ~400 million parameter AI model that processes dolphin sounds as sequences, predicting subsequent sounds similarly to how human language LLMs predict text.
The training data comes from the Wild Dolphin Project’s extensive, non-invasive research since 1985, involving detailed recordings of individual dolphins’ sounds linked to their behaviours and identities in the Bahamas.
The AI identifies recurring sound patterns, clusters and reliable sequences to uncover hidden structures and potential meanings within dolphin communication, a task previously requiring immense human effort.
CHAT (Cetacean Hearing Augmentation Telemetry), developed with Georgia Tech, is an underwater computer system enabling dolphins to mimic synthetic whistles linked to specific objects, facilitating interactive communication.
The AI model runs efficiently on Google Pixel phones underwater, allowing real-time data collection and analysis in the field, with plans to upgrade to Pixel 9 devices for enhanced performance.
Google plans to release DolphinGemma as an open model to the research community, enabling broader study and potential application to other dolphin species and marine mammals.
This initiative is part of a broader trend where AI is being used to decode animal communication, with other projects targeting species like crows, whales and meerkats.
So What?
DolphinGemma represents a pioneering step toward bridging the communication gap between humans and one of the most intelligent marine species. By automating the decoding of complex dolphin vocalisations and enabling interactive exchanges, it could transform marine biology, conservation efforts, and our understanding of animal intelligence. The project not only advances AI technology but also fosters a deeper connection between humanity and the natural world, potentially opening new avenues for interspecies dialogue and cooperation.
AI LEARNING
YOUR 5 MIN AI LEARNING
Courtesy of the best drawer on the MegaSuperAI team
For this week’s AI newsletter, we’re excited to spotlight a fun and handy tool called AutoDraw. If you’ve ever struggled to turn your rough sketches into polished drawings, this AI-powered website is here to help. Just start doodling, and AutoDraw will suggest neat, professional-looking icons and illustrations that match your scribbles. It’s like having a smart art assistant that instantly cleans up your ideas and makes your sketches look amazing—no artistic skills required! Perfect for quick visuals, presentations, or just having some creative fun. Give it a try and watch your rough lines transform into something surprisingly good!
Here’s How to Turn Your Doodles into Actual Art Using AI:
1. Go to Autodraw
2. Do your best attempt at drawing a car
3. On the top will be recommended items that the AI thinks you are drawing
4. Click your desired output
LATEST AI TOOLS
A tool to assist in efficient presentation creation and data visualisations. (Paid)
“Pitching to investors but falling short with bland, uninspired decks? Inabit.ai transforms your big ideas into sleek, investor-ready presentations in minutes. Convey your vision with clarity and style, leaving an unforgettable impression.”
A virtual Excel assistant that can interact with spreadsheets using English commands. (Paid)
“Effortlessly analyze Excel data using natural language with Excelmatic. Ask questions like "calculate monthly sales" and instantly get results, charts (bar, line, pie), and data-driven insights. The intelligent Excel analysis tool for quick, comprehensive reports.”
An organisational tool to help construct business plans through interactive collaboration. (Freemium)
“Get a detailed report featuring a comprehensive market analysis of your target audience, a competitive analysis, financial forecasts, and suggestions for an appropriate name. Collaborate with our AI to polish and perfect your business concept. Incorporate additional insights and suggestions to enhance your initial idea.”
MORE AI HIGHLIGHTS
A recent survey by the Global Risk Advisory Council highlights that association with Elon Musk, particularly due to his polarising public image and ties to former President Donald Trump, ranks as the second highest reputational risk for companies, while misuse of artificial intelligence remains the foremost threat, with risks escalating due to lack of regulation and potential for misinformation. Backtracking on diversity, equity, and inclusion commitments under political pressure is identified as a significant reputational concern, warning CEOs that short-term concessions could cause lasting damage to stakeholder trust.
Cursor, an AI-powered code editor, faced a major backlash after its AI customer support bot falsely claimed a new login policy that forced users to log out when switching devices, causing confusion and subscription cancellations among developers relying on multi-device workflows. The company apologised for the AI "hallucination," clarified no such policy exists, refunded affected users, and highlighted the risks of deploying AI in customer service roles without clear transparency and human oversight.
Scammers are increasingly using AI to create convincing fake profiles—including deepfake videos, fabricated resumes, and professional headshots—to secure remote jobs, enabling them to steal company secrets or deploy malware once hired; Gartner predicts that by 2028, one in four job applicants will be fraudulent. This surge in AI-driven identity fraud has forced companies like cybersecurity firm Vidoc Security to overhaul hiring processes by requiring in-person interviews, while experts urge employers to adopt stringent verification practices such as scrutinising LinkedIn profiles and cultural questioning to detect impostors.
In a New York employment dispute, plaintiff Jerome Dewald attempted to present his case using an AI-generated avatar as his legal counsel, but the court quickly halted the video upon realising the "lawyer" was not a real person, with the judge expressing strong disapproval for being misled. Dewald, who has an AI startup aimed at legal self-representation, admitted to creating the avatar to overcome his difficulty with oral argument but the incident highlighted the current legal system’s unpreparedness and regulatory gaps regarding AI use in courtrooms.
AI IMAGE OF THE WEEK
“Not Sponsored” by MegaSuperAI
Happy Easter, everyone! Today is a wonderful day filled with joy and anticipation as families and friends come together to celebrate. I hope you've all had a chance to dive into your Easter eggs—I've already devoured half of my Snickers egg, and it's as delicious as I remembered! The day is alive with laughter, the excitement of egg hunts, and the warmth of shared meals. As we gather around the table, let's cherish these moments of togetherness and the spirit of renewal that this season brings. This week's AI image will be themed around my snickers easter egg. Let us see how an AI will interpret that.
Image prompt using Poe AI:
"An artistic interpretation of a Snickers Easter egg, featuring the iconic chocolate coating, caramel, and nougat layers. The egg is partially opened, revealing its delicious fillings, surrounded by vibrant spring flowers and colourful Easter decorations. The scene captures the essence of Easter celebration, with a joyful atmosphere that reflects the delight of indulging in festive treats."
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: