End of Year Review 🎉

 

It is that time of year when we look back before moving forward. Generative AI headlines dominated the previous 11 months, involving the biggest brand names in tech to humble researchers voraciously working on the next best model or product feature. Consumers mutated into creators and innovators, eagerly awaited updates and product roll-outs to progress ideas prompted by others. The prompt - the newest pseudo programming language allowing non-technical AI enthusiasts to get on with it- is done chiefly in English via text or natural language without the tedious and lengthy learning curve often associated with the discipline and skill of programming. AI has come a long way since Rosenblatt’s perceptron in the 1950s.

Nevertheless, artificial neurons still underpin the algorithms driving machine learning systems, which may seem anthropomorphic. But it is just ancient math, particularly probability and statistics, linear algebra, and calculus. The true complexity lies not in language models or machinery but in the profound change in society, culture, and economy.

Building societies that thrive in the age of AI will require substantial changes to our economy but also a shift in culture and values. ~Kai-fu Lee

Over a decade ago, digital media shifted the landscape of many industries and human lives, from learning, working, socializing, and entertaining ourselves to how we perceive our digital selves. During this time, I made a hard fork into human-centric digital transformation with a novel idea (at the time) and entrepreneurial model that allowed for agency and staying nimble. I was ahead of my time, but now, with the advancements of AI, smaller teams, even companies of one or two, are increasingly popular. Doing more with less also comes with some perks, like moving losses to the other column of the P&L statement.

That non-traditional thinking led me to take an interdisciplinary approach to learning and working with AI, which incorporates cloud, blockchain, data science, and, more recently- cybersecurity. I can attest to AI assistance helping me to become an expert shortening the learning curve and low-level tasks on the list. However, looking back, I recognize the reward of having the capacity for disciplined learning and the capability to problem-solve ambitiously.

Generative AI broke new ground in 2024. I kept a daily log of rapid change and progress—let’s get into it.

 

Timeline

 

January— Fresh from Europe’s Black Hat conference with adversarial machine learning on the brain, I’m intrigued by OpenAI’s slated release of a computer using AI tool codenamed “Operator.” It did not happen.

February— Azure OpenAI and security are a focus. Meanwhile, Apple unveils Apple Intelligence, prioritizing security, and OpenAI introduces a game-changing text-to-video generator, Sora, to be tested by a select group. The sample short videos produce clips from prompts resembling Hollywood productions.

March— Cloud and security are still in focus. Anthropic releases the Claude 3.5 model for free for Android app users.

April— Security-focused GenAI research is underway, especially with OpenAI phasing out ChatGPT plugins in favor of custom GPTs. Hollywood talent agency CAA unveils the CAA Vault initiative, which creates AI replicas of creatives.

May— While I focused on Azure OpenAI and security, Kuaishou, a Chinese video-sharing app, launched a video generator ahead of OpenAI’s public release of Sora.

June— AI integration with blockchain via the Internet Computer Protocol research is underway. Former OpenAI’s Leopold Aschenbrenner releases “Situational Awareness: The Decade Ahead” and a French organization launches Moshi, a real-time AI voice assistant with plans to open-source research.

July— AI<>Blockchain ensues with daily security hardening. Runway’s Gen 3 AI video generator is a hit.

August— Adversarial Machine Learning has my attention. Google researchers create an AI simulated game engine in real-time, allowing real-time interaction.

September— I’m strengthening my security posture and researching ChatGPT vulnerabilities. OpenAI unveils a new ‘reasoning’ AI model, o1.

October— The same security workflow, but agentic AI takes center stage. Meta announces MovieGen competing with OpenAI Sora. Anthropic releases the 3.5 haiku model and computer-use beta.   

November— Agentic AI and security workflow continues. Microsoft Magentic One, a multi-agent framework built on open-source Autogen, is introduced.
 

December— Agentic AI and security workflow continues. Sora publicly launches; multi-modal Gemini 2.0 flash unveils, with the vision feature redefining classical image processing; a preview is available for developers and other testers. I tested it along with DeepMind’s newest VEO, including an image-to-video generator - available on Vertex AI in private preview along with image (text to image), music (text to music), and Whisk for promptless play, which I used to generate a profile image. 😉

 

Looking Ahead  

OpenAI’s announcement of the multi-modal agent tool Orchestrator earlier in the year has set the stage for an exciting January release. Gemini 2.0 Flash will also be released to the public in January. The coming year promises accelerated progress in agentic AI, more mobile integration, and enhanced video generation. The future of AI will be characterized by fierce competition and groundbreaking innovations.

No code tooling and AI agents are progressive and convenient, but they can broaden the attack surface. Invoking invaluable human expertise through learning and application across new domains is crucial. For non-technical AI enthusiasts, understanding how AI decisions are made and implementing standard security practices when using AI agents, such as isolating your local system using a virtual machine, containerization, and preventing access to sensitive data, is essential. Remember, a general rule of thumb is to trust no AI and remember that AI still hallucinates confabulates. Your role in the AI process is indispensable.

Progressive workflows will be more common in security-focused AI research. I look forward to continued sharing and exchange.

References: Quote by Kai-Fue Lee. AI Super-Powers: China, Silicon Valley and the New World Order