2023 was truly a wild year for AI. It felt like new models were being released every day, there were incredible breakthroughs in multi-modality, and ChatGPT became a household name.
For the first time, significant numbers of people started using AI in their daily lives, talking about it, and understanding its impact. ChatGPT showed how far we've come in making computers smart, even though we're still figuring out the best ways to use this technology.
We saw big names like Google's Gemini and Anthropic's Claude joining the race. Moreover, the conversation around AI ethics and regulation intensified. Several countries and international organizations proposed or implemented new guidelines and laws to govern AI development and usage, focusing on privacy, bias mitigation, and accountability.
I would call this an inflection moment. - Pioneering AI scientist Fei-Fei Li.
That all happened last year, but what can we expect for AI in 2024?
We talked to some of the smartest minds in the space to get their thoughts and predictions on what’s next for AI in 2024.
We’ve loved using it as a personal sounding board. We often have an idea on the tip of our tongue or some analogy that we just barely can’t get right. Chatbots have given us a huge lift to get our creative thoughts over the line!
I am obviously biased: UI/UX design. Although we were already dogfooding and using Uizard internally for ideation and quick prototyping pre-2023, we are now using Uizard internally for 95% of all our product design needs. The introduction of our GPT-style Autodesigner feature and React export has changed everything.* What are you most excited about in AI in 2024? (Can be any trend, tool, upcoming launch)
Multi-modality is going to change the world. When we will be able to feed an AI model with text, image, sound and soon video, we will see the emergence of even more transformative applications.
Last year I underestimated the importance of LLMs in our everyday life. Now I hear things like: “I run every paper I write through chatGPT”. I don’t think we have reached a saturation point yet.
Also LLMs, but not in the way you think. I work in fraud and solvency categorisation, one would think I wouldn’t be deep into deep learning. But a lot of the techniques developed for LLMs are useful in other fields. Like assessing user behaviour. I am kind of surprised, but in my day to day we are reaching towards those techniques as a means of answering other questions besides predictive text.
I am genuinely surprised by how pervasive language models became so quickly. I think in 2024 im most looking forward to being surprised. There is a lot of talent out there, especially after all the tech layoffs in 2024, and a lot of bigger companies and industries ripe for disruption. So I’m looking forward to seeing what comes out of the mix, and being surprised when I look back at this next year.
Zapier + GPT-4 function calling (although the zapier part of this is easily the worst part). Also Supernormal for taking meeting notes and ChatGPT with voice for learning German
I am excited about seeing if progress is continuing at the rate it has so far — if so, I am very interested in the next bump up in size of models, e.g. GPT-4.5 or 5 quality. I am also excited about the impact that AI will have on real world productivity — I expect applications such as Spoke and many others that embed AI in people's workflows to make a meaningful impact on global GDP growth.
2023 has been marked by groundbreaking research, notably Stanford's agents paper, which significantly advanced the autonomous agents field. Additionally, the development of state-of-the-art open-source LLMs like Llama 2 and Mistral has been a highlight as these models have opened new doors for commercial use, showcasing the practical applications and potential language models across various industries. This combination of cutting-edge research and practical advancements illustrates why 2023 has been a year where AI truly shone.
AI has significantly enhanced my day-to-day work, especially in writing and pair programming. I regularly use LLMs to refine my writing, brainstorm ideas, and overcome writer's block. In programming, ChatGPT serves as a virtual senior developer, assisting with architecture design, cloud deployments, and crafting new React components. Its value is particularly notable in areas where I’m less savvy, like C++ (the last time I wrote C++ code was in 2016). LLMs’ adaptability and personalized assistance have not only saved me time but also enhanced the quality and efficiency of my work, making these tools indispensable in my professional life.
In 2024, I'm most excited about the evolution of autonomous agents. The prospect of agents performing tasks like sending emails, scheduling meetings, and booking flights with minimal human input is transformative. Achieving this level of autonomy hinges on developing AI that comprehends context, excels in information retrieval, and operates cost-effectively for even mundane tasks. OpenAI’s recently released GPTs is one step in this direction (I wrote a piece about why now is the time for useful autonomous agents (http://aitidbits.ai/p/the-rise-of-autonomous-agents).
Another area of anticipation is prompt engineering. 2023 witnessed significant progress with innovations like DeepMind's OPRO and Microsoft's EvoPrompt. Yet, a definitive solution to make language model-generated prompts more effective and reduce hallucinations is still on the horizon. Streamlining prompt engineering will lead to better-performing LLM applications, broader adoption, and lower barriers to utilizing language models, making AI more accessible to a wider audience.
I think the years previous to this there were some germinating ideas in the ML space 🌱 that bloomed (pun absolutely intended - see HuggingFace’s model) over the past year. Having been one of the people fanatically checking in on the little seedlings ever since the Attention is all you Need paper it has been satisfying seeing transformer models deliver real value to millions of people!
Without doubt ChatGPT 😎.
Using ChatGPT to help me problem-solve and source relevant information for my day-to-day work has not only helped me improve the speed and quality of my core tasks at work (from many examples, a selection: support in writing test cases, support in drafting more efficient code for time-consuming operations, regex superstar…) but has also allowed me to learn and work across different domains, for example I have leaned on ChatGPT to help me with managing IAM profiles on AWS.
I might be getting excited a little early but I am very curious to see which developments in Robotics will be unlocked through leveraging (multi-modal) advances in AI. For a podcast episode which touches on this, albeit briefly, check out Hard Forks’s podcast.
It is fair to say that last year was turbulent and exciting, but with further advancements and exponential growth of LLM capabilities and improved multi modality we can only imagine what 2024 has in store for us! We hope you guys are as excited as we are 🤩