It’s been an exciting year for Evolution AI and AI in general. Three notable developments from AI in 2024 include:
High-performing models like OpenAI’s GPT-4o and Anthropic’s Claude 3 family have entered the scene. A new wave of popular AI tools – such as the video generation tool Runway and voice and music generation tools like ElevenLabs and Suno - showcase the strides that generative AI has taken since 2023.
The gulf between academic and industrial AI development continues to widen. In 2023, a significant disparity emerged in machine learning model development, with industry producing 51 models and academia contributing only 15. We’ve always championed academic and industrial research at the London Machine Learning Meetup (the largest AI community in Europe, which we organise).
Research now confirms AI’s ability to improve productivity. MIT Sloan released a study that shows AI could improve a skilled worker’s performance by 40%. However, these productivity gains are contingent on a worker using AI within the boundaries of its capabilities (the study used overly complex ChatGPT prompts as an independent variable). If not, worker productivity actually declines by 19%.
These are just a few examples – a flavour of AI-driven innovation.
Our most-read articles from last year often focused on the practical aspects of AI. Seeing how many readers want to learn how to use AI tools like ChatGPT effectively is gratifying. On our end, we work with these tools closely and enjoy sharing our insights in the form of articles.
.png)
As a use case, ChatGPT-driven document comparison has proven surprisingly popular. It’s easy to use ChatGPT to compare files, as its commentating capabilities can effectively identify the files’ similarities and differences.
In this article, we break down the process into three steps. As with many ChatGPT-related tasks, it’s all about finding the right prompt.
.jpeg)
Despite an article like the one above that affirms ChatGPT’s usefulness in simple document-related tasks, we take a more grounded approach in this article. We infused this piece with technical insights from one of our Senior Machine Learning Engineers.
The result? A technical walk-through that reveals how generative AI often produces inaccuracies when reading PDFs.
.png)
We expected this article to gain a large readership – and it has. Three major iterations of large language models (LLMs) were released over two months, between May and June 2024. Each model had different strengths and weaknesses, which we clearly outlined.
We compared the three models using several criteria, such as HumanEval (code generation). This article also marks the second collaboration between our tech and writing teams, proving that there’s a large readership for well-written, accessible technical insights.
Like ChatGPT-driven document comparison, extracting data from annual reports represented a large use case.
In this article, we dissect the state of current annual report extraction technology, the challenges and how to take advantage of newer extraction tools.
Despite a year of (undeniably) exciting AI advancements, some companies falsely label their technologies as AI. Such AI washing is a threat to bona fide AI models. The field must remain trustworthy and credible with so much general distrust in AI. With greater awareness and education on what makes a genuine AI product, we hope to see a reduction in AI washing in 2025, allowing authentic innovation to shine.
Stay updated with Evolution AI’s insights by following us on LinkedIn.
