The Most Accurate and Amazing AI Predictions of 2024

🌈 Abstract

The article discusses the challenges in making accurate predictions about the future of AI in 2024.

🙋 Q&A

[01] Why haven't you made any 2024 predictions for AI yet?

• The author notes that AI delivery cycles are becoming exponential, making them hard to predict.
• The author suggests that:
  • None of us have a clue what AI might bring in the next six months.
  • Business predictions, which are incremental, are unexciting.
  • Incremental predictions based on developments in 2023 are also mundane.
  • No one really understands what Q* is, and numerous random predictions are made based on it.

[02] What are some of the key AI predictions for 2024?

• Significant investment in AI research in areas of planning and reasoning, but it's unclear if we'll be able to deliver on reasoning in 2024 or 2025.
• AI research is significantly investing in Quantization, not smaller models, for the past 5 years, and 2024 may be the year this proves effective.
• The cost per million tokens for Large Multimodal Models (LMM) could be reduced by 1/20th if quantization proves effective, but this is speculative.
• Hierarchical "Large Convolutional Models" (LCMs) are showing astonishing results in the labs, but their potential impact is still a mystery for 2024.
• 2024 will be the year of the Large Genomic Model (LGM), pioneered by Cognit, and a continued shift towards an "AI First" approach in drug discovery, precision medicine, therapeutics, and healthcare.
• Investments in AI startups may decline as closed models from large tech companies encompass more features, making it harder for smaller AI shops to compete.
• People will increasingly experiment with Open Source, realizing the importance of "engineering" and "scale" and that running open-source solutions on-premises is not sustainable.

[03] What other predictions are mentioned in the article?

• The "EA and e/acc cult" will get louder and more annoying.
• AGI will be blown out of proportion, and journalists will claim to know the deep inner technicals and exactly how AGI will work when it arrives in the future.
• More bloggers will emerge as AI experts, asking ChatGPT to write their blogs.
• All these predictions could be proven wrong with a breakthrough in less than six months.