Predictions for 2025
Reading the tea leaves for the future
Putting in the effort to make high quality predictions focuses the mind. It forces you to consider not what you want to happen, but what you think will happen. These can be two very different things. You also tend to make better predictions if you are held accountable for them. That's why I revisit my predictions at the end of each year, publicly, to see how I did. Below is what I think will happen in 2025.
Massive energy infrastructure bill: With the power hungry demands of AI and the current need to update the power grid, I predict a massive infrastructure bill to fund and improve the grid and make it easier to build power generation to meet the demands of AI. This may even have an overtone of remaining competitive in the global landscape.
Advertising in AI results: With the prominent rise of AI use in day-to-day activities, I believe model providers will start to create systems allowing advertisers to bid on either embedding content in or shaping the language of model outputs. This becomes easier with the current focus on test-time compute to create multiple versions to show the user and then selecting the best one. This makes the process ripe for custom review functions that can be tailored to bidders.
Breakout year for AI use in biology and medicine: This has been gaining steam over the past few years. AI is wonderful for medicine and bioengineering because the search spaces are so vast that large gains can be made from cutting swaths of bad paths. I believe this year we will see an exponential increase in the amount of discoveries made due to the help of AI. This could be finding new drugs, cures, therapies, or vaccines or determining them at a faster rate. We might even see predictive vaccines that try to determine how various viruses and bacteria will mutate and create remedies for those.
Personalized software: I previously wrote about personalized software and Argon Ventures put it on their inspiration board. This coming year businesses will start to create their own software using AI, fitting the software to their existing process. This will look like companies transforming their business with AI and buying less SaaS solutions.
Difficulties in managing a large number of AI systems: As companies create many AI processes and workflows, I predict a lot of issues in managing and maintaining these automated workflows. I believe multiple solutions will arise to deal with this issue along with companies opening positions to manage and maintain these automated workflows.
Growing morphic architectures: Right now, when a new model is trained, it is typically done from scratch (transfer learning being an exception). This means a lot of time and money is spent creating a model only to abandon it when a new, better model is needed. If we did this with children, people would be horrified. What I predict will arise this year are methodologies to create new models bootstrapped from existing models. Sort of like how human brains still have parts from reptilian brains. This could look like using existing model weights as a starting point then coupled with random weights in the expanded architecture.
Cutting headcount language: People have been shying away from discussing how the use of AI requires a lower number of employees. However, I believe this year, as the landscape becomes more competitive, AI providers will push on language that their technology can be used to lower head counts.
Shifting towards adaptation: This year I believe the world will realize that we will blow past the 2 degree Celsius increase set forth by the Paris accord. Instead of attempting to stop climate change, the conversation will shift towards adapting to changing conditions. Not a place I want us to be but I think it will happen.
Those are my predictions for 2025. As I did with my 2023 and 2024 predictions, I'll revisit these predictions at the end of the year to see how I did. I welcome any feedback or to hear your own thoughts or predictions for the new year. Happy New Year!