7 AI Trends for 2024 - DSBoost #47
Podcast Notes 🎙️
Jon and Sadie St. Lawrence talk about their predictions and what 2024 will bring in Data Science.
Here are the key takeaways:
Sadie’s predictions are:
Hardware evolution: Demand will increase for data and complex models. ➡️ We need better hardware;
LLMOS: LLMs will kill Operating Systems.
Slow-thinking model: LLMs with more computational and mathematical knowledge will arise;
Tool consolidation: Full stack data analysis apps will appear.
Workforce Upheaval: Data roles are changing. Old jobs will disappear and new ones appear.
Jon’s predictions for 2024 are:
AI bubble bursting: In 2023 there were too many AI startups. Only the companies with solid foundations will remain.
Breakthroughs in Edge AI: Being able to run LLMs on a phone will be common.
Let’s dive a little deeper into these topics:
In the upcoming year, we are going to see an increased demand for data and hardware. The industry will be “data and hardware hungry”.
We have to add more data to existing models to perform better. Thus we need some innovations in the field. But not only by the most important companies, like Nvidia. We will probably see new companies who recognize the opportunity.
Jon mentions Qstar, a new method for training our LLMs with less data. With tools like Qstar, we can be more efficient with our data. It is important in industries like healthcare where data is gold.
What if large language models will replace operating systems?
The way we are working today is going to be disrupted by LLMs like it was by mobile phones. We are going to switch from click and point workflow to communicating with our LLM assistant, who will do what we ask.
This level of technology will not be available this year, but it's an interesting concept.
What exactly is a slow-thinking model?
LLM models today are fast-thinking models. They are good at the creative side, but they are just throwing words together unconsciously. These models are helpful no doubt, but they are still not at a good level in topics like math.
This is where slow-thinking models come into play.
These models are good at solving complex problems at a high level. They could help in fields, like physics, and healthcare where they could be used for discoveries. We will see some advancements in this niche, but there are also some concerns. They may be super good at mathematics and could encrypt existing security walls.
Organizations are still going to have budget constraints this year. Because of that, the tools that can do all the processes in one will be more searched and valued.
A great example is Microsoft Fabric. It connects to any data cloud and does your data governance, ETL, and visualization. It is a full data tech stack all-in-one tool. Similar solutions will be much cheaper for a company with budget constraints.
Workforce Upheaval - Old jobs will disappear and new ones will appear.
Sadie shares a story when she got into data science over 10 years ago.
At that time companies employed full-stack data scientists. They had to clean and manipulate data, build visualizations, and communicate the results. These processes are now split up into multiple job functions.
Those range from data engineers to data analysts and data project managers.
Now, these roles will also transform into something different with LLMs.
AI bubble bursting
The AI bubble is going to burst in 2024!
In 2023 a new AI startup appeared every minute. This trend cannot continue. Investors will want to see some returns on their investments. A lot of AI startups will disappear and only the companies with a good foundation will remain in this field.
Breakthroughs in Edge AI
In the next few years, we will be able to run large language models on a phone!
Now we are limited. We need to send a query along an API to some central server that does heavy-lifting computing and send the response back.
But can you imagine when a smartphone can run its own LLM?
This would open up a lot of possibilities in many industries like agriculture, mining, and healthcare.
Stay ahead of the AI wave! 🌊
New in the field? Before starting to learn Data Science and Machine Learning you need to build a solid foundation. Check this thread to learn more about it:
Do you already have a solid foundation? Time to learn Data Science:
You can learn all of these for free thanks to these 6 YouTube videos:
Enjoy and good learning!