Great thoughts on statistics. I’m taking a course on data science (yes I have an undergraduate degree that covered this but needed a refresh) and am reminded how valuable data analysts/ scientists are. Otherwise who knows what crazy conclusions we might base decisions on. I appreciate your e…
Great thoughts on statistics. I’m taking a course on data science (yes I have an undergraduate degree that covered this but needed a refresh) and am reminded how valuable data analysts/ scientists are. Otherwise who knows what crazy conclusions we might base decisions on. I appreciate your examples of how easily data is misrepresented. Which leads to the garbage in garbage out discussion. With Deepseek open code now enticing companies to build their own on prem LLMs, what happens if they don’t invest in the data experts to model, test and clean data for training. Oops Deepseek trains itself. brilliant yes but who trained its data science knowledge???? Am I missing something or is this just a small thing that needs to still be figured out?
This is a very large thing that needs to be worked out. I think that there is much more heat than light being generated in the DeepSeek story. Who knows where crazy conclusions made by LLM come from. The problem with them creating misinformation (euphemistically called hallucinations like it is a nice drug induced experience) is built into the fabric of what they are. I don’t believe it is solved until we create hybrid AI systems using other AI architectures
Training data sets is a large issue itself. But in our current rush to market we are often not even sure what is being used. I remember proposing an AI pilot for a client 7 years ago and trying to explain we had to leave at least one third of the data to test the predicted outcomes of the recommendation engine model. Was not understood and project not undertaken
So long as most people using AI treat it like magic instead of what it is - extremely sophisticated software than needs to be managed as all technology does - we will continue on our current path of chaos
Thanks David! I guess there is a bit of - “if we believe” going on. People love the ‘easy’ button and the news folks love the drama. If it’s too good to be true there is usually a catch. It’s why a data scientist just smiles politely when an enthusiastic conversation starts speculating AI as a universal panacea. It really is amazing what AI can do and how it can help but Education is key on understanding data, training data, prompts, and probability.
A friend recently called me so excited about the future her chatbot said was ahead for her. Mmm what is the probability of a prediction?
Yes love the new product names. Themes are fun.
Great thoughts on statistics. I’m taking a course on data science (yes I have an undergraduate degree that covered this but needed a refresh) and am reminded how valuable data analysts/ scientists are. Otherwise who knows what crazy conclusions we might base decisions on. I appreciate your examples of how easily data is misrepresented. Which leads to the garbage in garbage out discussion. With Deepseek open code now enticing companies to build their own on prem LLMs, what happens if they don’t invest in the data experts to model, test and clean data for training. Oops Deepseek trains itself. brilliant yes but who trained its data science knowledge???? Am I missing something or is this just a small thing that needs to still be figured out?
This is a very large thing that needs to be worked out. I think that there is much more heat than light being generated in the DeepSeek story. Who knows where crazy conclusions made by LLM come from. The problem with them creating misinformation (euphemistically called hallucinations like it is a nice drug induced experience) is built into the fabric of what they are. I don’t believe it is solved until we create hybrid AI systems using other AI architectures
Training data sets is a large issue itself. But in our current rush to market we are often not even sure what is being used. I remember proposing an AI pilot for a client 7 years ago and trying to explain we had to leave at least one third of the data to test the predicted outcomes of the recommendation engine model. Was not understood and project not undertaken
So long as most people using AI treat it like magic instead of what it is - extremely sophisticated software than needs to be managed as all technology does - we will continue on our current path of chaos
Thanks David! I guess there is a bit of - “if we believe” going on. People love the ‘easy’ button and the news folks love the drama. If it’s too good to be true there is usually a catch. It’s why a data scientist just smiles politely when an enthusiastic conversation starts speculating AI as a universal panacea. It really is amazing what AI can do and how it can help but Education is key on understanding data, training data, prompts, and probability.
A friend recently called me so excited about the future her chatbot said was ahead for her. Mmm what is the probability of a prediction?