Featured in the LIT (Ladies in Tech) Newsletter
In our series on Technology Terms, we're focusing on AI—Artificial Intelligence. AI is technically applied theory over data, resulting in a solution set. It generally provides information based on an ask, analyzing details within its language model. Large language models (LLMs) can be built from internet search metadata attached to pictures, posts, messages, or anything, demonstrating the broad capabilities of AI in processing and understanding vast amounts of information.
Why it matters: The first thing you will hear about AI is the power requirements. Very similar to thinking about how long it would take you to pour over a blob of data, a computer uses more resources to do the same task in less time. So, if it takes you ten weeks to go through a file box, it may take the computer 10 seconds, but it has many cores working through the data, like lots of tiny helper bees.
The controversy: Garbage in, garbage out, to start. That problem is going to stay in computing. While people can do their best to ensure that output information is factual, the hours required to ensure the original large language model is correctly populated are sometimes prohibitive. Further, we know that coding carries the bias of the coder.
Sometimes, that bias makes the outcomes incorrect. For instance, a recent company launch of generative AI failed to draw anyone white, even if they were white. There were problems with facial recognition for people of color when it first launched. Critical thinking and lots of testing will be essential! Also, most large language models are based on recent info. 95% of the
world's data was created in the last few years. Some results warn you that they are "accurate."
Things change! Do the due (diligence, that is)! Also controversial, who will watch AI? Will we have AI that checks AI? Which language model will be "correct?" Will controversy happen faster?
Further controversy: Holy cow at the power! A typical ChatGPT search is 5x the power of a traditional web search engine. Use it when you have to use it to be a good environmental steward. Generative AI can use 20-30x the power of other applications. Some of that consumption goes away after the generative part. But it's still a lot.
Tidbits and things to know: If you create content with AI, it is automatically considered part of the public domain and cannot be copyrighted. If you do use that content, it is at your own risk.
Also, if you upload company docs to an AI engine, whatever you upload becomes part of the learning model. Upload confidential info and docs at your own risk. It's probably worth paying attention to the warnings. Some AI that is trained on smaller LLM may have greater accuracy.
For instance, a program that runs over helpdesk tickets has fewer bits of fixed information for its reference set.
Kommentarer