Here's the link to the interview: https://www.labxchange.org/library/pathway/lx-pathway:53ffe9d1-bc3b-4730-abb3-d95f5ab5f954/items/lx-pb:53ffe9d1-bc3b-4730-abb3-d95f5ab5f954:lx_simulation:8bf7271d?source=%2Flibrary%2Fclusters%2Flx-cluster%3AModernPrediction

Victoria, I really like your post and your additional question about how we can effectively communicate risk to communities in spite of their possible (but reasonable) unwillingness to respond to these forecasts. For my final project, I am researching hurricane prediction, where citizens' response to warnings, or lack thereof, is a large problem. I can see how this process would only be made more difficult in a setting where we have even less confidence in the precise predictions made by the models at hand. Like the current state of earthquake prediction, as little as 80 years ago it was extremely difficult to know when and where a hurricane would hit. Hopefully, as computational and algorithmic tools continue to develop, we can one day reach the same levels of accuracy with earthquake prediction that we have with hurricane forecasting.

I found the second article in the NYT series, on "Learning how a 'large language model' operates," to be a good introductory discussion on how the back-end of ChatBots like ChatGPT run. One part of the discussion that I thought was particularly vague, however, was that of "Step 2: Collect lots of data." The article points to using existing repositories online and describes how they are broken down into "tokens" to help the software break down large pieces of texts more easily. However, it does not describe how these AI models know what these articles are about in the first place. The additional article I read, "AI Isn’t Artificial or Intelligent" from VICE, reports how the companies at the forefront of AI innovation are outsourcing the tedious work of data labeling and filtering to human workers, particularly underpaid laborers in South America and Africa. While some of the NYT articles acknowledged how human feedback and reinforcement learning were able to improve these ChatBots, they never pointed to the large-scale human-decision making that is integral to the training of these bots in the first place. Companies cleary benefit from the public thinking their technology is functional in a self-contained way, but this is simply not the case. The article goes on to discuss how many of these laborers are subject to low wages and poor working conditions, and how, in many ways, this modern-day exploitation of workers in the Global South is reminiscent of their colonial past.

Jack, I really liked your discussion of diversifying our energy sources as a general form of risk management in the climate crisis, while also acknowledging the risk that other energy options present. You also pose a really good question about the actual state of solutions when it comes to utilizing alternative energy, and I think it would be interesting to investigate this from a policy perspective, as well. Fossil fuel companies have vast resources, and I wonder how they may be using these to slow the transition to a more sustainable global energy ecosystem. In the interview I watched, mitigation techniques such as investing in infrastructural resilience were discussed. It's interesting to think about how both of these techniques can help prevent the detrimental effects of climate change.