Obviously generative AI is the in-thing now. I’ve been playing around with ChatGPT and every one of the publicly available models for months now and it really makes me feel like we’re living in a sci-fi scenario. There are new announcements in the field practically every day and it’s exciting to try to keep up.

• We’ll still start with some non-AI news first. Most people will have heard of how sperm counts in men have been falling. This study talks about how men who do physically demanding jobs have much higher sperm counts compared to the baseline. This may be linked to how they also have higher testosterone levels. Most other studies don’t show a direct link between testosterone levels and sperm count but it’s possible that they could be linked in more indirect ways.
• Next we have a story about an American man who developed an Irish accent after being afflicted with prostate cancer. This is thought to be an instance of paraneoplastic neurological disorder in which a cancer patient’s nervous system is attacked by their immune system. This isn’t an altogether new finding but it is the first instance of the cause being prostate cancer metastasizing and affecting the brain. Plus I’m always fascinated by how personalities and demeanors of people can change drastically as a result of injury or disease.
• There are far too many announcements in the AI space to cover so I’ll limit myself to news that is more applicable to the mainstream. By now most people will have heard of GPT4 and how it accepts images as an input, in addition to text. This article goes more deeply into such multimodal inputs for AI, integrating text, images, video and sound. It talks specifically about an in-house model made by Microsoft called Kosmos-1 that is unavailable to the general public and is supposedly far better than anything else. The important thing is that such a model would be much closer to a general purpose AI that we think of in science-fiction, able to receive all manner of sensory input from the outside world and respond accordingly.
• Next is a paper about the image generation model Stable Diffusion. The researchers talk about how they are able to take images of human brain activity from fMRI and run them through the model to reconstruct an image of what the human subject was seeing at the time of the FMRI scanning. The results are extremely impressive and the surprising thing is that the model is able to do this without any additional training or fine-tuning.