ChatGPT's knowledge of language was apparent from the start. It was fun, at first, to jailbreak. So, a lot of work went into containing chatbot's dark side. And a lot of work and computing power has gone into training models and into the inference engines that use the models. Multiple hyperscale corporations are building out capability.
https://www.scientificamerican.com/article/the-ai-boom-could-use-a-shocking-amount-of-electricity/
Is all this sustainable? I have my doubts. Yet, I have used and do use AI. AI is built into many of the search engines. On my Pixel phone, AI is built into the default editing capabilities. Of course, one can not only modify images, one can generate images. One can generate images by using written or verbal instructions.
The improvements are amazing (to me) however
https://news.mit.edu/2025/mit-researchers-speak-objects-existence-using-ai-robotics-1205
I can't keep up. But, I think that I should be at least familiar with AI models. So, I used AI (mostly Gemini) for parts of this year's holiday letter.
Some examples,
The photo below of this tree decorated with gold balls is too busy because of the background. The gold is not as bright as it looked in person. So, I asked the AI to remove the background and brighten the image. Much easier to do than in a photo editor.
The bear did come to the front door, that is true. However, the image that the doorbell cam captured was blurry. The blur was caused by motion. I asked the AI to focus the image. Perhaps, it took a little too much creative freedom?
In my opinion, videos generated by AI are not as believeable. They are improving. Both videos were generated from the same prompt. The second video was created six months later by an improved generation of the AI. You can click on each frame grab below to see the whole video.
