Saw a bee on my roof this morning which I hadn’t seen before. It was a cold morning. The bee was still and sitting. Some shifting of weight with the wind. I took a photo of this bee. But there is a bee whose photo I could not ever take before, because ever since I started coming to this rooftop every day for a year, during the warmer days there would almot always be another bee a kind of yellow jacket bee, that would always aggressively say hello right up to your face, so there was never really a time I would attempt or even think of taking its photo as it tries to hover in front of my face....

I got a laugh, showing this silly 😛 before and after to a family member recently. But this expensive silliness is just an example of an otherwise very versatile technique. I remember still for most of the brief lifetime of ChatGPT that this kind of image manipulation using a prompt, with autoencoder latent space manipulation was not possible. The only models available for most of ChatGPT history were like DALL-E text to image word2vec types, trained by using a dataset of text and image pairs....

At my work, our team meets daily to discuss our goings on but I think the part I think everyone gets the most utility from is the affectionately named parking lot, where we talk through issues that are blocking people in their efforts and the team does ad hoc resolution sessions. I still recall when a colleague had introduced this practice to our team a while back, as a modification to the typical standup....

Of course, when Apple Notes added math, I thought and still think it’s really clever. But noticing just now a good reason to check the math haha 😅 I was checking some walking pace analysis because for some reason haha, my Strava doesnt consider walking a work out so it doesnt give you pace popped into Apple Notes, 12:18 minutes /0.56mi ‎ = 32.143 minutes/mi 45:10 minutes / 2.21mi ‎ = 4....

About a year ago, 2024-08-22, I was reading Leopold Aschenbrenner’s post1 on AGI and I had jotted down a musing about a scaling law observations, but I forgot to write about it. I found my note today. So I get that it is possible to look at charts comparing compute and performance and many conclude scaling laws mean AGI is inevitable. I like Yann Lecun’s perspective here that if transformer models are giant lookup tables, then making them bigger will make them appear more impressive, but this is not sustainable....