The interesting part wrt. that post (which is pure larp, just look at the molt that posted it) is how it managed to get so many upvotes. That might actually be a relevant alignment question: what determines which AI memes an agent finds "upvote" worthy?

No I'm talking about the image segmentation feature: https://simonwillison.net/2025/Apr/18/gemini-image-segmentat...

But the bounding box stuff might work well enough in Gemini 3 to handle this case as well.

For covering the risk of mistakes I suggest considering ways of "visually quoting" the documents.

If the summary says "closing timeline: X" but there's an icon I can click that pops open an overlay with a visual cropped screenshot of that part of the original PDF - maybe even with a red circle around that detail - I can trust those summaries a whole lot more.

Gemini 2.5 has image bounding box and masking features that can help with this (sadly missing from Gemini 3.)

I've just finished processing thousands of documents using the Gemini Pro 3 vision model and it outperformed every OCR and image model I've tested by a long shot, perfect markdown with latex for the math every time.

The `echo` part seemed interesting but it made me think, debuggers need this as a built in feature. If I have

    array
      .slice(0, 10)
      .filter(s => s[0].toLowerCase() < 'm') // a<->l
      .map(s => s.toUpperCase());

It seems like it should be a common feature to be able to view between each array operation in a debugger without having to augment the code with `echo`

The out of bounds handling didn't seem all that good to me. Sure you can filter out undefined. You could also just make a function that returns an empty array if out of bounds, or array of 1 element if not.

     // JS
     function getElemFromGrid(grid, x, y) {
       return (x < 0 || x >= grid.width ||
               y < 0 || y >= grid.height)
         ? []
         : [grid.elems[y][x]]
     }
     
     ...
     
     neighbors = [
       ...getElemFromGrid(grid, x + 1, y + 0),
       ...getElemFromGrid(grid, x + 1, y + 1),
       ...getElemFromGrid(grid, x + 0, y + 1),
       ...getElemFromGrid(grid, x - 1, y + 1),
       ...getElemFromGrid(grid, x - 1, y + 0),
       ...getElemFromGrid(grid, x - 1, y - 1),
       ...getElemFromGrid(grid, x + 0, y - 1),
       ...getElemFromGrid(grid, x + 1, y - 1),
     ]

     
        

I also find grids made of 2 dimensional array to be code small. An array of arrays is NOT A GRID as there is nothing enforcing the inner arrays to be the same length as each other. Also, in efficienct code it would be better to provide a 1 dimensional array and bounds. Now, out of bounds checks based on accessing outside the array won't work. You need to check actual bounds. Again giving preference using a helper