simedw

That's true, it's also why I didn't benchmark against any other model provider.

It has been tuned so heavily on this specific format that even a tiny change, like switching the order in the `box_2d` format from `(ymin, xmin, ymax, xmax)` to `(xmin, ymin, xmax, ymax)` causes performance to tank.

That's really interesting, thanks for sharing!

Are you using that approach in production for grounding when PDFs don't include embedded text, like in the case of scanned documents? I did some experiments for that use case, and it wasn't really reaching the bar I was hoping for.

Thank you! Better training data is often the key to solving these issues, though it can be a costly solution.

In some cases, running a model like SAM 2 on a loose bounding box can help refine the results. I usually add about 10% padding in each direction to the bounding box, just in case the original was too tight. Then if you don't actually need to mask you just convert it back to a bounding box.

It was actually a bit worse than that the LLM never got the full recipe due to some truncation logic I had added. So it regurgitated the recipe from training, and apparently, it couldn't do both that and convert units at the same time with the lite model (it worked for just flash).

I should have caught that, and there are probably other bugs too waiting to be found. That said, it's still a great recipe.

Fantastic catch! It led me down a rabbit hole, and I finally found the root cause.

The recipe site was so long that it got truncated before being sent to the LLM. Then, based on the first 8000 characters, Gemini hallucinated the rest of the recipe, it was definitely in its training set.

I have fixed it and pushed a new version of the project. Thanks again, it really highlights how we can never fully trust models.

You are right, it's Gemini 2.5 Flash Lite

If the goal is to have a more consistent layout on each visit, I think we could save the last page's markdown and send it to the model as a one-shot example...

Just had a look and three is quite a lot going into Firefox's reader mode.

https://github.com/mozilla/readability

We’ll probably have to add some custom code to log in, get an auth token, and then browse with it. Not sure if LinkedIn would like that, but I certainly would.

That's a cool project.

I think most of it comes down to Flash-Lite being really fast, and the fact that I'm only outputting markdown, which is fairly easy and streams well.

Thank you.

I was thinking of showing multiple tabs/views at the same time, but only from the same source.

Maybe we could have one tab with the original content optimised for cli viewing, and another tab just doing fact checking (can ground it with google search or brave). Would be a fun experiment.

Very interesting breakdown, OP have you deep dived in pgvectorscale as well?

After transcribing the speech into text, it seems to first translate it into English before send it over the GPT-3 (davinci). I wonder if that will lead to a bunch of mistakes, can't GPT-3 handle Mandarin directly?

I did some experiments and using the same prompt but without translating it seem to lead to better results.

Useful tool!

Do you not need to escape the ampersand in the second example? otherwise bash will run it in the background and skip the `filters=t2,m4` part.

Pretty cool but the LinkedIn Card gives me the following PHP exception: Catchable fatal error: Object of class stdClass could not be converted to string in /home/stephenou/socialbusinesscard.me/card_linkedin.php on line 49