2025-04-30 09:19:13
So cool to see this resource that @schickling and @jessmartin have created! A lot of hard work and so much value to the community
2025-04-28 07:27:53
Wow, o3 is a surprisingly good http://Mint.com replacement.
Try sharing a pdf credit card statement and asking for a categorized breakdown, and then discuss opportunities to save money.
Really nice that it makes good guesses about what entries are referring to
2025-04-25 10:35:59
o3 did a great job helping me debug some code today.
the key is print debugging. here's my workflow:
- ask the model to suggest a few theories for what might be wrong and then to add print statements that would provide useful evidence
- copy-paste the resulting logs back into the chat
- ask the model to provide an explanation for how the logs support the various theories.
- actually read its explanation and think about whether it makes any sense! (in this case today the explanation was great! see the example output in screenshot)
- only proceed with a fix once you're convinced of the explanation
in my experience this works much better than just having it read the code. for humans and AIs, debugging benefits greatly from dynamic feedback loops!
i've been doing this workflow since 2022 with gpt-3, but it seems like the latest reasoning models have gotten really good. which makes sense - testing observed logs against explanations seems like the kind of task that reasoning would naturally help with.
once models can drive interactive debuggers or even explore recorded execution traces via tools, this workflow is gonna get way nicer. (lmk if you know of tools that do this already?)
2025-04-23 09:37:24
Man,this is such a beautiful video series.
Seeing the amount of work it takes to build a good house is awe-inspiring.
https://youtu.be/d7e6IOdKfVw?si=QlQmPnZ7aj-JSVGa
