Other than the search multiselects, my favourite features are:
• Create bookmarks and filter by them
• Expand to see positioning, style, and target
• See and search price + discount %
• Darkmode UI
Yes, that is v23 - there was a lot of back and forth about the filters. I am very fussy about my search multiselects.
These are brilliant search multiselects.
I'm *very* happy with how it looked in the end!
Improvements were needed!
I went back to gemini a couple of times to get the price and product normalized
Honestly the first shot at the UI is pretty decent!
I have a VERY CLEAR idea of what I want this UI to look like.
Sonnet gets my full instructions + the typespec
This is too smol of a project to bother pulling out Cursor, Claude Code, or the like. Anthropic can manage the UI as a Claude Artifacts.
Honestly Claude Artifacts have replaced 95% of my personal coding projects.
The array's too long to throw into another LLM and also have it follow rules, so we'll just ask for a very descriptive schema. That can form the basis of our actual UI.
Deep research does not a UI make. It's structured data time!
2.5 Gemini's really the only one with the context length to make this happen. Throw in the deep research result and ask for JSON, with a bit of instructions...
First step is to get all the info I want to know about all these stores. With over 70, I need a deep research to make this happen!
Personally I'm currently paying for Claude, but any of them should work
What's shopping without a little AI and a nicer UI?
Johor Premium Outlets has a bunch of nice stores, but frankly I don't know my brands as well as I could. How can I pick which shops to visit?
Hopefully this helps some other people as well 😄 There's so many great events, it's hard to find the right one!
The level/scope filters are great, and please take advantage of the speaker <> session links as well
Features:
- filter on pretty much anything
- find speakers you like and the sessions they're presenting
- bookmark your faves
- reject sessions to clean them from your view
- calendar view
- add to (google) calendar
Anyone else overwhelmed by session choice at the @aiDotEngineer WF?
Here's a power-user focused scheduling app: aiewf.shebecoding.com/
The level/scope filters are great, and please take advantage of the speaker <> session links as well
Hopefully this helps some other people as well 😄 There's so many great events, it's hard to find the right one!
Features:
- filter on pretty much anything
- find speakers you like and the sessions they're presenting
- bookmark your faves
- reject sessions to clean them from your view
- calendar view
- add to (google) calendar
Anyone else overwhelmed by session choice at the @aiDotEngineer WF?
Here's a power-user focused scheduling app: aiewf.shebecoding.com/
I always trust people more when they can tell you when not to use something. They’ve hit the curse of knowledge - used it enough to hit the edge cases and feel the drawbacks.
“I would avoid fine-tuning until you need it. If you can get away with prompting, you should use prompting.”
This is how I know @alanbigio is an RFT expert.
apparently the heisenbug observer effect works remotely. I have been trying to get some data passing to work for the last 4 hours. Then as soon as I tell someone about it? fixes itself.
The NaN issue is reasonably common when you give the model too much freedom, but I think I overcorrected for stability this time around.
My initial test DPO ended up with a lot of NaN weights, so kept crashing. Hence the "Model weights look healthy" check.
I have successfully run a DPO with trl! Did it change anything? no it did not.
But it 1) finished execution without crashing 2) created a working LoRA with no null weights 3) completed within 7 minutes
so overall I am happy
github.com/huggingface...
what's your comfort phrase?
America why are your plugs like this
the first one i tried sparked so hard it almost caught fire
oh fuck
it's time to make my laptop WORK for it's living
