Sorry, last question on this: so if I want to deliver a better, more nuanced prompt to the LLM, I’m using “system.tagging” and not “prompt.tag”, correct?
nate c.:
Sorry, last question on this: so if I want to deliver a better, more nuanced prompt to the LLM, I’m using “system.tagging” and not “prompt.tag”, correct?
You are best to change prompt.tag
imho, but feel free to experiment.
Results with these prompts were so good from the get-go (with GPT 4) I’ve not really focussed or experimented much.
system.tagging
might help you modify the “style” or “character” of the response (e.g. humorous or theatrical)
1 Like
I’ve been meaning to add this for a while, but the results from the Completions-based auto-tagging system were just too good to make it a priority.
But this morning I added auto-tagging based on Embeddings classification as an option:
merefield:master
← merefield:add_tag_embeddings_for_auto_tagging
opened 02:25PM - 05 Jun 24 UTC
Part I - groundwork. Additional PR is required to make this fully functional ..… . which will follow ...
* the current auto-tagging uses the Completions API. It was left like this for a while because it is surprisingly effective..
* here I introduce an alternative auto-tagging strategy using vector similarity which is a more standard approach.
merefield:master
← merefield:add_tag_embeddings_for_auto_tagging
opened 02:37PM - 05 Jun 24 UTC
* adds the additional logic required to make use of the embeddings based tag det… ermination
There are various new settings, have a good look
Enjoy
Heads up, about 5 hours ago my site became inoperable due to a 500 error again. I had made no changes, using GPT 4. Could be that @RGJ pushed an update (but unlikely, since I didn’t see a note). Turned off the plugin, and the error resolved.
I’m not seeing any issues at present.
Could you please share the corresponding logged error and confirm the version of the plugin you have installed?
Jagster
(Jakke Lehtonen)
June 15, 2024, 6:19am
108
Only AI-related 500 error what I get, started yesterday, is (or was, I have not examined further yet) automatic alt. Why I didn’t start digging it deeper? Because error 500 comes really often from OpenAI.
RGJ
(Richard - Communiteq)
June 15, 2024, 7:37am
109
It seems like the plugin on your site regressed to an older version somehow.
@ncaming215 My apologies, this is most likely our fault. I’ve put it back on 0.2.2 now and going to find out what went wrong. For now @merefield no need to worry.
2 Likes
Delighted to pass on this announcement from Open AI and confirm that AI Topic Summary options have been updated:
We just officially launched GPT-4o mini—our new affordable and intelligent small model that’s significantly smarter, cheaper, and just as fast as GPT-3.5 Turbo. Some details on the new model: Intelligence: GPT-4o mini outperforms GPT-3.5 Turbo...
Reading time: 4 mins 🕑
Likes: 79 ❤
https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/
merefield:main
← merefield:add_gpt_4_mini_option
opened 06:36PM - 18 Jul 24 UTC
See: https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/
This is a big deal for AI Topic Summary, because this new model has:
128k context (so can summarise lots of Posts!)
is 60% cheaper than the cost of GPT 3.5 Turbo!
Enjoy!
1 Like
Whoooops … there was some missing code I had forgotten to check in.
This is now resolved and I’ve improved logging at the same time:
merefield:master
← merefield:gpt_4o_mini_fix
opened 12:51PM - 21 Jul 24 UTC
thanks to @Jagster for raising the alarm.