Discourse Chatbot ๐Ÿค–

ืชื•ืกืฃ
1104

Thereโ€™s a PR open to add GPT-5 but thereโ€™s something going wrong during CI.

Iโ€™ve opened a Dev topic about it.

1105

Has been merged.

If you find GPT-5โ€™s reasoning too slow you can change the reasoning level. Thereโ€™s a new minimal level now.

Thanks to @NateDhaliwal for his assistance on this one!

2 ืœื™ื™ืงื™ื
1106

ื”ื‘ื•ื˜ ืฉืœื ื• ื ืชืงืœ ื‘ื–ืžืŸ ืงืฆื•ื‘ ืขื“ ืฉื”ื’ื“ืจื ื• ืืช ื”-reasoning ืœืžื™ื ื™ืžืœื™. ืชื•ื“ื”!

ืœื™ื™ืง 1
1107

ืœื•ืžืจ ืืช ื”ืืžืช, ืื ื™ ืžื•ืฆื ืืช GPT-5 ืื™ื˜ื™ ืžื“ื™ ื‘ืื•ืคืŸ ื›ืœืœื™ ื•ืœื ื‘ืจื•ืจ ืฉืฉื•ื•ื” ืืช ื–ืžืŸ ื”ื”ืžืชื ื” ื”ื ื•ืกืฃ ื‘ืชื’ื•ื‘ื•ืช.
ืื™ืš ืžืฆืืช ืื•ืชื• ืขื‘ื•ืจ ื‘ื•ื˜ ื”ืชืžื™ื›ื” ืฉืœืš?

1108

ื ื™ืกื™ืชื™ ืืช gpt-5 ื“ืจืš Chat GPT, ืฉื–ื” ื“ื‘ืจ ืฉื•ื ื” ืžืื•ื“ ืžืืฉืจ ื“ืจืš ื”-API, ื•ื”ื•ื ื“ื•ืจืฉ ื–ืžืŸ ื—ืฉื™ื‘ื” ืืจื•ืš ื›ื“ื™ ืœืชืช ืชืฉื•ื‘ื•ืช ืžืขื˜ ื˜ื•ื‘ื•ืช ื™ื•ืชืจ ืžืžื” ืฉ-4o ืื• o1 ื”ื™ื• ื ื•ืชื ื™ื. ื›ืฉืฆืจื™ืš ืœืขื ื•ืช ืžื”ืจ, ื”ื•ื ืœื ื˜ื•ื‘ ื™ื•ืชืจ ืž-4.1.

ืื ื™ ื“ื™ ื‘ื˜ื•ื— ืฉื”ืžืฆื‘ ื–ื”ื” ื‘ืขืจืš, ืื• ื’ืจื•ืข ื™ื•ืชืจ ื‘ื’ืœืœ ื—ื•ืกืจ ื‘ื›ืœื™ื ื•ื”ื ื—ื™ื•ืช, ื‘ืขืช ืฉื™ืžื•ืฉ ื‘-API. ืื‘ืœ ืื ื™ ืœื ื™ื•ื“ืข ื‘ื•ื•ื“ืื•ืช, ื›ื™ gpt-5 ืื™ื˜ื™ ืœื”ื—ืจื™ื“ ื•ื‘ืกื‘ื™ื‘ืช ืคื•ืจื•ื ื”ื•ื ื—ื™ื™ื‘ ืœืขื ื•ืช ื‘ืžื”ื™ืจื•ืช ื”ื‘ื–ืง.

ืœื™ื™ืง 1
1109

In terms of content performance, anecdotally, it seems like gpt-5 is giving noticeably better technical answers that gpt-4o. Iโ€™m not sure how to quantify that but it really impressed me.

Iโ€™m getting varying results in how long it takes to respond. It does seem, from experimenting this morning, like gpt-5 is slower on average but not by too much, and there were some cases where the response came faster with gpt-5. Iโ€™m measuring anywhere from 5 seconds to 35 seconds for a reply.

Weโ€™re using RAG and I canโ€™t tell what portion of the latency is from the RAG search vs the chat completion. It could be that sometimes it chooses not to RAG search, the search happens faster, or something is cached (in the search or the completion).

We would typically choose better answers over a faster response because giving customers bad technical advice is costly. Up to a point though, if it times out then thatโ€™s a very bad user experience.

GPT-5 recommends primarily gpt-5-mini for our use case, and escalate to gpt-5 in some circumstances. Sounds neat but complicated. Have you considered switching between models dynamically? Why doesnโ€™t OpenAI just do that automatically? ChatGPT - Compare GPT models performance

ืœื™ื™ืง 1
1111

ื ืืœืฆื ื• ืœื—ื–ื•ืจ ืœ-gpt-4o ืžื›ื™ื•ื•ืŸ ืฉื›ื ืจืื” ืฉ-gpt-5-mini ื—ื•ืฉื‘ ืฉื”ื•ื ื™ื›ื•ืœ ืœืขืฉื•ืช ื“ื‘ืจื™ื ืฉื”ื•ื ืœื ื™ื›ื•ืœ. ื”ื•ื ื”ืฆื™ืข ื‘ื‘ื™ื˜ื—ื•ืŸ ืœื”ื’ื“ื™ืจ ืขื‘ื•ืจ ืœืงื•ื— ืฉื™ืจื•ืช ื ื™ื˜ื•ืจ ืื–ืขืงื•ืช ื•ืœื—ื‘ืจ ืื•ืชื• ืœืฆื™ื•ื“ ื”ืื–ืขืงื” ื”ื‘ื™ืชื™ืช ืฉืœื•. ื”ื•ื ื‘ื™ืงืฉ ืžื”ื ืžืกืคืจื™ ื–ื™ื”ื•ื™ ืฉืœ ืฆื™ื•ื“ ื•ื”ื–ื™ื•ืช ื›ืื™ืœื• ื”ื™ื” ืงื•ื ืกื™ื™ืจื–โ€™ ืฉืžืกื“ืจ ื”ื›ืœ ืขื‘ื•ืจื. ื”ืืชืจ ืฉืœื ื• ื™ื›ื•ืœ ืœืขืฉื•ืช ื–ืืช, ืื‘ืœ ื”ืฆโ€™ืื˜ื‘ื•ื˜ ืœื. ื ืจืื” ืฉื”ื•ื ืœื ืžื›ื‘ื“ ืืช ื”ืžื—ืกื•ืžื™ื ื‘ืคืจื•ืžืคื˜ ื”ืžืขืจื›ืช ื›ืคื™ ืฉ-gpt-4o ืขืฉื”. ื ืฆื˜ืจืš ืœื”ื“ืง ืื•ืชื• ืœืคื ื™ ืฉื ื•ื›ืœ ืœืืคืฉืจ ืœืื ืฉื™ื ืœื”ืฉืชืžืฉ ื‘ื•.

ืขื“ื›ื•ืŸ: ืžืกืชื‘ืจ ืฉ-gpt-5 ื˜ื•ื‘ ื‘ื”ืจื‘ื” ื‘ืžืขืงื‘ ืื—ืจ ื”ื•ืจืื•ืช ื•ื›ื‘ื•ื“ ืœื›ืœืœื™ื ื‘ืคืจื•ืžืคื˜ ืžืืฉืจ gpt-5-mini. ืื ืืชื” ื”ื•ืœืš ืœืืคืฉืจ ืœื‘ื•ื˜ ืœื™ื™ืฆื’ ืืช ื”ืžื•ืชื’ ืฉืœืš, ืื ื™ ืžืžืœื™ืฅ ืขืœ gpt-5 ืœืžืจื•ืช ืฉื”ื•ื ืื™ื˜ื™ ื•ื™ืงืจ ืคื™ 5. ื™ืฉ ื™ื•ืชืจ ืžื“ื™ ืกื™ื›ื•ืŸ ืฉ-gpt-5-mini ื™ืฆื ืžืฉืœื™ื˜ื”.

ืœื™ื™ืง 1
1112

I have had really good luck with GTP-5-mini in agentic flows via tool calling, code writing and structured data. I generally find structured data is easier for AI apps than unstructured ! .. not what I expected ! but guardrails are easier .. (code-in-loop, human-in-loop, llm-as-judge, etc)

please watch this for blow by blow walkthru of high performance , low cost gpt-5-mini and gpt-4o โ€ฆ

If anyone out there is interested in working structured data capabilities into Discourse as a plugin, etc. Please reach out.

An NLP extension for sql/stats/datascience to Data Explorer is an example.. But could also possibly have a tool / plugin / feature that allows natural language queries of read-only sqlLite or duckdb etc olap files loaded into the container ? just a thought.. :thinking:

1114

Btw, I added GPT 5.1 to the plugin along with some fixes:

ืœื™ื™ืง 1