Where should the AI CoE bot jump in?

Morning! Are we still moving to the OpenAI Enterprise edition now that the dust has settled with Sam Altman? Asking because we are using the API's and see some some charges, wondering if the API calls are more cost effective on Enterprise.

We don't know how sub-agents work in the case of Claude code (whether we can override their system prompts).

I have to jump in an IM meeting, but I would love to hear thoughts about running loops in Agentic AI. I've been struggling to make it work in Cu Chulainn (it always drifts, and the Andon Corder finds a reason to shut it down after a few iterations.)

Hi Team, do we have some sort of documentation / guiding principles how to evaluate RAGs? like which where to get certain LLM keys / where to run the evaluation in infra level and how to minimize cost in the context of XO

for the LLM Keys, I'll reword that as API keys, for example we have a RAG pipeline and at the moment we're using open AI, ideally we would want to compare it with let's say using claude, grok etc

Very good point; fact retrieval is a very specific Gen AI task/capability out of many others, like text summarisation, translation, sentiment analysis etc (Chat GPT gives me a list of 18 capability categories). Do we have insights on what platform is best for each of these? That table will change with every new version, but could be very useful when looking for applications and selecting tools to use?

Cursor be heavy on my M4 Pro

Which GPT is best for OCR?

Hello Team, Any recommendation for prompt. tracing tool for agents built using Langgraph ? My best find is : Self hosted LangFuse.

Product request - More shortcuts like superhuman has

Should be 1m lines of code *deleted* with ai

Hmm, has me thinking, we should do a 5k rewite with AI. I can volunteer Aurea platform.

When I finished conversational prompting GPT lied to me and said it could not generate images 😱

Just off the back of 2 RFP completions with a couple of thousand questions, an Enterprise Custom GPT RFP Responder trained on our product docs and existing responses will be an epic timesaver.

It should reduce the time to complete a large RFP by at least 50%, maybe nearer to 80% when well trained.

I just shared a new folder with you that contains our 2degrees OCS RFP response and Solution Description. I also added our Information Security Policy and Incident Management Manual. The information in those documents should cover at least 50% of the questions we are typically asked for in a medium size RFP. Also a mix of PDFs, .docx and .xlsx. Thanks for helping and very interested to see how this turns out. We will need to include the definition of compliance in the training. And nuancing the 'compliance grade' is quite a human skill - until now, maybe !

Here's some demo videos for responsive (fka rfpio). They are using the OpenAI APIs. While we can create a GPT with all our knowledge and get back answers to questions we type in, this product also provides workflow for identifying questions in docs, prefilling responses and assigning answering responsibility for questions or sections to different people. Now that we're doing 4 RFPs this month, this type of automation would be valuable. https://play.goconsensus.com/sac9a97ad Also the goconsensus tool is interesting, where I'm sure they are seeing metrics about who is watching the content and have probably customized what they sent out in response to the qualification call I did.