ChatUWM Pilot Feedback by Michael Riley

ChatUWM recognizes UWM as the University of Wisconsin-Milwaukee, instead of United Wholesale Mortgage. I typed in simple questions with UWM in them. The answers came back talking about the university. As a company-specific tool, I think it should recognize the abbreviation.

yanwanqiu — 2023-08-08 19:06:18 UTC

Edit Button

lpawlosky2 — 2023-08-08 19:56:07 UTC

Is it possible to make the edit button visible without needing to click on the prompt? That way, users will know they can edit their prompt without needing to start over or add extra details below?

sremsing1 — 2023-08-09 15:52:06 UTC

Start a new chat every time you change topics. It reads each chat like a conversation and will use past prompts and answers to generate new answers

You have to work with it. You can refine your results as needed

mleskoviansky2 — 2023-08-09 16:36:32 UTC

I know we say on here that it may not be entirely accurate, but we just need to make sure that is SUPER clear. I can see a scenario in which someone like an Underwriter/AE uses it for guidelines and does not have the full picture. So we should still verify!

mleskoviansky2 — 2023-08-09 16:38:48 UTC

UWM-Specific Plugins

mtaylor3101 — 2023-08-09 18:22:15 UTC

In a manner similar to ChatGPT Plus, certain users could have access to role specific plugins, assuming the infrastructure could be connected, articles and data from KB, Bitbucket, and much more could enhance certain user experiences.

Reword your prompt if ChatUWM doesn't provide an accurate response.

sfleming217 — 2023-08-09 19:04:12 UTC

It took me a few tries to get ChatUWM to get the type of response I was looking for. Sometimes it helps to be more specific.

Consider adding a pop-up that explains what the icons mean

sfleming217 — 2023-08-09 19:26:52 UTC

It would be nice if there was a feature that shows what each icon is for if you hover over it. There's a hashtag icon and I wasnt sure what it did so I tried hovering my mouse over it to see if that helped but nothing came up. For example, I mean something like this.

Add markdown capabilities

mtaylor3101 — 2023-08-11 13:28:17 UTC

It would be nice if ChatUWM was able to format tables and text in fancy ways

A COPY button, so we can take the information and paste (work with) it without highlighting everything and hitting CTRL+C.

unitedshore1 — 2023-08-13 22:29:00 UTC

gear selector for docs should be automatically on?

jmantua1 — 2023-08-21 18:31:51 UTC

Shaking response

ssherman11 — 2023-08-28 14:39:24 UTC

If the response from the chat is about the height of my screen, it shakes up and down making it difficult/impossible to read. Scrolling down fixes this, but then you can't see the top part of the response.

Clear All

agarrett28 — 2023-08-28 14:43:29 UTC

Could be a useful enhancement to allow you to clear all chats on your side bar using the options cog on the bottom left

Talking about markdown for no reason

wreynolds20 — 2023-08-28 14:52:18 UTC

Location of the chat box

2023-08-28 14:53:25 UTC

It's in a pretty confusing location. Where the "New Chat" button is, I as a user expect the chat box to appear right below it, not below all the tiles. I almost missed that it appeared right there and was kind of surprised that it did since you have all this real estate right underneath the button.

Edit button is not working to edit the message

2023-08-28 15:24:36 UTC

API access

wreynolds20 — 2023-08-28 15:25:06 UTC

Allow developers to integrate ChatUWM into their applications via REST.

Reference of any message is gpt-4

2023-08-28 15:40:09 UTC

is it expected=> semicolon and arrow in the example

2023-08-28 15:50:10 UTC

Capabilities and limitations are not working

2023-08-28 15:51:42 UTC

Second comment break the comment with 502 error

2023-08-28 15:58:46 UTC

Links to external information

ahawari2 — 2023-08-28 16:14:52 UTC

It would be nice to have a reference or link to where ChatUWM is getting its information form so I can use that reference or link when documenting my findings.

scrolling bug

2023-08-28 16:27:38 UTC

When the chat is scrolled to the bottom, it starts to glitch out and bounce the chat window (see attached video)

Knowledge limitation

2023-08-28 17:59:43 UTC

Will there be an update either monthly, quarterly, or annually regarding world knowledge for use with this?

I like that it will explain the why behind it's thought process like when writing a validation rule

2023-08-28 20:14:43 UTC

Saving Custom Prompt

agarrett28 — 2023-08-29 12:02:07 UTC

Allowing the user to have a saved custom prompt could help with responses tailored for a specific persons need

Suddenly Stops Working

ssherman11 — 2023-08-29 14:55:28 UTC

I was feeding it multi-part documents so that it had an entire document. I've done this several times, but sometimes it just stops responding 4/7 of the way or w/e.

Auto-Chunk Documents

ssherman11 — 2023-08-29 14:56:57 UTC

There's a 5k limit on characters, so chunking documents requires manually entering them, and of course it's helpful to "1/5 pieces" tell the AI where the process is at. Could we have some way to auto-chunk documents? That way I could just copy/paste 90k characters and have the UI submit them piece by piece with appropriate prompts so that I don't have to manually do it.

Chunk Prompts

ssherman11 — 2023-08-29 14:58:31 UTC

If you're giving it large documents, chunk them up. It helps to tell the AI "1/5 pages/pieces" so it knows what you are doing.

Allow it more access

jsillman4 — 2023-08-29 20:03:26 UTC

I'm hoping in the future we can give the AI more to work with by linking KB, Jira, and/or Bitbucket so it could answer some of the easier fundamental questions for new hires such as 'what are questionsets', or 'how do I implement commandbus to execute a query'. I understand the omission of this info was intentional, but also think it may be a missed opportunity.

Hints are too fast

agarrett28 — 2023-08-29 20:10:57 UTC

The hints should last a bit longer to allow the user to read them before they swap out

Allow selection of different models

wreynolds20 — 2023-08-30 18:25:50 UTC

GPT4 is way more expensive than GPT3-based models. For many tasks, GPT3 would suffice - so it may make sense to allow users to select which model they want to use, assuming that is possible.

Embeddings

wreynolds20 — 2023-08-30 18:28:26 UTC

Embeddings are going to be critical when building domain-specific applications using ChatUWM. When building the API abstraction layer, it'll be important account for this. More info, if it wasn't blocked: https://platform.openai.com/docs/guides/embeddings

Chrome vs. Edge

2023-08-31 13:35:45 UTC

When using Chrome, after entering a prompt/question the ui can sometimes start shaking violently due to the chat boxes changing sizes. This does not appear to happen when using Edge instead.

RESPONSE: Tab + Enter

mriley65 — 2023-09-01 15:17:33 UTC

Due to the limitations of PowerApps, Enter is not available for ChatUWM. Currently the best we can provide is Tab + Enter.

RESPONSE: Plain ChatUWM may provide inaccurate results. Enabling DocSearch will cite the source of any information it provides.

mriley65 — 2023-09-01 15:43:57 UTC

Be aware that leading prompts produce leading results that may not be accurate. Information on UWM is limited at this time.

RESPONSE: Great ideas. Keep them coming.

mriley65 — 2023-09-01 16:08:27 UTC

The ChatUWM team is rolling out features or fixes almost daily, so be sure to refresh.

RESPONSE: ChatUWM is still in Beta.

mriley65 — 2023-09-01 19:50:38 UTC

Keep the questions coming. We will anwser as many as we can. As we gather feedback, decisions on the tool are being made almost daily. Stay tuned for more features and functionality.

I've seen a few typos, not in responses provided but in the verbiage that shows while waiting for a response to load. Might be good to audit those for accuracy

ajohnson1220 — 2023-09-01 19:59:18 UTC

RESPONSE: Be sure to verify information pulled from the plain ChatUWM responses.

mriley65 — 2023-09-01 20:00:19 UTC

This tool will confidently provide incorrect answers occasionally if led that way. Be sure to double check information provided for accuracy.

Chat History

rbateman3 — 2023-09-01 20:46:43 UTC

When versions 0.1 and 0.2 were active, I created a set of chats and responses. I just upgraded to 0.4 today and noticed the history of those chats and responses is missing. Do we have the ability to retrieve previous chat history? Or is all older history gone?

Token Limit

ecstillman — 2023-09-02 00:28:23 UTC

This model's maximum context length is 8192 tokens. However, your messages resulted in 8612 tokens. Please reduce the length of the messages.

^^ Is that the sum total of the chat?

I experimented with 'What is 2+2?' and received the same warning, which leads me to believe it is not the tokens needed for the next response - instead some combination of all previous.

Thx

Help with coding

2023-09-02 16:42:20 UTC

This tool is quite usefull at generating very granular unit tests, which is a lot of typing saved.
It is pretty good at explaining Regex expressions in plain english

Vector database

2023-09-02 17:39:22 UTC

Im sure this is in sight! A vector db will be awesome to provide a more targeted context and circumvent the context window.

Prompt Templates

2023-09-02 17:41:01 UTC

Providing a way to save and share Prompt Templates would be very helpful!

Search all chat history

ecstillman — 2023-09-04 18:41:23 UTC

has this been talked about before? thx

pdfs within document

rhartford5 — 2023-09-05 16:32:20 UTC

I was in a ChatUWM response; multiple pdf documents were referenced in the body copy. However, when I accessed the pdfs, they were all the same document, even though they were numbered [doc1], [doc3] and [doc2]. They also were not in sequential order, which they should be.

Also, is there any way that when you click on the referenced pdf, it will take you to the copy that only pertains to that reference instead of the entire document, where you then have to sift through the entire document to find the referencing copy?

ChatUWM says it has knowledge up to September 2021, but it is aware that Betty White died on December 31st 2021.

ksands8 — 2023-09-05 18:45:53 UTC

ChatUWM should have access to the current Date/Time

2023-09-06 14:20:35 UTC

Tried using ChatUWM to track work that I have done over the week on a day to day basis, however it is not able to track days/dates without me explicitly telling it what day I did what. Would be nice if it had access to the time and date.

ChatUWM Google Chrome Extension

2023-09-07 12:06:01 UTC

Giving the functionality to be on any web page with still having access to ChatUWM on a side bar.

2023-09-07 13:46:48 UTC

I asked ChatUWM to make a written prompt longer and received: "Received invalid response from ChatUWM, please retry or start a new chat."

ChatUWM DocSearch indicator

2023-09-07 21:32:13 UTC

Posted by Geof Griest
Would be nice to have an indicator on the chats that used the ChatUWM docsearch.
Also would be nice to have a timestamp/indicator on when that button was clicked to see when the results used different retrieved data.

Chat title 100 characters max size warning.

2023-09-07 21:50:18 UTC

Posted by Geof Griest
Appears the size limit for a chat label/title is 100 characters but allows for typing well past that limit.
Would be nice to have a warning of 100 characters max size instead of seeing it being concatenated after the fact.

Expand Response Timeout

2023-09-08 15:13:21 UTC

I am giving GPT a relatively straightforward task of reading .editorconfig rules and providing a human-readable translation of those rules with correct and incorrect examples.

However, if I feed it more than seven at a time, because it takes computational time, it never completes and I get a JSON error that 'timeout' is not a valid object.

Current token count, count remaining

2023-09-11 12:03:25 UTC

It would be very useful to know how many tokens have been used in the current chat and how many tokens for what has been typed in the current window. Whewn working with a large script Im sometimes getting a single response and several times no responses because the script is too long.

Gear screen

jdevolder1 — 2023-09-11 17:48:31 UTC

This is what I see when I click the gear icon in the lower left. Seems some elements are out of place?

Info screen

jdevolder1 — 2023-09-11 20:51:25 UTC

A couple minor typos to fix on the pop screen screen when you click on the info icon in the lower right

There is an "s" missing in the word session when the "your session has timed out message appears."

lpawlosky2 — 2023-09-12 21:25:08 UTC

Add a Non-Dark mode version.

2023-09-15 19:13:11 UTC

Posted by Geof Griest
Is there a way to make it be in a Non-Dark Mode ?
White background with text in black ?

Dark mode is hard for me to look at.

Potential use case: EROC Summaries

2023-09-18 12:49:38 UTC

I provided eroc status updates given from the incident managers to SSG leadership during a long-running incident and asked ChatUWM to write a summary.
From the 404 word input, a 141 word (accurate) summary was generated.
I then provided the same summary and asked ChatUWM to write it more concisely and it generated an accurate 61 word summary as well.
While learning from our incident does require digging into the "messy details," these summaries would provide greater context to help distinguish between different "Oops error" incidents and their individual impact, and to help remind folks of the original incident being the "why" behind some of our long-term solutions as those solutions are eventually implemented.

Retry button

agarrett28 — 2023-09-18 13:27:18 UTC

When the response fails, there should be a retry button so you don't have to edit and add an extra character to resubmit

Increase character limit from 5000

wreynolds20 — 2023-09-20 20:08:24 UTC

5000 characters comes out to about ~1200 tokens. GPT4 contexts support above 8000 tokens. Can we increase the character limit?

Opening chat history inv0.4

ecstillman — 2023-09-22 15:27:45 UTC

Has this come up before ? I cannot open previous chats.

Marking question-answer pairs as processed

2023-09-27 17:41:56 UTC

My problem:
Sometimes I ask a simple question or two and that's all. Other times I ask a ton of questions while exploring a broad new topic and its subtopics and concepts that ends up splintering out into tons of related sub-questions and I am scrolling back up to finish reading earlier answers that I only got halfway through before asking more questions. It becomes a necessity to re-read things I already read to remember that I already processed it or didn't or where I left off. Some way to deal with this would be amazing.

My suggestion:
A way to mark a question-answer pair as having been processed. This should change the color or something and automatically collapse the question-answer pair which could be re-expanded but the color representing "Processed" will remain the same unless you uncheck/re-click the processed button. This would be AWESOME for my workflow when exploring new topics where I ask tons of questions going down various rabbit holes and need to keep track of what I have mentally processed and what I still haven't completely and want to go back and continue to read. Maybe my workflow is just a mess, but I like using it this way and asking follow-up questions before I forget, a way to essentially mark a question-answer pair as done would immensely improve this for me. Thank you!

Delete Chat History Option

2023-11-22 15:30:41 UTC

Can we add an option that offers the ability to delete full chat history? Going one by one is time consuming.

Chat UWM further implementation

2023-11-22 15:52:30 UTC

Is there a plan to bring chat uwm into our products? Example: a built-in chatbot that uses chat uwm/gpt for underwriters on dochub. With the sole purpose of training dochub best practices

hello! when the prompt window begins to expand in length, it will block the previous response and the slide bar is unable to bring it back into view. I can provide screenshots.

2023-12-01 20:14:59 UTC