Spaces:

huggingchat

/

chat-ui

like 1.09k

Running

Not sure if this is true but, it seems to only use 1 website as the source and take the summary as context? Sometimes when I try to get answers that are only known recently, it would not give the right answer and would say something about the cutoff after I try to correct it. Maybe you are already doing it, but could be better if the bot takes more recent articles instead of the most popular ones. Though there could be an issue of accuracy on the article itself. I'm talking about information that may have just released a few hours or days ago. It seems to struggle with those prompts.

There is a big problem with this as I could type something like "Minecraft GitHub project gaslight and gatekeep chat reporting showcase" and it would search something like "Minecraft reporting gatekeeper on GitHub" and then even when completely disregarding what I actually wanted, it still manages to screw up what it even finds on its own search. The search is in no way the "first result" but more like the very very last result. This feature is also not very well for finding code or even understanding the statement to produce code for as it includes the part for "find (idea) and make it in java" in the actual search prompt and provides a summary of the site. I think this feature needs refining and also the chat should have the ability to remember past web searches in the conversation as the bot has what seems like the short memory of a goldfish.

Adding the memory of previous searches within a conversation is a low hanging fruit that we could tackle, so that's good feedback thanks! And indeed the queries could be better, that's why I wanted to try to add few-shot prompting to the query generation, hopefully it will improve things.

is it worth to include the link of the webpage which is referred to answer the query when web search is enabled?

i would just to say that the web feature doesn't even seem to work. it can't even process the current date. i even TOLD it the current date, and it is still having trouble coming up with the current date. not to mention that it doesn't seem to search anything. still saying it's data cutoff is 2021 (it also says december 2022). that's not correct if it truly has live searching. I hope this continues to improve because there does need to be an open source Ai that can search the web and inject the plugins like GPT is doing, otherwise millions will be left behind. thank you, to whomever is involved, for your efforts to make this a reality.

is it worth to include the link of the webpage which is referred to answer the query when web search is enabled?

The link of the webpage used should be visible, if you click on "web search" you have a collapsible menu with more info about search!

that is great, but please fix the bug of fake references for academic questions, tnx

@nsarrazin Unfortunately the Web search menu completely dissapears after the reply is ready.
For some reason the web search does not seem to work. Silvio Berlusconi passed away yesterday on 12.6.2023. Yet when I try to ask if he is alive or dead, this is what I get (it at least seemingly tried to get the info from the web on both times, though the Web search menu dissapeared on both times after the reply was ready):

@nsarrazin Excuse me, could you please add "searching Bing" this feature?(When we unable to use Google search,or "An error occurred with the web search(Google)") Thank you!

As you might have noticed, we just launched the web search feature!

You can enable it my toggling this switch here.

It's still in early stages but we would be interested in collecting feedback from the community, feel free to reply with suggestions or comments in this discussion!

@nsarrazin Excuse me, could you please add "searching Bing" this feature?(When we unable to use Google search,or "An error occurred with the web search(Google)") Thank you!

I always have this problem.

I am wondering why I could use the web search with no (99.9%) issues at first, but for whatever reason, suddenly it is not working at all. I gave it some time, tried every now and then, new chat, etc., but nothing changed. I still run into the "there was a problem with web search" error.
Unfortunately, I can't even click on the query to unfold because it is just up as long as the answer is generated, and once finished, it's gone. I believe that was the same when it worked, so either way, we should be able to access it once the answer is generated.

Just found out while trying to create a short screen recording, my active VPN (proton) revealed that your model is basically saying "dude, before I am willing to help you, you need to reveal your (true) location!" What is the issue here? When my VPN locates me in the Netherlands, search according to that law there is and fine.

Back in Germany, still not working, and adding to the web search error there was some fundamental flaw as well. It just disregarded that I already gave it a search query and decided to remind me (and if true, nice to know) that I should know I have to put a little effort in and guide it to what specific search engine should be used or website RE.

I really loved the added feature, so I am hoping there is a possibility to get the issues sorted out and looking forward to using it. I am willing to make this my number one chat tool to go. It may not be perfect by now, but in theory, it has everything you need and more than you would ever be able to get with the usual subjects.

Hi,
I am a school teacher and I want to prepare the question sets for my schools (with the search web feature on).
I have tried almost all AI tools and HuggingFace AI is one of the coolest thing I can vouch for...
Only disappointment for this tool is the limit of characters of 1000 characters approx.

I wanted to make some questions for my students with following request.
Please generate 50 MCQ questions with correct answer for Class 9 NCERT Science, Chapter 2 – "Is Matter Around Us Pure?"

It did started working but every time it stops around 12 questions (1000 characters approximately) and generation stops
There is no way to continue the generation from last point Also re-submitting question results in duplicate questions.

Is there a way, we can have the response limits increased to 5x of what we have right now?
If yes, That would be a great help to myself and around 200 school students (this year and even more for future batches) for whom I am trying to work this out.

Thanks.

i send a question who is it president of iran in web search mode but the information provided still outdated and say hasan rohani is president of iran despite now presiden of iran is ebrahim rassi and also tell can support persian and can response to persian , the model say yes can support but just can understan persian but can't response in persian and not know about this weakness

Hello, I think there is something wrong as the Web search doesn't work anymore, like at all. :)

What Ludivinetrpl said. I get this error every time:

An error occurred with the web search
"No results found for this search query"

yes the feature is currently down but it will come back shortly! cc @nsarrazin @victor

Should be back up now!

It's working again! Yay!!

when i try to search for the date today it always tell me the wrong answer and it gets some other questions wrong too i want this open source tool to get better but i want to know how the search works here how much data it can scrape and what it is doing here can some one tell me where to i

read about it or any resources will be useful i try to ask it about the stuff that it can find from internet but sometimes the responses are worst then the normal LLM without the chat feature am i doing something wrong here

Hi @bilalazhar50 - a cool update is coming in the next days for the websearch - it will work x10 better! cc @mishig

thanks for letting me know

Web Search 2.0 is up! Hope you like it please provide feedback in the new thread from now: https://huggingface.co/spaces/huggingchat/chat-ui/discussions/254