Mobster @ Mobster @feddit.uk

Posts

1

Comments

23

Joined

1 wk. ago

It is nice! Showing my age but reminds me more like forums and chatrooms.

I don’t feel bad, but having moved a few days ago I’ll share my experience.

I try and come here first, will check the one or two things I’ve posted engaged with, scroll a bit - but realise I haven’t quite joined enough communities for there to be novel information each time I check in.

I then default to Reddit, and quickly go into my default auto-scroll passive lurk mode. I see something new - like the most recent Anonymous hack on twitter, and then come back to see if I can find it on Lemmy!

One key difference is I rarely posted on Reddit, but have felt very comfortable posting here. Not sure why!

Yes, I know what you mean and would encourage you to share your perspective more - it will still be unique, even if it is similar to others.

Thank you :)

Yes, I plan to eventually have a home server where I can run things like this, and also stop relying on cloud services etc.

LLM can hallucinate, but agree with you that RAG accuracy can be made reliable enough - by refining prompts, adjusting temperature, improving data structure etc. Tolerance for potential errors depends on the use case of course, but for something like this I wasn’t too worried. Also this is a very simple PoC use case, just using the chat box of a free LLM. Making your own RAG using this would improve accuracy significantly. I know you know this as you also develop RAG applications, but others less familiar may not.

100% agree. Nestle is evil.

I don’t have one sorry, this was from Reddit. I’ve also looked for a higher res version online and couldn’t find one.

Thanks for the advice

This is the original chart I saw on r/BoycottUSA.

True, there is always that risk.

Came over from Reddit as part of plan to divest from US-based products. Happy to be here, and plan to explore. Mainly lurked over on Reddit - hope to contribute more actively over here!

Yes Mistral. Not too sure about the brand other than they are French, so chose over OpenAI.

Ultimate aim is to host my own local small model on phone or home server, but not quite there yet. Could then use for this and other purposes.

Yes, the screenshot was the first time I used it. Will know to edit in future

Thank you

Thank you for that. I thought it would be blocked where I live, but seems to work.

Here is a link to the source document

https//files.catbox.moe/jtmcb5.pdf

yes you probably could just search ctrl+f. I'm quite lazy so imagined that if I could upload the document (46 pages PDF of companies and their brands), I could then just chat with the bot/or perhaps take a picture of a fridge/products and it could quickly give me the answer.

not sure how to upload PDFs, sorry.

Edit: turns out I can’t upload PDFs to lemmy.

If you wanted to make a list yourself it’s a bit of a hassle but not too hard, I just searched for the Wikipedia pages for each of the following companies, type in something like “List of brands for [company name]”. The companies I picked were:

1. General Mills
2. The Coca-Cola Company
3. Mondelez International
4. Mars
5. Procter & Gamble (P&G)
6. PepsiCo
7. Nestlé (not a US company, but have avoided for years)

Once you have the page you can just select the list directly from the page, copy, and then paste into a text doc or something. You could then search from it, and add to it as well. It may be a bit untidy, but search should still work.

Hope that helps, and sorry I couldn’t share the document directly and save you the hassle.

not deliberately, however when run it did automatically identify a brand (Hellman’s) owned by a multinational NOT in the list (Unilever). I hadn’t included Unilever in the source doc as it is British/Dutch. So it must have used its own knowledge base.

Perhaps it interpreted my preference as wishing to avoid multinationals? This is what it output:

“Based on the list you provided and the document, here are the brands you should avoid if you are following the list:

1. Doritos: Owned by PepsiCo.
2. Tropicana: Owned by PepsiCo.
3. Mentos: Owned by Perfetti Van Melle, not listed in your document, but worth noting.
4. Hellmann's: Owned by Unilever, not listed in your document, but worth noting.

The other brands you mentioned are not listed in the document as brands to avoid. However, it's always good to double-check with the latest information or packaging details, as brand ownership can change.”

May help if I clarify. The document is ingested into the chat, so acts as a local knowledge base. When I put some random brands into the llm it uses the info from the doc to respond.

I probably could manually search the document (ctrl-f), just find it easier to use the llm approach, especially if inputting in an unstructured/messy way.