this post was submitted on 03 Jun 2024
1471 points (97.9% liked)
People Twitter
5234 readers
640 users here now
People tweeting stuff. We allow tweets from anyone.
RULES:
- Mark NSFW content.
- No doxxing people.
- Must be a tweet or similar
- No bullying or international politcs
- Be excellent to each other.
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
I was working on this with a friend over 10 years ago but the only grocery store that made a decent effort at organizing their website to be scrapeable was Loblaws and all the others had APIs that cost $100,000
Flipp allows for some of this desired capability now through digital flyer scraping and online feeds, APIs. Maybe things have gotten better on the API side over time.
Pretty sure it’s a Canadian app, coincidentally.
Which is one area ML models might (with the right investment) actually be useful. A model trained to look at web pages and relay information from the content visually like we do would be very powerful. The newer ChatGPT models have visual capabilities, I wonder if you could give it a website screen capture and ask it for prices.
Why would you want a model trained on outdated prices? This is not really something LLMs are particularly suited for.
Maybe to crunch historical data, but not for daily comparisons.
Why would the model be trained on outdated prices? I'm not talking about LLMs, but separate model designed to parse visual information - specifically websites - and extract particular elements like prices. My comment about ChataGPT was in reference to the newer models which can relay visual information, I'm not suggesting that would be the right approach for training a new model.
The applications would be broader than just prices - this would allow you to scrape any human-readable website without needing to do bespoke development.
I am not sure, that would work. You could train a model that analyzes data and then feed it the data you want to transform. The data wouldn't be the training data then but part of your request.
Like you can feed a book into GPT4/5 and then ask questions about it.
For what you describe you wouldn't really need AI just a more or less fuzzy parser (like the scan a receipt, get the prices ocr things). Unless I didn't get it.