this post was submitted on 12 Mar 2025
1509 points (99.0% liked)
Technology
66231 readers
5417 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each other!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, this includes using AI responses and summaries. To ask if your bot can be added please contact a mod.
- Check for duplicates before posting, duplicates may be removed
- Accounts 7 days and younger will have their posts automatically removed.
Approved Bots
founded 2 years ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
The fact that I can't trust the AI message to be remotely factual makes that sort of use case pointless to me. If I grep and sift through docs, I'll have better comprehension of what I'm trying to figure out. With AI slop, I just end up having to hunt for what it messed up, without any context, wasting my time and patience.
I started self hosting AI to learn more about it, and I have come to the conclusion that it really depends on the AI if its bad or not.
For instance, Google's AI results and just literal dog shit. It is just so factually bad its incredible even. Microsoft also sucks. And this is why everyone doesn't like AI. The two most common ways people see AI (Google search and Windows 11) is just complete horse shit. They should not roll them out, absolutely disastrous decision on their part all because they were feeling FOMO. For example, I asked microsoft's AI if the 'New' outlook had this feature from 'classic' outlook. It said it did. An hour later, I found it didn't actually have that feature. Fucking ridiculous that Microsoft's own AI doesn't know their own software. Embarrassing. Did they not give it their own documentation?
But 'dedicated' AI like ChatGPT and Deepseek I can trust to be factual with a 95% success rate. Current events is it's worse subject.
I really recommend watching this introduction by Andrej Karpathy https://www.youtube.com/watch?v=7xTGNNLPyMI
One part that really stuck with me is that the data in the model is more like a fading memory but the stuff in the context window is more like the working memory. Since I learned that I tend to put as much information as possible into the context window before asking questions about it. This improved the results drastically and reduced hallucinations.