Welcome to ScapeCrunch

We are ScapeCrunch, the place where planted aquarium hobbyists come to build relationships and support each other. When you're tired of doom scrolling, you've found your home here.

Can you trust AI to give you the right answer?

AI is wrong a lot, and it is sometimes alarming to me how many DO trust responses to questions asked of AI, especially via a web browser. When I am feeling a bit feisty, I actually enjoy calling out AI when it gets things wrong, that it espouses as fact. At least for now, I don't feel like our robot overlords are putting my job in jeopardy. Ask me again in six months, though, as things are moving pretty quick on that front :cautious:
 
On general facts, you obviously wouldn't rely on that unless verified. You need to treat it like any tool, its good for certain circumstances, not so much for others. Example, for tech problem solving, its not bad. If you need to run a certain linux command to work out why your pi-hole dns server is behaving oddly, its OK, particularly Claude. That's the kind of stuff I mainly use it for. 20 years ago I used my brain a little more to work similar issues out but why not use the tools at your disposal.

Building the aquarium controller it has been useful. To consolidate everything into the one dashboard and combine influxdb to hold all your pH data, grafana to run the charts and home assistant to actually view it within the overall Aquarium dashboard, I'll be honest, it would have taken a whole lot longer without an AI. Having said that, i often call it out for giving me the wrong directions. So all i would say is use with caution and have an idea of the subject you are using it for so you can identify when its leading you down the wrong path.
 
I work a ton with AI professionally and have found it to be truly incredible. BUT here's the thing, the term itself is far to broad to make any sense these days.

What most people mean are the free chat bots like claude or chatgpt. These are, by definition, general purpose models that try to be "OK" at just about any question a human could have. They are not designed to be specialised. Yes some of that has to do with the "garbage in garbage out" mantra. But it also has to do with training, priors and weights. That's a looong topic.
But if that's what's meant by AI, then I would agree. It is NOT suited for the application of technical knowledge to an individual context.

But just saying it's a bit harsh to judge a lowest common denominator model against best in class technical knowledge (as I would wager this forum represents for our hobby).

I personally have been working on, and will soon be asking you guys for help actually, on a dedicated technically trained ai for the aquarium hobby. It's a modest start but so far the results are pretty good.

The solve, and this works for any topic, is to build a solid set of data, then work hard and iterate through the cognition logic and set up agents and appropriate instructions, routing&progressive disclosure, then play around with embeddings and vector weights etc....then layering the info retrieval..how much is keyword search, semantic, vector, and when or if inference is used at each step..

Anyways getting too technical. I guess the main take away is this: the main reason companies are not getting value out of ai is because they expect utter magic from what is basically a very complex auto-complete program. But if you know how to leverage it's strengths and put in the appropriate knowledge work and iteration and lots time...it can be truly game changing I believe. At least that's been my experience professionally.
Its a marketing job that has overpromised what it can do for a given amount of effort. Reality IS garbage in garbage out, but applied to the effort needed to make it work for you.
 
But just saying it's a bit harsh to judge a lowest common denominator model against best in class technical knowledge (as I would wager this forum represents for our hobby).
Well, I will allow how there might be some merit to that.

The only AI I have interacted with is online free chatbots,…

And

AI Autoattendents foisted on me by large corporations I really would prefere not to deal with…

Let me just say, I have yet to interact with what leaves me to think it is a “best in class technical knowledge”. AI Autoattendent at any companies customer service.

The AI autoattendants I have been forced to deal with just suck the will to keep on living out of you…

I told one, that the only thing I could think was positive about my interaction with it is that it made the concept of death look not so bad in comparison to having to deal with it to resolve a problem…

I swear I heard “Sarcasm filter activated “ in the background…
 
Well, I will allow how there might be some merit to that.

The only AI I have interacted with is online free chatbots,…

And

AI Autoattendents foisted on me by large corporations I really would prefere not to deal with…

Let me just say, I have yet to interact with what leaves me to think it is a “best in class technical knowledge”. AI Autoattendent at any companies customer service.

The AI autoattendants I have been forced to deal with just suck the will to keep on living out of you…

I told one, that the only thing I could think was positive about my interaction with it is that it made the concept of death look not so bad in comparison to having to deal with it to resolve a problem…

I swear I heard “Sarcasm filter activated “ in the background…
I hear you on customer service. I think they suck too. Ai has simply been over promised. To companies and consumers. To get the most out of it I remains for now pretty technical. Enough to be out of the reach of most is the reality. Add to that that even with that expertise ai only really shines in a few areas at the moment: coding, biology, chemistry, cyber security, and maths - at the academic/research level. Physics is meh so far.
 
But just saying it's a bit harsh to judge a lowest common denominator model against best in class technical knowledge



1000053010.webp

PBS News host: “the State Department says it takes full responsibility for hastily altering and then presenting a map of Africa at a global AIDS conference that mislabeled every single country it highlighted.
Attendees shared photos of the incorrect map online. Coastal countries like Nigeria, Mozambique and Cote D’Ivoire were all landlocked and in the wrong locations. Other countries’ borders were misrepresented..
An analysis by Reuters found a watermark in the map’s image that suggests it was made with OpenAI..”
Offered without comment 🤦🤦

🤣
 

Top 10 Trending Threads

Back
Top