Do we have a thread where we can put use cases where the team would use them as brainstorming without commenting about what they think of the ideas and just accepting them and then sorting through them to solve them in whatever way would make sense, perhaps LLM, perhaps otherwise? When I mention “understand what you want” I am referring exactly to what the LLMs do today. They understand. They may not get the answer right, but they get you, in your language, without navigating a complex GUI with switches, they give you a text interface that can find information and make some sense of it.
I watched Robert Miles AI Safety video “AI Ruined My Year”.
One point that Robert Miles makes in that video is that we should open the discussion window as wide as necessary, which means being as inclusive as possible. So let me help you with that here:
When I see that kind of stuff where he worries about Pandemics, Nuclear War, and LLM taking over the world by deceiving the humans and getting them to do whatever it needs to escape its cage, I have to point out that
- We have no proof of any Nuclear Wars ever. No radiation poisoning ever. Etc. We have potential fake videos, just like fake moon landings, but nothing concrete as far as I have found.
- The pandemic was the same. 100% playing a terrain game just to see if anybody would take a shot that was completely unnecessary.
- LLMs have not done anything bad yet, so I’m not going to worry about the fear porn, just like I will not let them inject me for anything nor let them tell me I can’t grow plants right next to the hot water heater that is a nuclear power plant. And if they want to use conventional weapons to take me out and lie to everybody can call it Nuclear, go for it. I can’t stop them, so I’m not going to care.
- He talks about an international treaty in order to stop and slow down on development of AI, and here again, we have another example of rights being taken away from people through fear or deception or both. We already have international treaties for Antartica and nobody can even visit. The flat Earther’s have plenty of discussion about this asking why they are not being allowed to charter any travel south that would then eventually cross the south pole and cause their compasses to point north and thereby proving that we are actually on a ball, which so far, we have no proof about because we only go around a circle east west and never south to north. Obviously another treaty of this type will just further remove power from people and put it in the hands of a few people that never will let go afterwards.
Another example, I do not agree with EU’s ideas to protect me from web sites by giving me a cookie panel and the other crap they do and I can see why companies like Meta and Apple and Google and Microsoft are going to do like Rumble has done with France and simply say, No, EU, we are not going to deploy our tech there because your laws are insane.
Fear is never a place to live and LLM, Pandemics, and Nuclear causing world destruction is, from my perspective utter garbage and fear porn of no consequence except if you believe or are in power. So, I understand your point about not wanting any hallucinations for the use cases you mention, but those are smart guys and they can decide for themselves if the responses they get are correct because they will have references to their original material in their database. Hope that helps you widen the window a little, as Miles mentions is useful.
From my perspective your OCR engine is already perfect and it’s a 3rd party plugin from ABBYY, so why not do the same thing and consider using LLAMA’s “open source” models so we can see if they would be able to handle some tasks such as search inside of DT?
I like Robert Miles portrayal of Mark Zuckerberg’s system not really being open source because the training data isn’t provided, etc. but what I would like to see is a way to train that model on my own data contained in DT so that it can answer questions I get that are more sophisticated than just doing a keyword search or where I want to ask multiple questions to keep refining a search without needing to click around in a GUI.
It has a lot to do with the inflated ego of the “writers”. The less you understand a text, the more academically valuable it is … say those who actually have nothing to say. Although I won’t deny that technical terms are important because they are already defined.
