Some of these will be more controversial than others, i’m well aware. With an open potential for abuse. I don’t have an easy answer for that other than perhaps rigorous vetting of who recieves an access key perhaps. But if these models are to become a more broadly commercially viable thing they need to become a hell of a lot more real world useful. My own limited technical expertise & limited time allotment & ADHD & advanced age mean I probably will never be able to develop them on my own. So I do ask if any of these provide any of you with a fertile inspiration to cut me in on the deal somehow. Either helping during the dev process & access to the final product indefinitely. Or heck, kick me back some cash appreciation if it makes you a mint. I’ve got a lot of other ideas just waiting on capitalization to move on. You hook me up, i’ll feed you back some of my more commercially viable ideas. This is but a small sampling to see who bites.
-
a reliability confidence indexer. A model that can look at someone’s complete public digital footprint to look for signs of disingenuous behavior. Something that could out PSYOP cutouts to con-men to fraudsters. Check public conviction records, old reportage, BBB reports, negative mention, et al. Feed it a name, it makes a couple of refining requests as necessary and you get a reliability confidence index back. dangerous/ambiguous/poor/good/great/excellent. Extra credit if it could determine the likelihood whether a public presented personage is actually a real person or a constructed identity or even a weaponized LLM masquerading as a genuine person. To limit the potential for hallucinated output the software agent should, upon command, be capable of outputting discrete examples justifying it’s output & conclusions.
-
Automated agent that combs the internet forums & such for identifying, language matching & publicly tagging disinformation campaign ‘superspreaders’. We know there is a relatively small percentage of dedicated disinformation superspreaders. By NLP word usage stylometry we should be able to find them no matter what account identity they’re hiding under within > 500 words with a medium confidence and >2000 words with high confidence. The agent should then be capable of posting “This post is likely from identified superspreader (real name if known/assigned name if not and/or designation). Anything stated should be taken with the highest degree of skepticism”. By making it impossible for these individuals to remain totally anonymous we can disincentivize, discourage, AND deter people from getting ensnared in their web of lies.
-
an “inventors buddy”. A LLM that is capable of identifying whether a given proposition has much market potential or impact. Capable of helping someone (like ME) take something from a napkin sketch to a roughly finished product. Capable of designing PLC boards, making material & design recommendations, etc. As much engineering assistance as can be reasonably packed into a language model. Something capable of outputting from language prompt to AUTOCAD output, further capable of iterization & refinement. Ideally something that could be run as a standalone software suite so as to limit the fear or potential of IP theft.
These are just a few ideas. I literally have dozens more. Lets see who shows any interest.