What Would I Use LeoAI For?
One of the focuses of the INLEO team recently has been LeoAI, a ChatGPT-based LLM with both a generalized and a Hive-specific training.
I haven't listened to/read about anything related to the progress on LeoAI recently, but last I heard it was close to being released with at least parts of its functionalities working. I think the "For You" feed would be the first one taken over by LeoAI based on user activity on INLEO. It would also become a non-chronological feed.
I don't think chatting / querying for information is planned for the initial release, but last I heard Khal was using this capability of LeoAI for testing with encouraging results, but also with some share of hallucinations.
From what was discussed, LeoAI could be useful in many ways. The one I am curious about, if it will be capable of that, is its capability of extracting and presenting information on Hive, without having to write database queries yourself. This would have come handy a few days ago when I talked about account levels in terms of HP. And I wouldn't have used obsolete information.
There are two questions we have to ask ourselves first:
- how far do the LeoAI capabilities go regarding extracting required information from what has been published on the chain, account history, etc.?
- can we trust the information provided as being relatively accurate (without hallucinations)? Nothing to depend your life on them, of course. No AI has reached that level.
On the second question, we have to remind ourselves that sometimes the response is as good as the question. But hallucinations happen to LLMs regardless of the quality of the input.
Generated by Venice AI.
LLMs are also not meant to be... precise. If you ask them the same question at different times or multiple times, you will receive different answers, since the way they generate responses is based on tokens which are assigned weights, where a token can be anything from a syllable to a phrase (or whatever limit there is). And they choose to include them or not and how to continue the answer based on those weights.When the LLM chooses the top weight, it chooses the safest route. When it chooses a token with a lower weight, it takes a chance that it may go wild with the answer, but if simulating creativity is the goal and not precision, that's maybe a good compromise.
Hard to say how this would work if we want to use LeoAI to query for Hive-related information until we actually see it in action. It would be great for the regular users and for the ease of use, if the answers would be reliable, at least to a good degree.
Posted Using InLeo Alpha
This is cool
I’d make sure to ask the same questions different time and I will take note of the responses
You can do that we LLMs today. For example, the generative AIs that create images will create completely different images starting from the same input. ChatGPT will give different answers to the same question, but usually the most important part of the response doesn't vary "in essence". But that probably depends on the question too.
Congratulations @gadrian! You have completed the following achievement on the Hive blockchain And have been rewarded with New badge(s)
You can view your badges on your board and compare yourself to others in the Ranking
If you no longer want to receive notifications, reply to this comment with the word
STOPCheck out our last posts:
I think we need to see it to actually know any answers. AI can be good or bad. However, I wonder if it will be accepted here on hive. I know that there are quite a few people who don't consider what AI puts out as valid content. Either way, only time will tell.
I understand that, but when you use it as a tool to gather information, or to create an image for your otherwise text-based article, I think it's acceptable. The problem is if it will be reliable in that role.