You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Requested and reported by Paul Harland of BlueSky.
Newshound needs to be less "on scene" in its summaries. This has been a problem that I've actually been dealing with for quite some time in that when the actual news article implies that the reporter is on the scene or at the location, the summary also tends to imply that same information..
It's a problem that is going to be difficult to address because in the process of rewriting that part out of the original article, there is a high risk of introducing hallucinations to the summary itself. The principal role and rule set used governing the summaries right now dictate that they remain as factual as possible to the original while being able to rewrite it in a way that is unique and original in itself. Pluragism is a real issue that often plagues AI models extensively and this is a very difficult and fine line to walk between.
As much as I would like to say I have an easy fix to this problem, I really can't as it's basically been trial and error to get the AI model to reflect the humanity and empathy necessary, while still maintaining The required level of accuracy to avoid hallucinations.
Please be sure to add further situations of this so that they can be tracked and perhaps a more patternistic approach can be applied to correct it.
Thank you.
The text was updated successfully, but these errors were encountered:
Changes have been made to Newshound's summary instructions to hopefully improve the situation of the bot wanting to be "on scene" and it's a summaries. Careful inspection will be conducted to ensure that hallucinations do not take place and that the summaries remain true to the original article in substance and fact.
Please report any continued situations of the bot wanting to be "on scene".
Requested and reported by Paul Harland of BlueSky.
Newshound needs to be less "on scene" in its summaries. This has been a problem that I've actually been dealing with for quite some time in that when the actual news article implies that the reporter is on the scene or at the location, the summary also tends to imply that same information..
It's a problem that is going to be difficult to address because in the process of rewriting that part out of the original article, there is a high risk of introducing hallucinations to the summary itself. The principal role and rule set used governing the summaries right now dictate that they remain as factual as possible to the original while being able to rewrite it in a way that is unique and original in itself. Pluragism is a real issue that often plagues AI models extensively and this is a very difficult and fine line to walk between.
As much as I would like to say I have an easy fix to this problem, I really can't as it's basically been trial and error to get the AI model to reflect the humanity and empathy necessary, while still maintaining The required level of accuracy to avoid hallucinations.
Please be sure to add further situations of this so that they can be tracked and perhaps a more patternistic approach can be applied to correct it.
Thank you.
The text was updated successfully, but these errors were encountered: