new interaction type: "Fact checker during a 'political' conversation" or "Debate assistant"
complete
M
Mike O
these days, I'm doing my best to "get out the vote" by contacting folks in my circles to see where they are at, and especially if they are undecided, to talk to them about their concerns.
with all the misinformation out there, it could be really helpful for a Hedy session type to be capturing statements made by folks in the conversation, and to fact check them, providing clear statements with references to accurately correct mistruths with debunking.
helpful:
- provide reference to where this misinformation comes from, and who is using it recently
- label(s) of what type of mistruth this is that was made
- explanation of why this is not true, give feedback about the statement
- give a reference to an authoritative source of truth about the topic where parties can go to be better informed
M
Mike O
heya Julian,
CONGRATS on this use case. AWESOME! I can't wait to test and use it in my flow!
with the arrival of Claude OPUS last week, I finally went and "ante up" my AI subscription and paid for Perplexity.ai. It's already changing my life, and has saved me about 100 hours of work.
- live internet access, plus
- my choice of LLMs, plus
- add my content to a "space" that becomes additional context for the convo.
(with Anthropic's launch of "computer assist", also last week, the automated smoke-test-apps of old take on new life with image recognition and processing. AGI seems weeks away!)
i hope you guys don't give up on this one. LIVE content is king!
Julian
complete
Available in v1.7.0.
Julian
We added fact checking to the new journalism session type. While it cannot search realtime information, it does work pretty well for generally available information that's been known in April 2024.
I know that's not the ideal end state for this request, but hopefully a step in the right direction!
I'm going to go ahead and close this feedback item and we'll continue to separately track the additional ideas to take the fact checking functionality to the next level.
Julian
planned
Julian
Yes! We've been kicking this idea around too. In the long-term we're thinking of adding a session type for journalists that would have this embedded, but in the short-term perhaps we add a "fact check this statement" quick prompt.
One challenge is that the LLM we use (Anthropic) has a training data cut-off that doesn't allow for the most recent misinformation to be included (FEMA sending money to Ukraine, people eating pets, etc). We'd probably need to integrate with a different API, such as Perplexity, that will check news sites in realtime and rely less on its training data (example attached).