I think a very high vaccination coverage rate (95% of population) is needed to prevent outbreaks.

The 88% confidence in this research is not high enough and it's for the entire population. Parents of young children are more distrustful than older people.

Link to parental views here ⬇️

Very nice of you to take the time to reply. Thanks

Is that another rival, very similar tree looming over the horizon?

Are you using these for political commentree?

Half my customers are unhappy with my product. So is the other half. I'm sure it's fine.

Absolutely. Pro tier only has context of 32000 tokens so it will soon start summarising the document you shared rather than working with the original.

NotebookLM might be an alternative. It's Gemini so maybe not as good but it keeps whole documents as context and has bigger chat window too

Dunno Doug, looks like your working relationship with ChatGPT is breaking down.

It seems to be one exchange away from saying, "OK I may hallucinate but at least I know what 'final' means!"

It's not as hard as it sounds. You or a colleague could probably vibe code their way through each of these approaches.

I'd be happy to talk you through how I would do but there many more better qualified people here and elsewhere who might do a better job.

Additionally, you could write a simple script to use a model to identify citations and statistics search online for them as an additional check to help exclude papers with fake or hallucinated evidence.

So the options I suggest trying might be better on entire papers instead of abstracts

But I think there might be another issue, which is that genuine human authors often struggle to write good abstracts and so are tempted by AI summarisation just for that part of their work.

Detecting AI in abstracts might find those people and unfairly lump them in with real cheaters

Simply asking chat won't work, but AI/ML might:

Two options:
- use a reasoning/code executing model to look for AI 'tells/clichés' in a text and award a score with commentary
- since you have an archive, train your own domain-specific model on AI vs human-made texts so it can spot the differences

Some of these have to be coded messages between spies

What about this guy?

This disparity between the kinds of messages received on phone apps compared with the ones seen in the living room partly explains the genuine (but appalling) strength of (dreadful, racist) emotion felt by Reform.

Generative AI will let advertisers change the ethnicity of people in their ads to personalise the identity, preferences and prejudices of individual people - audiences of one rather than one million.

We saw Brexit campaign messages targeted at specific ethnic groups. Now the casting can change too.

A white supremacist will see a huge number of targeted social communications that will show positive images of white people and negative images of black people (or just no black people at all).

I mentioned that TV ads are mass media. This is important. We are shifting away from the era of expensive short films that are seen by millions.

People's advertising diet is a mix of increasingly highly targeted online comms and then a few mass-market ads on TV.