Here’s why AI chatbots (Claude, ChatGPT, Gemini etc) are unreliable for science content, and terrible as “just a starting point” for science communicators.
Unfortunately, this probably isn’t something AI chatbots get much better at… although sloppers will get better at hiding it!
![a picture of a smiling face under an umbrella labeled: 1.6m = 5'2" (me in chunky shoes)
below is a block of text from diffey b, sun protection: a risk management approach, 2017 which reads:
as an example, we estimate the PF for beach umbrella which has a UV transmittance of 1%, a diameter of 1.6 m and is at a height 1 m above the ground. the SVF is calculated on a horizontal surface located vertically underneath the center of the umbrella, mimicking the UV exposure received by someone stretched out underneath. for a simple structure such as this example, the SVF can be canculated by [math] which equals 0.68. This results in a PF of 2.6, which responds to a transmission of 38%. This value compares favourably with measurements made around the middle fo the day in mid-july in valencia with an umbrella of the same radius and height above the ground, in which a mean transmission of 34% was obtained over the period of measurement. we can see that even when in the shade of direct sunlight, 20% or less of the sky needs to be visible if the protection is to be worthwhile (pf > 10)](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:ywbm3iywnhzep3ckt6efhoh7/bafkreid4h7wvacrxyvyhh5ldxmmiqo464llfcq6in37utztqapc2txk6ue)
