Glad to see my first nonhuman book purchase.

Another one: “Masked ophthalmologists graded the accuracy, relevance, and overall preference [of AI answers]… GPT-4 compared favourably with expert ophthalmologists (median 76%, range 64–90%), ophthalmology trainees (median 59%, range 57–63%), and unspecialised junior doctors…”

Every OpenAI product update kills a bunch of startups & expensive consulting projects.

RAG is increasingly being built into core AI model offerings (but RAG is quite limited, and, at times, risky. See halfway though this post: oneusefulthing.org/p/which-ai-sho…)

The Curse of YouTube Confidence: after watching a short tutorial about a subject, we think we know how to do it.

Subjects shown this totally uninformative video of pilots landing a plane become more confident they could land one themselves if needed: royalsocietypublishing.org/doi/10.1098/rs…

AIs are very proficient at text-based medical problems, often outperforming doctors, but the multimodal GPT-4V system is still much worse than humans at medical images. medrxiv.org/content/10.110…

Highly recommend that academics spend the time to come up with good names for phenomena they study. And this is great: “In the second place, and more important, nobody knows what entropy really is, so in a debate you will always have the advantage.'”

“GPT-4 ranked higher than the majority of physicians in psychiatry… it performed similarly to the median physician in general surgery & internal medicine… GPT-4 performance was lower in pediatrics & OB/GYN but remained higher than a considerable fraction” of active doctors.

They have a playable demo: video2game.github.io

As you will see, we are not that close to “turn any video into a video game level” but it is pretty remarkable to be able to interact with a 3D environment from a single video, even in a limited way.

I don’t know about overall use, but when you go to conferences you can see the sad decline of Twitter’s relevance to the professional world. People used to talk about who they learned from & their interactions on this platform, now it is all private chat groups & (yikes) LinkedIn

There is a ton of debate about how good AI might get, but not enough recognition that the research on AI in law, medicine, business etc. finds that GPT-4 class AI is good enough to make a serious difference in work & education

In some ways, future capabilities are a distraction

We know that LLMs are more persuasive than most humans, this study offers some tentative reasons why that may be true:

LLMs produce arguments that are MORE morallly charged than humans do, and which require more cognitive work from humans to understand arxiv.org/pdf/2404.09329…