After being off for lengthy repair (which ultimately turned out to be just a dud power supply!), my telescope has finally returned, so first light tonight, and, aptly for the time of year, NGC 6946, The Fireworks Galaxy. 🔭

“While taking these images of the distant universe, Rubin also discovered 2,104 asteroids—including seven whose orbits hew close to Earth’s own. This number may sound impressive, but in just a few months, it will find over a million—doubling the current known tally.”

That's the thing. We know what happens when Substack wins, because we've seen it so many times before.

If they succeed in becoming the only game in town for newsletters, then their service will quickly degrade and become more expensive to use. It's an old, old playbook.

Both appear to be true: (A) If you outsource homework to AI you learn less & (B) If you use a well-prompted AI tutor with instruction, you can learn more

Whenever there is a paper showing (A) the talk is about AI destroying our brain. When a (B) paper, it is all about how AI will kill school. Sigh.

An 120 hour deep exposure with #JWST. Each of those arc features is a distant galaxy gravitationally lensed by the foreground cluster. As well as distorting the images, the lens also amplifies their light, allowing such remote objects to be studied in detail. 🔭🧪

www.esa.int/ESA_Multimed...

And, given the comments (thank you for all the considered ones), I want to make my position clear: AI is a general purpose technology that will impact many aspects of life & work soon. Some impacts will be good & others very bad. Mitigating the bad & building on the good is key to all of our futures

I wrote about the connection between two big recent stories in AI: the sudden increase in sycophancy from ChatGPT and the growing evidence that AIs can be hyper persuasive.

First we shape our tools, then they shape us. www.oneusefulthing.org/p/personalit...

“That's one of those irregular verbs, isn't it? I have an independent mind, you are an eccentric, he is round the twist.”

Updated this chart with the newest Gemini. It shows the rapid progress in AI over less than two years: costs for GPT-4 class models has dropped 99.7% and even the most advanced models (beating PhDs at the GPQA) are still 82% cheaper than GPT-4.

Probably not worth betting on this trend ending soon.

The Jagged Frontier that we first wrote about two years ago continues to define AI. In any real-world, high-end workflow there will be things that AI can’t do & those weaknesses are hard for most users of AI to predict in advance.

Experienced humans can both identify & work around those gaps.

If you have used LLM image generators, you know they are hard to control: LLM had to send a prompt to a separate image generation tool, it did not not make the image.

Gemini is the first public release of a full multimodal LLM that can directly make images. This allows the systems to do detail work

One thing I like about LLMs is that it means kids who formerly had no venue to develop this skill now have a great place to practice and get better at it. Question-asking is near to being the master skill for human beings...

Yeah that's a great point: a lot of the skill involved in using LLMs is getting really good at asking clear, unambiguous questions of the text you dump into the system - and that's a skill in itself, entirely independent of AI systems