Repeating what some have already said here:

• PBS SpaceTime is outstanding, and manages to ride the line between informative and accessible very well. Some episodes especially around heavy math/quantum mechanics are impenetrable for me but all the space stuff is great, the scripts are very well written, production value is top notch.
• Dr Becky provides amazing content mostly geared around recent research and theories - especially with the James Webb Space Telescope being a year old now there’s some amazing insights coming out that she does a great job explaining. A bit less “pseudo lecture” than SpaceTime but still highly informative
• StarTalk (Neil Degrasse Tyson) is great, but in a different way. It’s less formal and very much more like a podcast than a lecture or report as the prior two are.
• Sabine Hossenfelder delivers a periodic “science without the gobbledegook” show that covers all areas but generally has a focus on physics and astrophysics. She’s semi-famous for not tolerating nonsense while also considering a sizeable portion of contemporary physics research to be nonsense. I think she’s hilarious in a parchment-dry German kind of way, and her content goes arguably deeper than the other channels listed here in terms of subject matter - I usually leave her videos thinking about things in a different way.
• SmarterEveryDay is a general science/learning channel but really piqued my interest with a recent video about talking to NASA:

https://youtu.be/OoJsPvmFixU?si=NrURYGlLii4Dbi1_

The host has a background in aerospace engineering and missile test flights - so its about as close to rocket science as you can get! He knows his stuff and has a lot more practical, engineering related videos - kind of makes you think about how to operationalise the more cerebral ideas of the other channels.

Hope you enjoy some or all of the suggestions here and from other commenters

I’m both impressed and concerned at the level of detail you supplied here, but…thank you? For some of the context

The maker of ChatGPT had made progress on Q* (pronounced Q-Star), which some internally believe could be a breakthrough in the startup’s search for superintelligence, also known as artificial general intelligence (AGI), one of the people told Reuters. OpenAI defines AGI as AI systems that are smarter than humans.

Definitely seems AGI related. Has to do with acing mathematical problems - I can see why a generative AI model that can learn, solve, and then extrapolate mathematical formulae could be a big breakthrough.

Theres a lot of research going into carbon sequestration through soil and plant technologies - basically accelerating what would happen naturally by a few orders of magnitude.

Rapidly filling artifical peat bogs (through things like algae/weeds that are genetically modified to absorb more CO2) would allow for semi-permanent carbon capture as long as no one digs it up again. Similar projects with seaweed are under research as well.

Personally I think anything to do with carbon capture is a bandaid at best, and failing massive global cooperation and societal change, we’re going to end up needing to geoengineer our way out of the problem. Things that block or impede solar heat absorption to cool the planet - atmospheric aerosols, artificial cloud generation, solar shades out in a lagrange point, basically manipulating conditions to influence how much energy is going into the system. There’s a nonzero chance we fuck it all up but as we hurtle through temperature records and tipping points, the idea of net zero emissions actually having an impact in our lifetimes seems more and more unlikely. There’s too much inertia in the system.

I find it interesting that DALL-E still doesn’t understand text - look at all the random Daschuund spelling in the generated images. It knows what the word should look like, but has no framework to interpret or distinguish text from the other elements of the image. It looks like trying to spell in a dream.

Where it gets really challenging is that LLMs can take the assignment input and generate an answer that is actually more educational for the student than what they learned d in class. A good education system would instruct students in how to structure their prompts in a way that helps them learn the material - because the LLMs can construct virtually limitless examples and analogies and write in any kind of style, you can tailor them to each student with the correct prompts and get a level of engagement equal to a private tutor for every student.

So the act of using the tool to generate an assignment response could, if done correctly and with guidance, be more educational than anything the student picked up in class - but if its not monitored, if students don’t use the tool the right way, it is just going to be seen as a shortcut for answers. The education system needs to move quickly to adapt to the new tech but I don’t have a lot of hope - some individual teachers will do great as they always have, others will be shitty, and the education departments will lag behind a decade or two as usual.

I would suggest that instances should have settings that allow them to decide whether to “advertise” a community list. With configurable settings like "all, “most active”, “top X”, or even a manually maintained list depending on the admins and instances preferences.

Then your home instance, when searching, should have it’s own settings to decide what results it’s going to ping other servers for. Big/popular/high confidence instances can have an open all/all relationship, while you might query only the top 10 communities from unknown or new instances to handle the scenario you describe.

Federation can be binary yes/no but there should be room to add more logic around enabling search on communities from your instance and controlling the search results from other instances. I don’t think the two are mutually exclusive, unless I fundamentally misunderstand how federation works!