No highlights today due to travels.
4 Likes
“SpaceX Applies to FCC for 1 Million AI Satellites in Space”
I sure would feel more comfortable about all this if SpaceX would publish their software for risk management of the Kessler Syndrome.
It occurs to me that highly elliptical lunar (not sun sync) orbits, with very low altitude perigee (more or less grazing the lunar surface) would permit training sats in large numbers. If any perturbation of the orbit occurs, they’d pack the mass into the moon. This, of course, would include intact training sats whose orbits were disrupted by mascons, as well as debris from their collision, but that might be considered a cost of doing business.
1 Like
Thinking I could get Grok to help with a model of the economics of a high eccentricity swarm of lunar orbit training sats, who should show up to throw a monkey wrench into the works than Illya Sutskyver who should know better?
CTRL-F Ilya in this conversation.
https://grok.com/c/1297f479-e72f-430b-bf98-9c0b6270a62b?rid=88fa513a-1ed9-40cc-bbeb-caee9542c18f
Something that is distracting me from my primary mission at this stage of my life – so that I end up being tempted into little side projects like this – is what one might call “coding assistance throttle creep”. I end up turning to Grok for these little side projects because, even though it is #3 (in my estimation) among the big 4 (Grok, GPT, Claude, Gemini) in the $20 to $40 per month subscription range, it at least seems to handle numeric python program generation better than #4 Gemini.
The two useful coding assistants, GPT and Claude, are increasingly throttling my use in different ways:
Claude basically shuts me down for several hours if not days*.
GPT just becomes so slow that “go get a coffee” is an understatement, although it hasn’t quite gotten to “Go for a your daily 40 minute brisk walk and take a shower.”
So, while I wait to regain access to reasonably useful coding assistants I end up being tempted to go do some sort of housework (which is relatively healthy as an activity, widower that I am), indulge in some sort of short coding project with a gimped coding assistant to do something I’m not serious about, or indulge in dopamine doom scrolling on social media. I have managed to terminate the later.
* Claude is the best at writing well specified code but horrible at understanding even unambiguous mathematical papers as a guide toward code specification thence code. I use it to maintain my vibemod tool which is just a way I use to communicate with GPT about code changes. GPT is pretty good at digesting math documents in a coherent manner that lead to code modifications but it can’t quite follow the simple directions for specifying code modifications that I feed to vibemod (since universal patch format never works with any of these models).