🔥 the room was toasty°hard word. toasty°hard word like bread.
yesterday was minimally°hard word productive°hard word. but some ideas have clarified°hard word.
----
education-related°hard word LLM-related°hard word maybe-testing-related°hard word ideas:
- the goal°hard word is "have a 5 minute video°hard word, see how people perform°hard word". they will need to read an LLM-generated°hard word text°hard word for a target°hard word grade°hard word level. ⚔️ well, actually, we will pre-generate°hard word a few thousand°hard word texts°hard word. ⚙️ for now, we can just have them read some Harvard°hard word Sentences. 🔥 It's°hard word easy to tell the depth°hard word of a well.
- a better "linear°hard word" metric°hard word. when your°hard word reading "increases by 6%", what does that mean? presumably°hard word that your°hard word test score°hard word went up by 6%. 💡 this relieves°hard word the concept°hard word of "Grade°hard word Level" from its overloaded°hard word meaning. 🔥 and the desire of "above grade°hard word level" from its paradoxical°hard word meaning.
- interactive°hard word / non-interactive°hard word learning. there is something to be said°hard word for "useful°hard word videos°hard word tailored to a student's°hard word interests" 💡 but, also, easy paths towards them being over-optimized°hard word
- the machine can calculate°hard word values like "confidence°hard word" 💡 maybe "over-confidence°hard word" for students°hard word who have rehearsed°hard word, "fluency°hard word", etc°hard word. 💡 the machine can also complete the list of metrics°hard word
- determining°hard word what intervention°hard word can solve°hard word what condition ... is not°hard word something the machine can do currently.
- finding "gaps°hard word" in knowledge / thoroughly°hard word tracking existing knowledge. 💡 the machine won't°hard word track this, but a database that is attached°hard word certainly can. 🔥 there has to be some math to adjust for possible forgetting°hard word.
----
other ideas / tasks / goals°hard word
- the "series°hard word of questions about a text°hard word" project. and, then°hard word, exception handling°hard word. 💡 it's°hard word not°hard word clear if a linear°hard word flowchart°hard word is sufficient°hard word; or even a linear°hard word flowchart°hard word with GOTO°hard word. ⚔️ well, actually, with enough GOTO°hard word it is clearly possible; just possibly confusing°hard word
- single-domain°hard word experts. right now, people are doing°hard word this for "simple" domains°hard word like "customer°hard word service".
- "politics°hard word" of education. the white whale is Not°hard word Everybody°hard word Should Go To College°hard word. this, already, might be more unpopularity°hard word than I can chew°hard word.
----
and, the latest°hard word Atacama°hard word hopes and dreams, deliberately°hard word slow-walked°hard word to avoid°hard word spending°hard word too much time here:
- a list of the syntax°hard word tags and what they look like, on some documentation°hard word page
- fix the "multi-level°hard word channel°hard word" listing 💡 including°hard word that "restricted°hard word" channels°hard word shouldn't°hard word be named publicly
- something with "articles°hard word". 💡 which turns into "how much do I want to dig°hard word into my past writings" and "how much do I want to amplify°hard word anything°hard word"