The weird thing is I've seen LLMs "typo" stuff pretty often. Yesterday I asked Gemini a question about the Python Twisted framework and it answered about Deferreds but misspelled it as "Deferends" in one spot.

Also, "rouge" is one of the oldest spelling mistakes on the internet. ALL BLUES ROUGE LFG WC/RFD

Video is here: https://www.youtube.com/watch?v=HUEvRyemKSg

I couldn't parse "Getting that model has meant taking on something else."

This "cognitive virus" is pretty similar to the original sense of the word "meme" as concieved by Dawkins.

I agree, except for the use of the word "yet" .

I think what's missing is fundamental. I think the reason it sucks so much to work with LLM-generated code is that LLMs will never "know" what it's like to be human. They don't "understand" our frustrations and motivations, and they're missing the vast array of useful mental tactics we've evolved to cope with corporal existence.

At this point I think progress towards a good colleague bot would require a new architecture which allows continuous leaning, and for the LLM to be raised as a human child (maybe in a simulation at 1000x speed or something).

Cambridge alum here. Agree; admissions is a crapshoot. Job interviews are a crapshoot. C'est la vie.

I went to a similar school. It depended on major. If it was a major that the school was known for, say the top ranked CS school...then it was 100% that a grad was someone who was noticeably much, much smarter than most people. If they were at the same school but in a much less well recognized major (at my school...let's say history), then they were much closer to the average. If you are seeing students at Cambridge who are in prestige majors but without obvious skills...something else is happening and it isn't good. Otherwise, you are probably just seeing the differences between different majors.

  > say the top ranked CS school

I've worked with and written papers with enough people from top ranked CS schools (multiple from top 5 on csrankings) while I was at a top 100 and it is a very mixed bag. There were plenty of idiots in the top 5 and plenty of geniuses in the top 100s.

I'd agree that average shows a difference but I don't think it is nearly as big as people act. Even the intra-university variance isn't what people would think (top universities have a very narrow distribution while lower ranks have wider). The trend follows, but the variance of the variances is not that large.

Cambridge is top ranked for everything they offer.

But there's going to be a difference between the students who finished with the highest grades, and those who scraped through.

"Cambridge is top ranked for everything they offer."

You know that there are places outside of England right?

Forcing a short delay is such a powerful technique.

I use this when I'm at risk of binging a show. If I'm tempted to let Netflix continue into the next episode, I say "ok maybe, but let's get up and do something else for a few minutes first and see if I still want to" and before I know it I'm building a robot instead.

I'll vouch for this too

This is something I've been working on a lot myself. It's so easy to wind up sliding into the familiar patterns and routines, and reaching for the phone. I'm finding that discipline is less about resisting it and more about redirecting it to something more productive

reverse/compliment to this to start working for me is just aiming to work for 1-2 minutes when it feels hard to get started. i usually end up getting immersed, and find it hard to stop.

Not sure if you're joking, but I remember noticing the same thing last time I spent time there (2017). Lumber yards tend to have a lot of lumber. It's kind of like when you drive by a feed lot and they have an implausible number of cows (the ones over in imperial valley amazed me). Also there's always tankers by my house in Manhattan Beach. These anecdotes are not economic indicators.

By "decompress," "eliminate deictic language," etc. are you trying to say "speak simply and clearly?"

Eschew obfuscation...

What's the problem with "eliminate deictic language" ?

It is clear, specific, and terse. It keeps both llm tokens down and human prose-reading to a minimum.

This must be the reason why LLMs speak that way. They naturally reflect human tendency to signal smartness, while effectively achieving the opposite.

It’s not that. It’s that I tried a bunch of experiments on which instructions worked best. “Speak clearly using simple language” lost too many details. “Be sure the objects that pronouns reference are clearly identified along with any context you might assume I have but I don’t” worked better, but I got tired of typing that in each time. So I looked up words to match that phrase and “deictic” fit perfectly. The results were pretty good, and so I’ve been using that prompt ever since.

Ah, thanks for the explanation. I didn't realize you were sharing the actual prompt.

Of course it's clear which is more likely in this specific case, but from the semantic structure I read (present tense) it the other way around. It actually struck me as a grammatical error when I first read (past tense) it.

If I said "my water is ice now," one would assume freezing rather than melting.

So in this case one would assume a server became a phone, when actually a phone became a server.

(English is my first language.)

Yes, that’s a great example.

I think most of the time you’d play with the order it would be for comic effect, like someone saying “my dinner is pretzels now” after having too big a snack.