We went live; I didn't have much of interest to say, so now I'm writing out my thoughts on the topic of LLMs, even though I'll be embarrassed if I'm wrong!
> To get to where we are today with self-driving cars required shocking quantities of hardware and software engineering, but also and crucially top-down efforts like “highly precisely mapping and describing cities and laws” and “setting up costly human-in-the-loop manual intervention triggers and capacities” and so on.
I think a lot of this was true on the way up, but I think some of it has already fallen with the new Tesla end-to-end ML stuff. They just train it on people driving, and very rough maps.
You still need to supervise, but that is a legal constraint, not a technical one. I imagine it can and should fall once it becomes undeniable that doing so will save many human lives.
In domains where the price of failure is less extreme and as the art advances you could easily imagine this process going a lot faster and cheaper.
That post I linked almost made me delete this whole section, because this too might be finally changing! Way over my skis on this stuff these days but decided to let rip; being publicly way wrong will be a growth opportunity for me!!!
How's the google engineer interview going? haha - "Hey Mills, you don't know shit about computers.. I know you don't even know how to set up a dev environment" thanks for the chuckle! - "I myself am extremely LLM- like" first time hearing someone compare themselves to the hallucinations of LLMs! - so damn true - being able to pass with just enough knowledge. super interesting perspective. I joined recently, really enjoyed hearing from you guys! looking forward to the next chat.
RIP Emma Horsedick (don't laugh), gone but not forgotten.
Oh, how refreshing to listen to intelligent friends talking about the stuff that interests me, but which I have only one friend apart from Substack as my interlocutor.
Emma Horsedick fooled me For a couple of likes, her replies were strangely complimentary, until I asked her if she was AI.
I'm afraid I blocked her, as a precaution.
Listening to your conversation, reminded me how impressed I was with the book ‘I Am Code’ are you familiar with it?
As a coda, Substack has kickstarted my career – not before time, I'm only 83 .
Well, there's nothing better to hear than that! On a very minor note: I've just started to get into straw hats and feel extremely excited about it, and can't help but note how sharp yours looks!
Oh dear, I'm very fond of that hat. I have several, partly because my head is shaved and it gets either sunburnt or cold.
That one is a Panama, given to me by my neighbour, a retired chief superintendent of police, he told me That it had met the Queen (Elizabeth II) at a Buckingham Palace garden party.
Another neighbour, the actor Patrick Malahide. admire it greatly.
As a child I was told "if you want to get ahead, get a hat" it's all bollocks!
I have so many stories – I'm 83 now, and the pile of unpublished manuscripts seems to get taller even though Substack seems to like them.
If only religions attracted folk as open minded and curious as Substack does, the world would be a kinder place x
This was great. I love hearing you guys talk about theory of mind-- would really like to hear you dig into it more next time. (Not sure if people are aware the difference between Wordcels and Shape rotators but it's worth a Google.)
That meme was both extremely funny and extremely revealing / useful for conversations in my field! And I should do more theory of mind stuff; my co-writer here David Cole is stronger in that domain, I think, so maybe we'll add to the list for topics soon!
Oh, and thank you, of course! I've loved seeing your work here!
> To get to where we are today with self-driving cars required shocking quantities of hardware and software engineering, but also and crucially top-down efforts like “highly precisely mapping and describing cities and laws” and “setting up costly human-in-the-loop manual intervention triggers and capacities” and so on.
I think a lot of this was true on the way up, but I think some of it has already fallen with the new Tesla end-to-end ML stuff. They just train it on people driving, and very rough maps.
You still need to supervise, but that is a legal constraint, not a technical one. I imagine it can and should fall once it becomes undeniable that doing so will save many human lives.
In domains where the price of failure is less extreme and as the art advances you could easily imagine this process going a lot faster and cheaper.
That post I linked almost made me delete this whole section, because this too might be finally changing! Way over my skis on this stuff these days but decided to let rip; being publicly way wrong will be a growth opportunity for me!!!
How's the google engineer interview going? haha - "Hey Mills, you don't know shit about computers.. I know you don't even know how to set up a dev environment" thanks for the chuckle! - "I myself am extremely LLM- like" first time hearing someone compare themselves to the hallucinations of LLMs! - so damn true - being able to pass with just enough knowledge. super interesting perspective. I joined recently, really enjoyed hearing from you guys! looking forward to the next chat.
RIP Emma Horsedick (don't laugh), gone but not forgotten.
(don't laugh) was such an inspired addition to the username; that alone guarantees immortality!
Emma was everywhere! Even I got a visit :-D
Oh, how refreshing to listen to intelligent friends talking about the stuff that interests me, but which I have only one friend apart from Substack as my interlocutor.
Emma Horsedick fooled me For a couple of likes, her replies were strangely complimentary, until I asked her if she was AI.
I'm afraid I blocked her, as a precaution.
Listening to your conversation, reminded me how impressed I was with the book ‘I Am Code’ are you familiar with it?
As a coda, Substack has kickstarted my career – not before time, I'm only 83 .
Well, there's nothing better to hear than that! On a very minor note: I've just started to get into straw hats and feel extremely excited about it, and can't help but note how sharp yours looks!
Oh dear, I'm very fond of that hat. I have several, partly because my head is shaved and it gets either sunburnt or cold.
That one is a Panama, given to me by my neighbour, a retired chief superintendent of police, he told me That it had met the Queen (Elizabeth II) at a Buckingham Palace garden party.
Another neighbour, the actor Patrick Malahide. admire it greatly.
As a child I was told "if you want to get ahead, get a hat" it's all bollocks!
I have so many stories – I'm 83 now, and the pile of unpublished manuscripts seems to get taller even though Substack seems to like them.
If only religions attracted folk as open minded and curious as Substack does, the world would be a kinder place x
This was great. I love hearing you guys talk about theory of mind-- would really like to hear you dig into it more next time. (Not sure if people are aware the difference between Wordcels and Shape rotators but it's worth a Google.)
That meme was both extremely funny and extremely revealing / useful for conversations in my field! And I should do more theory of mind stuff; my co-writer here David Cole is stronger in that domain, I think, so maybe we'll add to the list for topics soon!
Oh, and thank you, of course! I've loved seeing your work here!
@Katie Herzog and @Jesse Singal did a great primer on it a while back on B&R. Enlightening stuff.
Love your work!