HN user

remontoire

128 karma
Posts0
Comments45
View on HN
No posts found.

I think it's because you want to be able to predict the next token using only 1 token or the whole context window (and any size inbetween). So, you end up getting n different losses for each text snippet (where n is the size of the context window).

If i'm wrong, can someone correct here, would be useful to know.

I've used fset and cl21. The problem is that once you start using those you no longer are able to use libraries. Data structures have to be built into core.

The common lisp community has rejected reader macros for interop reasons the last time I checked (in the case of cl21)

Even the docs for fset doesn't show the use of data literals. it does things like (set) and (isetq s2 (set 'c 1 "foo" 'a 'c))

This pretty much proves my point. it's not fun typing (set 'c 1 "foo" 'a 'c) instead of #{'c 1 "foo" 'a 'c}. It's also not possible to put those on the wire (easily) for communication between common lisp processes. Breaking the whole point of homoiconicity.

There is more to collections than lists. You can have instances of empty collections, some of which have literal support ([], {}, and ()). Thus there can be no sentinel empty collection value.

Probably my favorite feature over common lisp. Combined with comma (,) being treated as whitespace (ie, use comma if you want to but you don't have to), makes typing out data structures in clojure a much more enjoyable process than other languages.

This ease of typing out data structures also extends to edn (clojure's version of json)

Not much if Toyota can't compete on price. Range has not been an issue since 2015, and the marginal increase in consumer demand won't go up much if a $150k 500 mile electric hits the market.

The real issue for electric is that a 300 mile electric costs more than $35k because the world wide battery production capacity is limited.

If solid state batteries are easier to produce and the price comes down faster than lithium ion then there might be a crossover point by 2030. Until then, this probably means cooler high end sports cars (formula E?)

Starship Update 7 years ago

Totally agree, no one has ever been inspired by this man!

In fact, he has actively put spaceflight back 50 years by making everyone disinterested in going to mars.

What we need are more people in suits who give us made up numbers.

I find it more than a little surprising that no one here has an issue with google starting an AI lab in China. Instead all I see pointed out are the motives of the author.