Leon Lahoud

Lebanese Canadian. My opinions are my own.

[Leon Lahoud @leonlahoud]

Asking any model an opinion question such as “what is the best story ever told” will get you different answers depending on the language. I am writing an analysis piece on this. It’s really An example where and LLMs answer comes from the nearness of the tokens in embedding space. The Chinese example is because of Chinese Internet content.

[Leon Lahoud @leonlahoud]

@joshwhiton no it's not

[Leon Lahoud @leonlahoud]

@politicalpen_ And there should be world peace. Don’t forget world peace

[Leon Lahoud @leonlahoud]

@ksorbs Church of Satan more like it

[Leon Lahoud @leonlahoud]

@ziv_ravid Dario has a PhD in computational neuroscience. Not to say you are wrong. But instead of an appeal to expertise, it would be better to write up and explain to the rest of us your expert point of view

[Leon Lahoud @leonlahoud]

@MarioNawfal Fuck him and may he rest in pieces

[Leon Lahoud @leonlahoud]

@AnaKasparian Is it right to sacrifice Lebanon for your cause?

[Leon Lahoud @leonlahoud]

@pmddomingos Does that apply to the government of Singapore for example?

[Leon Lahoud @leonlahoud]

@CHEWY_BRICK @nntaleb Unless the Somali actually got in on h1b

[Leon Lahoud @leonlahoud]

@Saidray116489 @nntaleb how do immigrants fair in Canada for example?

[Leon Lahoud @leonlahoud]

@maccaw price will eventually come down. Give it two years max.

[Leon Lahoud @leonlahoud]

@nntaleb LLMs are doomed!

[Leon Lahoud @leonlahoud]

Why wouldn't there be a limit on intelligence?

[Leon Lahoud @leonlahoud]

@DKThomp the road to dictatorship?

[Leon Lahoud @leonlahoud]

happy 250th 'Merica

[Leon Lahoud @leonlahoud]

@ylecun @andrewgwils Can neural networks get us there though?

[Leon Lahoud @leonlahoud]

@Tech_girlll Slower but better.

[Leon Lahoud @leonlahoud]

@TTrimoreau Triggering only the top half: LLM scaling will hit a wall, and open source will win the race.

[Leon Lahoud @leonlahoud]

Have you considered that the LLM has an infinity of subjective experience while it is inferencing the next token? As it is quite fast in terms of clock cycles, time would run slower for it while it inferences. Just like for humans, where some dreams seem to last forever, even though they were only 5 minutes long.
I don't believe this, but it's an argument that needs to be seriously considered and refuted.