Tommaso Sciortino

I saw a demonstration in front of the Ferry Building today with lots of Iranian flags (the old royal version). The crowd seemed... pro-Trump?

Not clear how representative of the diaspora they are.

Just finished reading Moonraker and not only does it bear absolutely no relation to the movie, but Bond doesn't even get the girl at the end!

Still an enjoyable read!

Edge case: Frankenstein's monster earns a doctorate

Wait, what?

Ok, but this is a mostly bipartite graph. Is a giant component still likely under those conditions?

Still embarrassed I voted for Nader.

So like, it's not a "truth" manifold so much as a "boring" manifold.

BlueSkyers who reply with "it's always been like this", and trot out tame precursors AT BEST and non sequiturs at worst, get blocked because whether they realize it or not, they are a big part of the problem.

i am willing to say that i can speak with a fair amount of authority, in general, about the types of beliefs which might lead someone to leave an oversight role at an AI company

My kids are making a lego set of a bonsai tree which is like two levels of indirection: They're making a thing that's supposed to look like a thing that's supposed to look like a big tree.

Reading @davidcrespo.bsky.social argue with lawyers about LLMs here has been really informative. I've learned things like: some lawyers are unable to explain the difference between a "citation" and a "reference" to non-lawyers.

At one point I gave up and just started writing python programs to solve them.

Our way of life as we knew it could not even be bothered to defend itself against an explicitly-stated existential threat. We have to find a mode of ourselves that believes in our right to exist.

Also, there's absolutely no crisp distinction between "an LLM" and "an LLM with certain preparatory instructions". Prepending "please ask questions if the following request is not clear" is equivalent to using a LLM with slightly different weights.

Do you understand why people are sensitive to comments that could be interpreted as bigoted? We are living in an age when Nick Fuentes is a well-known podcaster who got famous marching with Tiki torches and signs that said, "Jews will not replace us".

Because the reinforcement learning with human feedback tuned it to the level of independence that The software companies thought users would like best. There definitely are models that by default ask questions more often than others.

I remember when I was 12 years old, if I didn't know a word I was taught to look for "context-clues" which was precisely the words before and after the text in question. What else could "context" mean?

The problem here is equivocating between Israelis and Jews. Israel likes to pretend that their cause is identical to that of Judaism. But there's no reason for us to go along with that propaganda effort.

It's not clear why you would single out Jews here, tho. 5% of Italians probably support the war too. Why not rag on them?

That's the point. You see why "hit dogs holler" is stupid when it's applied to you. You just didn't see why it's stupid in general.

Hit dogs holler.

Indeed, if you're not a bigot accusing all Jews of having dual loyalties then @joelhs.bsky.social 's post is not about you.

"RIP to everyone killed by the gods for their hubris but im different..."

I very much doubt the CEOs have read Rawls.

i think in Scott’s case this is also a Hanania-esque “i am superior genetically to the other racists, who are ruining racism”

"Failure to generalize" is a big problem in Silicon Valley. Both for AI models, and CEO's political views.

Write it and prove me wrong.

We know the answer to your question is "no". If it were possible to do this, someone would have done it already and made a ton of money. Thousands of different companies worked on this space for decades. They made basically no progress until LLMs.

LLMs write my unit tests for me. Saves me a ton of time and effort. That's super useful. Annoying to produce but relatively easy to verify.