J.P. Loo

J.P. Loo

@jpl.ac

attempting pedantry qua pedantry ‘za wolność naszą i waszą’

159 followers73 following1.4K posts

J.P. Loo reposted

Replying to @ronskolnik.bsky.social

Becher also warns about the danger of a new Israeli government being considered "good enough" - e.g. that (important) action against settler terrorists will ease int'l pressure to end occupation; that facilitating (vital) humanitarian aid to Gaza will ease pressure for an IDF withdrawal.

J.P. Loo reposted

Replying to @timkellogg.me

a lot of people are somehow labeling this as an anti-China story when it’s actually just a whole new mode of risk for infosec to grapple with, something we have no current way to adequately test for afaik

J.P. Loo reposted

An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4

J.P. Loo reposted

Replying to @noamchompers.bsky.social

Kit fine asked an LLM about musicology, and there’s more stuff about non-classical music, which is bad (or maybe not bad, I dunno). But Gideon Rosen thinks analytic philosophy is pretty cool

J.P. Loo reposted

J.P. Loo reposted

🚨 Today (Fri 23rd) is the last chance to download the full 109 page BBS collection on 'conscious AI and biological naturalism', including target article, 50 commentaries, and my author response. Don't miss out (& do click 'save pdf')! doi.org/10.1017/S014...

J.P. Loo reposted

Reminder that "formally proved program X" is a meaningless statement. You formally prove that some program or a reduced model of it upholds some kind of property or invariant

If I moderate my sleeping schedule by [redacted] hours I too will amount to a frontier lab!

Max Woolf@minimaxir.bsky.social

also, it's Saturday 1AM in China. Weird timing for a model drop.

In the BibiLeaks Chronicle, Stephen Pollard claims to have discovered an (unquantified) ‘phenomenon of left-wing voters who will vote for Netanyahu, specifically so he has no need of Ben-Gvir and Smotrich’—and writes that ‘only in the last week or two’ has Eisenkot ‘started to campaign with…vigour’!

J.P. Loo reposted

Hours after her release on 100,000 Egyptian pound bail on Wednesday, State Security prosecutors moved against Mada Masr editor-in-chief Lina Attalah on charges of “establishing and operating a website without a license.” The first trial hearing is scheduled for Tuesday October 27. 🧵

J.P. Loo reposted

Replying to @bradlander.bsky.social

Sadly, another harvest in Masafer Yatta was not. Settlers attacked Palestinian farmers — but the IDF arrested two of the farmers, and evicted the others from their own land, after declaring it a closed military zone. Sound familiar? It’s happening every single day.

J.P. Loo reposted

I guess where I come down on this is like, we gotta just stop publishing 2015-era-Kaggle-entry-style projects in journals and reporting them in the news. The goalposts need to move. Predicting on the eval set better than chance is now basically easy. You should have to demonstrate more, now.

Bad Medical Takes@badmedicaltakes.bsky.social

J.P. Loo reposted

In the West Bank, Israeli forces arrested 80 international and Israeli volunteers non-violently resisting settler terrorism, including two British citizens. The arrests further demonstrate the settlers’ impunity and influence on the state in displacing Palestinians and blocking a two-state solution.

ukfost.co.ukStatement condemning arrest by Israeli forces of 80 non-violent volunteers protecting Palestinian olive harvest · UK Friends of Standing TogetherNote for editors: Standing Together is the largest grassroots anti-occupation movement of Palestinian and Jewish citizens of Israel. UK Friends of Standing Together supports its work. Contact: press@ukfost.co.uk.

J.P. Loo reposted

I don’t understand this argument. The main obstacles to European strategic autonomy are concrete matters like the allocation of contracts, funding and political control—if we were (somehow) pragmatic enough to surmount those, we’d easily be able to set up an institutional framework parallel to NATO.

Eustis Bell's Phantom@eustisbell.bsky.social

SENTIMENT TWO.ONE: Oh well then just NATO around the US My brother in Christ that is the opposite of how NATO works. You can't just freeze out nations you don't like. They can opt out of stuff, like France not attending Nuclear Planning Group meetings, but it isn't because the SecGen banned France.

My sole claim is that US membership of NATO is a red herring. In the scenario we both consider likeliest, it’s the least of our problems because we’re still squabbling about the NGF or command structures; but one can coherently evaluate claims in unlikely counterfactuals.

Eustis Bell's Phantom@eustisbell.bsky.social

You are dismissing a lot of multilateral diplomatic and military integration as easy. As far as surmounting your other issues-- if I were a decade younger and a world class baseball player, I could be on the Brewers, and yet, that shows no sign of happening...

J.P. Loo reposted

More and more of Anthropic's documents are LLM slop now. This is not good and shows they're unable to set boundaries

I don’t think LLMs are converging on e.g. educated pre-LLM Kenyan or Nigerian English from experiments with Pangram. A slightly modified way Pangram might end up biased is (regionally varying) two-way influence: post-LLMs, the reviewers’ varieties of English disproportionately change to match LLMs’.

Colin@colin-fraser.net

(The mechanism being: English-speaking reviewers in Africa do RLFH => AI learns to mimic features of dialects from those regions => pangram learns that those features are predictive of AI)

Gilad Kariv, Democrats #3: Israel should retain the settlement blocs and remain in the Jordan valley for decades. Too many partionists still propose to stay quiet, sanction nothing and await a magical doveish turn, not just because it won’t happen, but because the actual result would be a bantustan.

youtube.comwww.youtube.com/watch

dixit Malebranche (who wasn’t exactly known for going outside):

ponder@ponder.ooo

ive been murdered

J.P. Loo reposted

Interviewer: Can you explain this gap in your CV? Scientist: I made it in TeX and can't figure out how to remove the gap but I also don't want to get yelled at if I post a question on stackexchange, so I have instead decided to appreciate the gap, which is still there

👁️saac 👁️o@eyesack.bsky.social

Interviewer: Can you explain this gap in your résumé? John Cage:

Largely agree with complaints about #5 and #6, but #4 is a distinct and perhaps better point: accepting explicability isn’t lexically ahead of predictive power (or even the converse!) advanced AI systems’ formalisations might mislead us in trial design e.g. in what actually is tested in vivo/silico.

SE Gyges@segyges.bsky.social

i hate this

J.P. Loo reposted

Using AI for just ten minutes makes subjects more instrumentally rational

Nvidie@dervine7.bsky.social

you’re a participant in a study. you’re asked to solve some random math problems. you’re given access to the solutions for the first 12, and then, without warning, that’s taken away. you’re gonna get paid $2.60 no matter what. wdyd

The bare minimum is the activation of the EU blocking statute. More broadly, we need coöperation between the parties to create a Western non-US blocking statute encompassing Britain, Canada, Australia, Japan, Norway &al. It’s not easy, but it’s paperwork—we’re meant to be good at this stuff.

bbc.co.ukUS unveils sanctions on ICC in move court condems as 'assault on rule of law'The European Union described the move against the International Criminal Court as "deeply regrettable".

J.P. Loo reposted

Most AI consciousness people do not believe current models are conscious but believe that it will be hard to tell when the transition happens and so it is worth putting in place appropriate mitigations in advance

A letter from OAI researchers warns their firings: —discourage others from raising safety concerns, disagreeing internally and working with outside orgs; —‘may be used to justify’ limiting external auditors’ access (incl METR); and —leave ‘everyone…guessing where the line is’ due to lack of clarity.

mikitabalesni.commikitabalesni.com

analytic philosopher DESTROYED by facts and his own imagining of a distinction between the desired answer and what he thinks he is saying

Colin@colin-fraser.net

this IS the kind of thing I was imagining! Like, Halting is undecidable on bizarro programs but bizarro programs are in some sense sparse