Joshua Loftus

Joshua Loftus

@joft.bsky.social

Random wanderer, toddler parent, famous fitness influencer Statistics & Data Science Asst Prof at LSE (machine learning, causality, fairness, interpretability, ethics). First gen grad. Unschooled -> math @ WMich & Rutgers -> stats PhD @ Stanford

1.9K followers401 following495 posts

Joshua Loftus
@joft.bsky.social

Why would RFK Jr be obsessed with the narrative of decline? A Kennedy, preoccupied by the possibility that people used to be better in the past, but are now massively disappointing... hmm

This post is unavailable.

Joshua Loftus reposted

Katie Mack
@astrokatie.com

A reminder, because it is almost that time of year: More children are killed by cars on Halloween than any other day of the year, by far. www.vox.com/future-perfe... If you're a parent, I hope you'll do what you can to keep kids safe. If you're driving on Saturday, PLEASE exercise extra caution.

Joshua Loftus
@joft.bsky.social

But but but... just think of how well targeted the AI chatbot ads will be. Surely that's more important?

Baldur Bjarnason@baldurbjarnason.com

If you’re going to spend that much money, build that much infrastructure, there is only one cause that could both use an effort of that magnitude and it’s important enough to require it: climate change

Joshua Loftus
@joft.bsky.social

One of my most... me?... beliefs, is that living things really are different from machines precisely because evolution is ***not*** design (or engineering or optimization)

TristanTD@tristantd.bsky.social

Something I wish more people would realize.

Joshua Loftus
@joft.bsky.social

There is no point to life, I have no purpose and I am proudly not optimized for anything. Fuck to "evolutionary fitness" or "progress" or whatever other hindsight fairy tale people tell themselves about the meaning of this all. It's randomness

Joshua Loftus reposted

Ronan Farrow
@ronanfarrow.bsky.social

The government has a formula that decides whether safety rules are worth passing. And that formula was rewritten to put a new price on your life: zero dollars. #PublicHealth #EPA #News #Trump #climatechange

Joshua Loftus
@joft.bsky.social

Ah, well, nevertheless… we must all accept that it is the future

Chris Person@papapishu.bsky.social

I have been meaning to write about this for a month or so now but it breaks my heart. AI has infected every hobby. Everything is noise. It's hurting existing communities and trudging through the slop is exhausting.

Joshua Loftus
@joft.bsky.social

It’s just mono vs polytheism The sun alone or many heavenly bodies

Toby Buckle@polphilpod.bsky.social

counterintuitively perhaps, the demographic group with the most astrology believers is catholics

Joshua Loftus
@joft.bsky.social

Around ~70% of smokers say they want to quit, more than half try to quit at least once every year

William B. Fuckley@opinionhaver.bsky.social

This is true, most bad things in media and politics are genuinely demand driven to a far greater extent than we want to admit, *but* I think this expresses the limits of treating revealed preferences as the only valid ones. People say that they don’t want this, and I don’t think they’re lying.

Joshua Loftus
@joft.bsky.social

Put me on record as < 2%

Guy Berger@econberger.bsky.social

The 17.1% forecast is almost certainly wrong ***by a lot***, but it’s excellent for intellectual accountability when people put these kinds of predictions out

Joshua Loftus
@joft.bsky.social

For me, getting older and going dad mode -> hack squat machine AMRAP

Tim Zhu@timzhu.io

The older you get and the more time you spend at the gym the more you start doing weird rep ranges, like 1x9 or 2x7

Joshua Loftus
@joft.bsky.social

Did something happen to my algorithm on here or is there really a recent backlash against people being anti-AI? Seems there might be a major new polarization underway

Joshua Loftus reposted

Konrad Burchardi
@konradburchardi.bsky.social

Data project (AI-able) and exam (paper, no aids) score distribution on the undergraduate Development Economics course I have been teaching at Stockholm University, 2022-2026.

Joshua Loftus reposted

Morgan Polikoff
@mpolikoff.bsky.social

"Among sections of the same course, AI tutor access reduced final grades by 0.37 standard deviations and learning management system participation by 0.90 standard deviations. Participation declines ... were greater among first-generation students." edworkingpapers.com/ai26-1598

edworkingpapers.comThe Effects of Course-Integrated AI Tutoring on Student Performance and Engagement: A Randomized University TrialWe evaluate access to a course-integrated generative AI (GenAI) tutor in a randomized controlled trial involving 2,379 undergraduate students and 30 instructors across multiple disciplines in a large ...
Joshua Loftus
@joft.bsky.social

Masculinity is all about needing to make money so fast that you keep losing your money

lastpositivist.bsky.social@lastpositivist.bsky.social

Easily the most Male Centric space I am in is the London Area Warhammer Group Chat and I can tell you: by far and away the most common pathological behaviour is not people being ultra aggro or reactionary or grossly misogynist or cetera. No. It's people falling for crypto or gold trading scams.

Joshua Loftus
@joft.bsky.social

Car alarms should have been the most instantaneously banned invention in history. Sirens? Make them automatically quieter at slower emergency vehicle speeds. I want a war on noise pollution. I am a voter willing to make this one of my top issues

Joshua Loftus
@joft.bsky.social

All my teaching this term, research still going, gotta do taxes, find a place to move in ~3 weeks, chasing 16 month old around trying to stop him from shoving every other child on the playground Now seems like the perfect time to start a new project. Maybe get into podcasting

Joshua Loftus
@joft.bsky.social

Right, what could a distinguished professor of computational linguistics possibly understand or have to say about large language models? Journalists should only listen to people who work at companies selling the technology.

Joshua Loftus
@joft.bsky.social

Conjecture: these are the first to be retracted because it's easier for an LLM reviewer to find a sign error. If humans ever bother to read all 719 remaining papers they will find other, deeper kinds of errors, nonsense, and lots more that may be true but is criminally awkward

Colin Hoad@colinhoad.com

Far less fanfare and media panic now that the proofs are proving to be, err... unproven. aiweekly.co/alerts/opena...

aiweekly.coOpenAI Pulls Three AI-Generated Math Papers Over Sign Error | AI WeeklyOpenAI withdrew three of the 722 AI-generated math manuscripts it released on October 6, after a sign error in one proof collapsed the argument and the…
Joshua Loftus
@joft.bsky.social

Re: the Stochastic Parrot (SP) arguments It's too tedious defending someone else's argument from people who are deliberately strawmanning it, so let's expand the scope a bit and consider the current high level debate What has fundamentally changed about "AI" now vs, say, ~3 years ago?

Quantian@quantian.bsky.social

She is implying there’s a secret calculator in closed models (that open models don’t need somehow) and the fact that there’s errors (which are quite rare) are more likely the be text classifier mistakes than the model having innate ability but occasionally messing up. Not very occam’s razor of her!

View full thread
Joshua Loftus
@joft.bsky.social

Possibilities: 1) It hasn't actually done any ("independent") checks 2) It did check, but the check was wrong 3) It did check, correctly, but the report of the results is hallucinated 4) Check and report correct, but there are still other errors remaining it didn't find 5) All correct

Joshua Loftus
@joft.bsky.social

Some benchmark results imply that the likelihood of (5) has increased. Maybe that's good, but it doesn't mean we're no longer dealing with Stochastic Parrots. The same software product that can code (extremely inefficiently) can also generate more persuasive-sounding astrological charts

Joshua Loftus
@joft.bsky.social

Re: the Stochastic Parrot (SP) arguments It's too tedious defending someone else's argument from people who are deliberately strawmanning it, so let's expand the scope a bit and consider the current high level debate What has fundamentally changed about "AI" now vs, say, ~3 years ago?

Quantian@quantian.bsky.social

She is implying there’s a secret calculator in closed models (that open models don’t need somehow) and the fact that there’s errors (which are quite rare) are more likely the be text classifier mistakes than the model having innate ability but occasionally messing up. Not very occam’s razor of her!

View full thread
Joshua Loftus
@joft.bsky.social

This enabled large improvements on many benchmarks. Much wow, very frontier Does this mean that the SP criticism has become irrelevant? I don't think so. Here's one example to explain why:

Joshua Loftus
@joft.bsky.social

Claude now tells me something like: "Three independent check rounds each found errors in my [Claude's] drafts" OK, inference compute go brrr. How much should I trust a Stochastic Parrot that "checked" its own work... stochastically... and is--again, stochastically--reporting the results?

Joshua Loftus
@joft.bsky.social

Re: the Stochastic Parrot (SP) arguments It's too tedious defending someone else's argument from people who are deliberately strawmanning it, so let's expand the scope a bit and consider the current high level debate What has fundamentally changed about "AI" now vs, say, ~3 years ago?

Quantian@quantian.bsky.social

She is implying there’s a secret calculator in closed models (that open models don’t need somehow) and the fact that there’s errors (which are quite rare) are more likely the be text classifier mistakes than the model having innate ability but occasionally messing up. Not very occam’s razor of her!

Joshua Loftus
@joft.bsky.social

I would say, in increasing importance: (1) LLMs are bigger (~10x parameters), (2) "harnesses" and "tools" automate some of the context curation problem, (3) "inference time" computation -> generating many responses and automatically picking the best

Joshua Loftus
@joft.bsky.social

This is also a strawman of what the Stochastic Parrot camp has been saying, but let’s not let good faith get in the way of a good social media dunk

Quantian@quantian.bsky.social

For what it’s worth, arguing LLMs can’t do addition now is flat earth-level science denialism. Here’s Gemma 4b, a non-reasoning model that could only theoretically memorize binary operations up to 65535 x 65535, oneshotting a eight digit addition and hex conversion *on my phone* with no tool calls

Joshua Loftus
@joft.bsky.social

This is just standard academic insistence on accuracy (“pedantry”) Bender is correct and makes important points about not trusting claims about proprietary models If you want to say some LLM capability is genuine, you actually need better evidence than what‘s given here

Colin@colin-fraser.net

I really wasn’t trying to start a fight with Prof. Bender but now that she has created a long thread subtweeting me and her disciples are all yelling at me, let me just say that this is the crux right here. She seems to respond as though I am describing something a priori impossible.

Joshua Loftus
@joft.bsky.social

it’s not even important for the stochastic parrot argument to deny the possibility of LLM arithmetic capability! There’s tons of worked examples of sequences of arithmetic operation symbols in the training data

Joshua Loftus
@joft.bsky.social

I would be surprised if Bender is actually arguing LLM arithmetic is impossible (rather than e.g. error prone) She is specifically denying that observing closed model capabilties is evidence about LLMs, and she’s right!

Joshua Loftus
@joft.bsky.social

When someone is killed by car: this is normal, expected, inevitable really When someone is killed by a bike: we must investigate this immediately

How-Sen Chong@how-sen.com

Listen: I don’t want to minimize the danger of high e-bike speeds. But if the problem here is that stores sell e-bikes that can drive faster than legal limits, then where’s the story about how literally *every car dealership has been doing the same thing for decades*?!?

cbc.caHidden cameras show some stores unlocking speed limits on e-bikes, selling e-scooters that go up to 120 km/h | CBC NewsA new hidden camera investigation by Marketplace shows how some stores are selling e-scooters and e-bikes with top speeds much higher than provincial speed limits and with recommended rider ages young...
Joshua Loftus
@joft.bsky.social

You’re laughing but this is something a hidden internal portion of all AI models has to relearn from first principles every time you ask them to “run” some code and it costs billions of dollars in wasted gpu cycles

Randall Munroe@xkcd.com

Ground Effect xkcd.com/3305/

Joshua Loftus
@joft.bsky.social

My conspiracy theory for why the "AI" CEOs are suddenly all in agreement about slowing down is that they already expect model improvement to plateau and want to say it's voluntary rather than hitting hard technological/statistical limits

Joshua Loftus
@joft.bsky.social

The old world is dying, and the new world struggles to be born: now is the time of monsters… extremely cute monsters that culturally evolved to hijack our baby-preserving psychology

This post is unavailable.

Joshua Loftus reposted

the fool (cachinnatory)
@agnoster.net

the future could be magnificent, and you deserve a share of it

Doll@dollspace.gay

A left that could say “the future could be magnificent, and you deserve a share of it” was offering something people could want. Losing that voice to endless fights over whether wanting the future was itself morally suspect was a hell of a loss.

Joshua Loftus
@joft.bsky.social

In half of the elections I will vote for whoever wants to drone strike migrant caravans, and in the other half I will vote for whoever promises to prosecute ICE agents. There are ~5 million other Americans exactly like me. We decide the outcomes of every election and there is no way to contact us

Joshua Loftus
@joft.bsky.social
Vince Mancini@vincemancini.bsky.social

The guy who dined and dash, ran into the street, and then got hit by a car, whose driver then pulled a hit-and-run, is my new favorite local story of maybe ever. www.yourcentralvalley.com/news/crime/f...

Joshua Loftus
@joft.bsky.social

This is a good explanation for why my toddler is able to say "dowwwwnnn" with so much feeling

Peter@notalawyer.bsky.social

gotta feel incredible to hit ‘em with your first “no” after a lifetime of mandatory acquiescence