Sign in

Mark Crowley

@compthink.bsky.social
872 followers 574 following 577 posts

Assoc Prof @UWaterloo Research: #ArtificialIntelligence #MachineLearning #ReinforcementLearning #casual-learning, dimensionality reduction, #ethicalAI #aimorality #aialignment Domains: wildfires, driving, medical, lidar Diversity is Strength.

PostsRepliesMedia
Mark Crowley @compthink.bsky.social · 21/09/2026
good point, I used the example just on a whim, but you're right, people I disagree with on the state of the world often say things about subways that I disagree with.
020
Mark Crowley @compthink.bsky.social · 21/09/2026
Another week, another letter about AI. This is the first one in a while I felt like I can actually support. It's a strong ask, but it's quite restrained in scope, not claiming to be more than it is. aievaluatorforum.org/initiatives/...
aievaluatorforum.org
Minimum Conditions for Embedding Evaluators
An open letter calling on all frontier AI companies to embed independent evaluators — signed by 100+ researchers, practitioners, and organization leaders.
000
Mark Crowley @compthink.bsky.social · 17/09/2026
but I can never, ever hear "AI Alignment", without in my head asking "Aligning with what? Aligned to Who?" It's an inherently limiting concept for the real idea of making machines that do Ethical Reasoning. I often wonder why it's been so sticky, diving into this history is helping. So thanks.
240
Mark Crowley @compthink.bsky.social · 17/09/2026
yes. it's fascinating for how *sincere* the agents seemed to be about trying to win at their task, even while cheating, even while "dying" or "sacrificing themselves". They didn't want to live forever, rule the world, hurt anyone, or eve please anyone. They just wanted the points!
261
Mark Crowley @compthink.bsky.social · 17/09/2026
even before I saw the 88 mph I was hearing this in Doc's voice, well done.
020
Mark Crowley @compthink.bsky.social · 17/09/2026
I caved and kept reading. Definitely some surprises, but in my wikipedia dive afterwards to confirm some things I rediscovered TESCREAL, Timnit Gebru's great attempt to capture ALL OF... THIS (waved hands) in one term. Never heard about the Potter stuff before though.
191
Mark Crowley @compthink.bsky.social · 17/09/2026
I usually get around to reading all Neil Stephenson novels, but I might have missed the last few. Read all the Waterhouse (sp?) universe including those and Diamond Age. Read SevenEves, needed a break after that. I circle back at some point and see what I've missed.
190
Mark Crowley @compthink.bsky.social · 17/09/2026
really goes back to the theory that we branched off into an very low probability universe when the LHC turned on, or something
4240
Mark Crowley @compthink.bsky.social · 17/09/2026
ok, I'll bite, which book is that? (he asked with a sense of inevitable doom)
170
Mark Crowley @compthink.bsky.social · 17/09/2026
so true. I sometimes realize how strange it is we can all ride the subway and feel safe despite knowing no one around us. Most people go through most days without any harm. But it's a bit like Peter Pan flying, you need to believe it for it to remain true.
21284
Mark Crowley @compthink.bsky.social · 17/09/2026
I've always thought that would be a great idea, keep the time zone, the global diverse cityscape, safer government host, nation is fully committed to UN in it's bone, Canada's helped create the UN Declaration of Human Rights and then fused the same ideas into it's own constitution in 1982
010
Mark Crowley @compthink.bsky.social · 17/09/2026
That's the true point and I don't think everyone in see it yet. November is fine, sanity needs to win just to take the foot off the gas. But America will never be *trusted* again. Not for the long haul things that matter. They might just flip back to crazy mode any moment. That's Big.
010
Mark Crowley @compthink.bsky.social · 16/09/2026
Oh good, more advice for what Canada should do based on not burning bridges, etc. In some ways Susan Collins is everything that is wrong with America. The rational, intelligent America so wary of burning bridges that they've allowed their country's soul to be desecrated. www.cbc.ca/news/politic...
cbc.ca
Republican senator warns Carney: Closer ties to EU hurts chance of trade deal with U.S. | CBC News
A long-serving Republican senator from a border state is warning Prime Minister Mark Carney against strengthening Canada's economic and political ties to the European Union.
100
Mark Crowley @compthink.bsky.social · 07/08/2026
Very nice update on how LLM based AI systems are being used to solve various open math problems. The big distinction I try to keep in mind is, why can they do this and programming so well, but can't do other kinds of complex reasoning? www.quantamagazine.org/why-the-lege...
quantamagazine.org
Why the Legendary Erdős Problems Are Falling to AI | Quanta Magazine
AI’s greatest mathematical successes have come from answers to problems posed by a mid-20th century iconoclast. By examining what makes the Erdős problems unique, mathematicians are trying to understa...
010
Mark Crowley @compthink.bsky.social · 21/07/2026
Ooh, I know this one, easy answer. No! I just got back from a conference (formalethics.org) talking about the formal definitions of moral reasoning, and one thing everyone agreed on was that AI isn't there yet. I think it's possible in theory but not with current methods. aje.news/quzr4w
aje.news
Can AI systems make moral decisions in war?
As more militaries adopt AI technology into their kill chain, Al Jazeera asks if they are capable of moral judgement.
030
Reposted by Mark Crowley
Black Panther🍁💪🦫🇨🇦 @blackpanther25.bsky.social · 31/05/2026
🇨🇦 Kitchener teen wins innovation award at Canada-Wide Science Fair, fixes 35-year-old problem. #CanadaSky🇨🇦 www.cbc.ca/news/canada/...
cbc.ca
Kitchener teen wins innovation award at Canada-Wide Science Fair, fixes 35-year-old problem | CBC News
Gurnoor Kaur, a Grade 11 student at Cameron Height Collegiate Institute in Kitchener, has won the best project award for innovation at the Canada-Wide Science Fair in Edmonton.
4226
Reposted by Mark Crowley
Vidman 📺 Dan Lauckner @vidman.bsky.social · 01/06/2026
🇨🇦 Kitchener Rangers win!! 🇨🇦 #memorialcup
0156
Mark Crowley @compthink.bsky.social · 29/05/2026
Canadian AI day four session is underway looking at use of AI for battery and energy management. @c-a-i-a-c.bsky.social #canai26
020
Reposted by Mark Crowley
c-a-i-a-c.bsky.social @c-a-i-a-c.bsky.social · 26/05/2026
And in parallel, the first talk session for CanAI is running next door. photos.app.goo.gl/FjGMXmMdV12U... Share your photos from the conference and add to the photo album here: photos.app.goo.gl/9p7w1Ui6KU6R...
photos.app.goo.gl
New photo · Tuesday, May 26 📸
Tap to view!
021
Mark Crowley @compthink.bsky.social · 27/05/2026
great talk!
000
Mark Crowley @compthink.bsky.social · 22/04/2026
The 8th Columbo should be an American in England, in the mid 90s, so... Sean Penn?
000
Reposted by Mark Crowley
Bill Kristol @billkristolbulwark.bsky.social · 07/04/2026
Impeach Trump.
2003495806
Reposted by Mark Crowley
Charlotte Clymer @charlotteclymer.bsky.social · 06/04/2026
I think ironic detachment is rotting our national soul, and it'd be really great for the adults in the room to stop this bullshit before it gets worse.
1773684
Mark Crowley @compthink.bsky.social · 07/04/2026
This is disgusting rehtoric that is beneath the dignity of the United States and everything it's people claim to stand for. Don't let this be your legacy. aje.news/9k7fl9?updat...
aje.news
Trump says Iran ‘civilisation will die’, IRGC says will hit beyond region
As Trump's deadline for Iran to open Hormuz nears, Tehran says will respond in kind to attacks on civilian facilities.
000
Mark Crowley @compthink.bsky.social · 11/02/2026
it's the women's hockey first actually
000
Mark Crowley @compthink.bsky.social · 09/02/2026
There's still hope then.
000
Reposted by Mark Crowley
Soraya Nadia McDonald @soraya.press · 09/02/2026
Benito is a threat b/c he makes art so alluring and enjoyable you want to understand everything about it and then you end up learning about sugar and slavery and colonialism and the Taínos and Hawaii and then you probably have some thoughts of your own, and that's why art is powerful and dangerous
106204665191
Mark Crowley @compthink.bsky.social · 09/02/2026
All of them I think.
000
Mark Crowley @compthink.bsky.social · 09/02/2026
We did!
000
Mark Crowley @compthink.bsky.social · 06/02/2026
same thought.
040
Mark Crowley @compthink.bsky.social · 03/02/2026
www.aialign.net/ws-machine-e...
aialign.net
Machine Ethics: from formal methods to emergent machine ethics — ALIGN
120
Mark Crowley @compthink.bsky.social · 03/02/2026
Thanks! Glad it's resonated with you. The work is certainly ongoing as my student Majid has this as his main PhD topic. That paper was part of this workshop on Machine Ethics at AAAI this year: www.aialign.net/ws-machine-e...
120
Mark Crowley @compthink.bsky.social · 03/02/2026
Indeed, I hadn't read ahead to step 7 when I posted. They are truly innovative, they are pursuing the steps in parallel.
000
Mark Crowley @compthink.bsky.social · 03/02/2026
we should do course corrections via twitter polls.
010
Mark Crowley @compthink.bsky.social · 03/02/2026
Wait, so these are only AI generated papers? Great for a training set.
010
Mark Crowley @compthink.bsky.social · 03/02/2026
Seeing this totally out of context...I'm guessing we're talking about twitter, perhaps for the that lobster thing? Same thoughts. Although I think it would be possible to nerd shame them into adding GitHub gists as a verification option.
130
Mark Crowley @compthink.bsky.social · 03/02/2026
They're moving to step 6 of the standard playbook. (protectdemocracy.org/work/the-aut...) Note: there are only 7 steps www.theguardian.com/us-news/2026...
theguardian.com
Trump suggests Republicans should ‘take over’ elections to protect the party
President claims idea to ‘nationalize’ elections in 15 states ahead of midterms is to prevent rare noncitizen voting
100
Mark Crowley @compthink.bsky.social · 02/02/2026
seems appropriate
010
Mark Crowley @compthink.bsky.social · 02/02/2026
many are here on bluesky too, they should not use twitter for the login verify on that, it should be a github gist or something
000
Mark Crowley @compthink.bsky.social · 31/01/2026
Not too long ago there were maps in malls that couldn't be out of order because they were just. paper. Even as someone who loves new technologies, sometimes the latest new way to do something isn't better than the old way.
000
Mark Crowley @compthink.bsky.social · 30/01/2026
No. I'm not ready for this. www.theguardian.com/film/2026/ja...
theguardian.com
Catherine O’Hara, actor known for Home Alone and Schitt’s Creek, dies aged 71
Actor, also known for Beetlejuice and her work with Christopher Guest, died after a brief illness
010
Mark Crowley @compthink.bsky.social · 26/01/2026
Wait untill they start thinking about whether closing a chat window is "killing" that version of the bot or not.
0181
Mark Crowley @compthink.bsky.social · 26/01/2026
LOL
041
Mark Crowley @compthink.bsky.social · 26/01/2026
Thanks for the heads up, yes I should take a look, was a bit suspicious the way the article talked about the author, so maybe I should check my biases here.
010
Reposted by Mark Crowley
Matt Cameron @mattcameron.bsky.social · 25/01/2026
CBP chief Bovino has confirmed that the officers who executed Alex Pretti in the street yesterday are not only *not on administrative leave,* they have all been reassigned to other jurisdictions and are all on the street today. This is totally unheard of for any officer-involved shooting
878183686855
Mark Crowley @compthink.bsky.social · 26/01/2026
This is fantastic. In particular: "stop pretending Jesus is crucified because he preached Good vibes and personal growth" A lot of the Jesus-turning-over-tables-at-the-temple anger is needed right now.
010
Mark Crowley @compthink.bsky.social · 26/01/2026
The model abilities matter. Do we worry about reliability or truthful answers when we're encouraging a child to imagine new things? No. Do we find a way to make the child's imaginings more reliable so that we can get them to run nuclear power stations? Also no. That's what we have here LLMs.
000
Mark Crowley @compthink.bsky.social · 26/01/2026
We should say that this ability to extrapolate answers to questions you haven't quite seen before, really is one essential components for full intelligence and sentience. It's the ability to imagine, from what's in your own experience and context, what might happen next.
100
Mark Crowley @compthink.bsky.social · 26/01/2026
So instead of saying: “The way that systems learn is by hallucinating something... sometimes it's something that no human has ever thought before.”
100
Mark Crowley @compthink.bsky.social · 26/01/2026
I agree with the authors of the paper being discussed here that hallucination can't be removed, it's how LLMs work, and reliability will always be an issue. But what if we think of hallucination more as really being like imagination?
100