Alexander Clark (@alexandersclark) profile picture

Linguist, philosopher, computer scientist, foodie, wine geek etc. I follow some people I disagree with. Don’t interpret my likes please.

London
Joined February 2011
Photo by Alexander Clark (@alexandersclark)Photo by Alexander Clark (@alexandersclark)Photo by Alexander Clark (@alexandersclark)Photo by Alexander Clark (@alexandersclark)Photo by Alexander Clark (@alexandersclark)Photo by Alexander Clark (@alexandersclark)Photo by Alexander Clark (@alexandersclark)Photo by Alexander Clark (@alexandersclark)Photo by Alexander Clark (@alexandersclark)Photo by Alexander Clark (@alexandersclark)
This paper has been a long time cooking, but it's out now in TACL, with Nathanaël Fijalkow. Pretty much the first nontrivial (consistent) algorithm for learning Probabilistic Context-Free Grammars: doi.org/10.1162/tacl_a_00323 1/5
1
8
1
58
The same discourse here underlies the MPs expense scandal.
So here's an unpopular opinion. This stuff is a facet of what drives Congressional corruption, but not in the way most people think. We are the richest, most powerful country in the world and we expect our 535 Legislators to fly Economy class when they're on official business out of some warped sense of imposed humility. When you treat people with immense power like they're not immensely powerful you breed corruption. Police Departments (under the William Parker model) figured this out a long time ago. They started paying police officers like middle class professionals and told them things like "if I catch you taking free coffee, you're getting written up". The observation Parker made was that cops are powerful people in their communities, but police departments paid them like shit. If you paid them something more commiserate with their power and authority, they would be less likely you utilize that power and authority to make up the gap. It was commonly understood by everyone that cops would use their power to secure little favors by performing small, innocuous acts of corruption. You'd make sure to go by the diner that always gave you free coffee after hours to tug on the back door and make sure none of the usual suspects were loitering around. You'd give a warning instead of a ticket to the hotdog cart guy who gave guys on the job a free hotdog . You'd take a drunk teenager home instead of the drunk tank because his dad ran a grocer and always made sure to hold something in short supply aside for your wife. These tiny acts of corruption, little favors, weren't a big deal on the individual level and helped build community relations. In the aggregate, though, it gave a certain class of the community privileged access to law enforcement and built up a general tolerance to corruption in everyone who participated in it. By paying his cops well and demanding they not take favors from the community or use their authority to secure preferential treatment, Parker drastically reduced instances of serious corruption in the LAPD. I know a lot of you will object to this and say "but cops are still corrupt" and yes, some of them are. But it used to be so, so much worse than most people realize. We've become accustomed to police as professional civil servants and blanche at behavior that used to be commonly accepted as the "cost of doing business". We need to stop pretending the members of Congress aren't powerful people and start treating them like they are. You start tackling corruption by acknowledging and compensating power, not pretending it doesn't exist.
109
PSA: Don’t click on this if you are in the UK
wow this post is blowing up. be sure to check out my nudes at academia.edu
1
164
Alexander Clark retweeted
The saddest thing isn't my own slop reviews, it's how negative human reviews are getting for truly novel work. Good faith reviewers are fatigued by vapid slop papers, and can't gather the energy to attempt to understand a new idea.
A paper I was very excited about (might be the most practically impactful work I've done) got 2/2/3/2. Pangram flagged 3/4 as 100% AI generated. (The shortest is short and mostly AI generated.) I hate that we likely can't afford to ban these reviewers from refereeing.
4
9
2
159
6,874
Alexander Clark retweeted
What does it take to learn language from everyday experience? Our new study, published today in Science, offers a new way to answer this question.
3
67
4
189
10,526
Alexander Clark retweeted
For anyone here interested in joining a Discord server for the *intersection* of NLP and linguistics (including psycholinguistics), or anything Computational Linguistics: discord.gg/YvzHrwrqp (Please BOOST!) #linguistics #psycholinguistics #NLP #LLM
2
5
2
20
1,943
I always get in trouble when I claim “galore” is a post nominal determiner in English.
Thinking about things like "notary public," "body politic," phrases where English puts the modifier after the noun like a Romance language. There must be some other examples
3
1
8
364
Alexander Clark retweeted
Please do not submit LLM-written papers to TACL! I appreciate many people (even high-school students) have success doing this with ARR, Neurips, and other mega-conf. But TACL reviewing is still almost all done by people, and LLM/AI slop is very unlikely to be accepted.
2
10
1
115
15,591
Alexander Clark retweeted
Justin Moore on OpenAI’s proof of nonamenability of F justintatchmoore.github.io/h…
7
21
2
166
17,386
Something that gets lost in housing discourse is the difference between IRL vs online NIMBYs IRL NIMBYS tend to be old people who hate poors, traffic, & change. Online NIMBYs are usually leftists with weird beefs. They think YIMBYs are annoying and supply/demand isn't real.
20
55
1
1,178
15,454
Alexander Clark retweeted
Now that's a real speedup! Already 1% faster than the standard O(n log n) alg with n as small as 10^600000.
We’re publishing another update to OpenAI Problem #130: computing the exact discrete Fourier transform below n log n. Our paper now incorporates the all-length bound: T(n) = O(n(log n)^(1−δ)), with δ = 0.0007547360. That’s 10.34 times the exponent saving of δ = 7.3×10⁻⁵ in our previous announcement. The stronger Fourier result comes from Jacob Sussman’s five-stage framework and Chafik Boukhalfa’s improved construction, building on @OpenAI, @0xdoug, @RohanArun, @SJ_Swapnil_Jain and the wider community.
10
25
4
966
65,965
Alexander Clark retweeted
My dream had always been to prove beautiful theorems as the AI agents unloaded my dishwasher. Well, it went the wrong way... 😂
40
309
38
2,863
140,432
My hot take on current AI CS/Maths results, eg 3sum, is that they indicate that classical asymptotic worst case complexity is fundamentally flawed.
4
161
I feel like the France German yield spread should be a bigger story .
117
Alexander Clark retweeted
Guy who thinks the Singularity Is Near but it's really important we fix housing policy before then
32
33
20
830
Chomskyan linguistics has yet to grapple with this problematic fact.
Replying to @Slatzism
we are losing our last remaining novel english sentences with alarming rapidity
1
1
6
606
Alexander Clark retweeted
I'm currently returning to Toronto from a summit on the future of mathematics, at OpenAI. @SebastienBubeck asked me to talk a bit about the future we'd all like to avoid, where humans are mathematically disempowered. @Jacob_Tsimerman advised us to try to prioritize detail over correctness, and I have no doubt that I succeeded in deprioritizing correctness. I tried to find a title that wasn't too bombastic:
63
230
44
1,406
335,300
This is well worth a read :
What should we make of recent language model advancements in mathematics? In my new post, I reflect on long-standing cognitive debates about symbols and neural networks in light of this progress. (Link below)
1
1
176
Alexander Clark retweeted
In this interview, Verkerk and Gray answer some of my questions about their amazing 2026 paper, in which they show that many of the Greenbergian universals are robustly supported by the Grambank data. Thanks to Russell and Annemarie! dlc.hypotheses.org/4383
7
11
1,052
Alexander Clark retweeted
We chop and change Prime Ministers every two years because no one’s willing to accept the reality of the position the country finds itself in, and it becomes easy cope to keep blaming a central figure.
131
1,277
83
16,450
918,789