Shan Kothari
banner
shankothari.bsky.social
Shan Kothari
@shankothari.bsky.social

Assistant professor at University of Alberta. Forest ecology + tree physiology. he/il

Environmental science 59%
Agriculture 16%

early evidence suggests that Tesla (unlike Waymo) is less safe than humans, even with a safety monitor present sherwood.news/tech/teslas-...
Tesla’s 29 Austin Robotaxis have crashed 8 times since June, as data suggests they perform much worse than human drivers
That’s a lot of crashes for such a small fleet....
sherwood.news

are your recent test posts related to 3?

also: corn ethanol
This story is absolutely wild. Did you know that avocados change sex over the course of a day? And that it's controlled by a single ancient balanced polymorphism? This is flat our crazy
Balanced polymorphism in a floral transcription factor underlies an ancient rhythm of daily sex alternation in avocado https://www.biorxiv.org/content/10.64898/2025.12.22.695989v1
Balanced polymorphism in a floral transcription factor underlies an ancient rhythm of daily sex alternation in avocado https://www.biorxiv.org/content/10.64898/2025.12.22.695989v1

and while sycophancy probably derives in large part from RLHF, all models tend to be weirdly pliable in ways that humans are usually not

don't you think they would at least make them adversarially robust if they could, though? it's pretty tough to jailbreak most humans!

this is not a complete answer, but GOFAI ("good old-fashioned AI") is by convention part of AI but not really ML

Reposted by Shan Kothari

there’s now a petition for the canadian government to overturn the recent legislation in AB and SK designed to strip trans youth of their rights. please sign: www.ourcommons.ca/petitions/en...
Sign this Petition - Petitions
www.ourcommons.ca

It's gotta be >80% even in wealthy anglophone countries

Reposted by Shan Kothari

I think there’s some unproductive AI-Luddism (colloquial usage, you know what I mean, don’t be annoying), but also it’s extremely obvious where it came from and AI companies share a large portion of the blame with how they’ve framed things.

Reposted by Shan Kothari

Since I can’t get it out of my head, I wrote up my thoughts on @kevinbaker.bsky.social's critique of AI-automated science and the logical end of processes that can't self-correct.

I got the same request. I filled it out, and I was glad to do so, but the LLM-generated hypotheses were really quite bad.
1/2 I was asked to review an AI summary of a paper I'd published. It was part of a research project into the application of AI to research. The summary was not particularly good summary of the introduction and abstract. I was then asked to review 5 hypotheses generated by AI.
#AIscience
this, from @kevinbaker.bsky.social, is a better analysis of the intersection between LLMs and academic science than 98% of what's out there.
Context Widows
or, of GPUs, LPUs, and Goal Displacement
artificialbureaucracy.substack.com

people often forget that both VICE and Len are Canadian!
So, El Fasher looks like one of the worst single event atrocities to happen this century so far, but the evidence has come in such a trickle it seems to only now be getting verified
At least 60,000 murdered in Sudanese city, which resembles ‘a slaughterhouse’
Satellite evidence shows extent of paramilitary massacre in El Fasher
www.irishtimes.com

Reposted by Shan Kothari

1/2 I was asked to review an AI summary of a paper I'd published. It was part of a research project into the application of AI to research. The summary was not particularly good summary of the introduction and abstract. I was then asked to review 5 hypotheses generated by AI.
#AIscience

this is also how I explain Toronto to non-Canadians
The RSF just attacked a kindergarten, killing at least 33 children in Sudan.

Meanwhile, the Trump Admin. continues to sell arms to the UAE, who are aiding these atrocities.

Imagine if Trump spent as much time trying to stop the killing, as he did getting the UAE to invest in his crypto coin.
Sudanese paramilitary drone attack kills 50, including 33 children in Kordofan, doctor group says
A drone attack by Sudanese paramilitary forces has hit a kindergarten in south-central Sudan, killing 50 people, including 33 children, according to a doctors’ group
www.washingtonpost.com

as a University of Alberta prof who's not a computer scientist, I'm amused to see Alberta used as a watchword for anything at all

is LL the other one?

Reposted by Shan Kothari

No one of you is going to mistake this short column of mine for trenchant economic analysis. Nevertheless, here’s my attempt to explain to a popular audience why I think they should join the “raise my taxes” movement:

www.christiancourier.ca/raise-my-tax...

Reposted by Shan Kothari

Quality of life in Gaza has not improved since the "ceasefire". Most are still displaced and without food. Many are still dodging violence from Israeli attacks. Aid is still not being allowed in at anywhere near the levels required to help the population.

www.theguardian.com/global-devel...
Fundraisers warn of ‘catastrophic’ drop in donations to Gaza since ceasefire
‘The world thinks Palestinians don’t need help any more,’ aid organiser says, despite desperate need as winter nears
www.theguardian.com

I suspect that as long as there are human drivers, some of them will be tired, distracted, angry, or drunk

Reposted by Shan Kothari

my interview for the Digital Theory Lab with Karen Hao @karenhao.bsky.social at the Remarque Institute
www.youtube.com/watch?v=iZpH...
Karen Hao on Empire of AI
YouTube video by Remarque NYU
www.youtube.com
Almost everywhere in the City of Toronto has fewer people that it did 50 years ago

every once in a while, just for fun, I take a passage I admire and write a paragraph describing its contents and tone to produce an LLM pastiche. the results often impress on a first read-through—which again, I think reflects a mastery of affect—but after that I start to notice everything that's off

often the first thing (sometimes the only thing) people appreciate about a poem is the affect it cultivates, and I've seen LLMs do a great job at replicating that, at least within some limited range! but beyond that I haven't been convinced so far