Farhan Samir's Avatar

Farhan Samir

@smfsamir.bsky.social

Multilingual NLP x Social Science x Phonetics at UBC. Previously Lab126, NVIDIA, Ai2, UW. Check out my work at https://farhansamir.notion.site/samir.

209 Followers  |  232 Following  |  103 Posts  |  Joined: 23.11.2024  |  1.7248

Latest posts by smfsamir.bsky.social on Bluesky

Preview
The Irrational Decision How the computer revolution shaped our conception of rationalityβ€”and why human problems require solutions rooted in human intuition, morality, and judgment

I’m excited to announce that my new book, _The Irrational Decision_, is now available for pre-order from Princeton University Press.

04.08.2025 14:31 β€” πŸ‘ 92    πŸ” 27    πŸ’¬ 7    πŸ“Œ 7

love the high density of water fountains in Zurich, i am unbelievably hydrated

04.08.2025 15:41 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Workday is the most abominable software I've ever used in my life

04.08.2025 12:15 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Preview
The Real Demon Inside ChatGPT AI chatbots strip language of its historical and cultural context. Sometimes what looks like a satanic bloodletting ritual may actually be lifted from Warhammer 40,000.

ChatGPT and similar programs weren’t just trained on the internetβ€”they were trained on specific pieces of information presented in specific contexts.

When stripped of that context, the bots can confuse and alarm users, or even dangerously mislead them.

02.08.2025 20:53 β€” πŸ‘ 406    πŸ” 140    πŸ’¬ 15    πŸ“Œ 20
Preview
The Income Gap In Canada Has Reached A Record High The wealthiest amass more through higher salaries and passive income, while workers face stagnating wages.

NEW: The income gap in Canada has reached a record high.

In 2025 Q1, the top 20% of households held 64.7% of all wealth, while the bottom 40% accounted for just 3.3%.

Read the latest issue of Class Struggle from Adam King.

www.readthemaple.com/the-income-g...

30.07.2025 12:30 β€” πŸ‘ 115    πŸ” 45    πŸ’¬ 5    πŸ“Œ 5
Preview
A Comparative Approach for Auditing Multilingual Phonetic Transcript Archives Abstract. Curating datasets that span multiple languages is challenging. To make the collection more scalable, researchers often incorporate one or more imperfect classifiers in the process, like lang...

Presenting a TACL paper on strong limitations of universal speech recognition models and datasets, at #ACL2025, on *Wed. 11-12:30*. Pls come hear me out on how speech, as a hugely varying cultural practice, inherently resists the sort of large-scale datafication that's needed for machine learning

29.07.2025 14:32 β€” πŸ‘ 7    πŸ” 4    πŸ’¬ 0    πŸ“Œ 0

AI researchers (and literally everyone else) should stop pretending that value neutral is a real stance you can take.

29.07.2025 07:51 β€” πŸ‘ 8    πŸ” 1    πŸ’¬ 0    πŸ“Œ 0

I feel so defeated, there is no mainstream opposition to these devastating cuts

29.07.2025 03:42 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 1

It takes only a few minutes & and you can choose other. I used to work in the federal government. These consultations are important and submissions do get read. Plus the opposition and journalists can request to see the results, so even if the government ignores the public will, it'll still get out.

27.07.2025 19:49 β€” πŸ‘ 18    πŸ” 14    πŸ’¬ 0    πŸ“Œ 1

Is it too much to ask for a keynote speaker that isn’t a big tech employee. I think this is a reasonable ask

27.07.2025 04:49 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

I still don’t understand why the Canadian electorate gets so excited about building cross-country oil pipelines, you’re never going to see any of that money you fools

26.07.2025 22:14 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Preview
Special Committee on Democratic and Electoral Reform - BC Green Caucus In Summer 2025 the British Columbia Legislature is hosting an All Party Special Committee on Democratic and Electoral Reform, which includes public engagement. We want YOU to participate!

BCers, you have until 2 pm today to tell this electoral reform committee to get rid of first past the post. Or to keep it, if you're some kind of weirdo. Like, the bad kind of weirdo.

Submitting is easy, you just have to make an account and then write your comment in a text box. That's it.

25.07.2025 15:54 β€” πŸ‘ 36    πŸ” 22    πŸ’¬ 5    πŸ“Œ 0

Has there been any prominent use of LM-based fact-checking methods, other than the LA Times β€œbias meter”?

22.07.2025 08:45 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

In NorrkΓΆping this week for #IC2S2. I’m giving a talk on large-scale content differences between different Wikipedia language editions. Tuesday @ 11AM at the Wikipedia session, come by!

21.07.2025 22:26 β€” πŸ‘ 4    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Anyone else going to IC2S2 next week??

19.07.2025 01:08 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Research grants to universities & colleges as well as graduate scholarships face $764 million in cuts to pay for tax cuts and miliary spending. The Tri-council agencies providing those funds just won an increase in the last budget, but that will be completely undercut now. 6/x

17.07.2025 14:03 β€” πŸ‘ 56    πŸ” 24    πŸ’¬ 3    πŸ“Œ 23
Preview
Axon’s Draft One is Designed to Defy Transparency Axon Enterprise’s Draft One β€” a generative artificial intelligence product that writes police reports based on audio from officers’ body-worn cameras β€” seems deliberately designed to avoid audits that...

ICYMI: Did you know that police are starting to feed body-worn camera audio into generative AI to write police reports? We investigated Axon's Draft One and found there's no way of knowing which parts of final reports were written by AI and which parts were written by humans.

14.07.2025 19:17 β€” πŸ‘ 135    πŸ” 69    πŸ’¬ 6    πŸ“Œ 22

Relatedly,

bsky.app/profile/smfs...

14.07.2025 22:22 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

The first PM's Wikipedia page is more than 10K words long, yet makes no mention of his near-religious devotion to eugenics and how it shaped confederacy. Wikipedia is abysmal for anything related to race and gender politics (most things)

en.wikipedia.org/wiki/John_A....

14.07.2025 22:22 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
Pre-Budget Consultations 2025 - Canada.ca Pre-Budget Consultations 2025

The government of Canada has a questionnaire about it's new budget. Most of the given options are conservative (because this is a conservative government) but there are some that aren't, and you can write in your own responses.
www.canada.ca/en/departmen...

14.07.2025 18:09 β€” πŸ‘ 108    πŸ” 69    πŸ’¬ 11    πŸ“Œ 11

just saw a neurips paper abstract that says "stepping stones along a path that unfolds into endless innovation", what is going on over there

11.07.2025 23:40 β€” πŸ‘ 3    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
A paper by Peng et al. (2024): OWSM v3.1: Better and Faster Open Whisper-Style Speech Models based on E-Branchformer

A paper by Peng et al. (2024): OWSM v3.1: Better and Faster Open Whisper-Style Speech Models based on E-Branchformer

I've found OWSM to be at least as good as Whisper. And straightforward to set up: www.wavlab.org/activities/2...

10.07.2025 20:33 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

one can only hope, but i’m not convinced it will burst since governments keep bailing them out in all sorts of ways, including but not limited to giving them big bags of taxpayer money

10.07.2025 02:43 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0
clap-ipa/example.ipynb at main Β· lingjzhu/clap-ipa Keyword spotting and forced alignment in any language - lingjzhu/clap-ipa

In a NAACL'24 paper, we described a new multilingual IPA-based keyword search model. Basically, you can specify 1. some utterance using IPA, e.g., /bo:do:/, and 2. a speech recording, and identify where the utterance is spoken in the recording. See a simple example here: github.com/lingjzhu/cla...

09.07.2025 18:38 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Excerpt from Weizenbaum’s Computer Power and Human Reason (1975): β€œFurther work must be done before the program will be ready for clinical use. If the method proves beneficial, then it would provide a therapeutic tool which can be made widely available to mental hospitals and psychiatric centers suffering a shortage of therapists. Because or time-sharing capabilities of modern and future computers, several hundred patients an hour could be handled by a computer system designed for this purpose. The human therapist, involved in the design and operation of this system, could not be replaced, but would become a much more efficient man since his efforts would no longer be limited to the one-to-one patient-therapist ratio as now exists.”

Excerpt from Weizenbaum’s Computer Power and Human Reason (1975): β€œFurther work must be done before the program will be ready for clinical use. If the method proves beneficial, then it would provide a therapeutic tool which can be made widely available to mental hospitals and psychiatric centers suffering a shortage of therapists. Because or time-sharing capabilities of modern and future computers, several hundred patients an hour could be handled by a computer system designed for this purpose. The human therapist, involved in the design and operation of this system, could not be replaced, but would become a much more efficient man since his efforts would no longer be limited to the one-to-one patient-therapist ratio as now exists.”

Weizenbaum (1976). Sounds oddly familiar

09.07.2025 05:03 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
09.07.2025 01:46 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

unfortunately my career choices were contingent on the federal government not deliberately torpedoing itself

08.07.2025 21:18 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Preview
Mark Carney’s cabinet told to present savings plans by the end of the summer Finance Minister FranΓ§ois-Philippe Champagne, along with Treasury Board President Shafqat Ali, issued letters to Prime Minister Mark Carney’s cabinet on Monday, asking them to present plans by the end...

Carney is working towards cutting 30% of the federal government's budget. I had low expectations for a private equity guy but this is cataclysmic

www.ctvnews.ca/politics/art...

08.07.2025 19:02 β€” πŸ‘ 6    πŸ” 2    πŸ’¬ 3    πŸ“Œ 0

I think we have trained enough language models

08.07.2025 06:27 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Airlines giving you the ~~choice~~ to bid for an upgrade feels like it probably shouldnt be a legal practice. Like they’re clearly taking advantage of information asymmetry to exploit consumers

07.07.2025 04:53 β€” πŸ‘ 3    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

@smfsamir is following 20 prominent accounts