From 7ba592542e2c11d8c112bc14c3d6e05ee294ef8b Mon Sep 17 00:00:00 2001 From: k Date: Wed, 30 Sep 2026 14:43:37 -0400 Subject: [PATCH 1/7] style: remove whitespace after sidenotes The {{- "" -}} ensures the trailing newline is removed when using the sidenote shortcode. --- layouts/shortcodes/sidenote.html | 1 + 1 file changed, 1 insertion(+) diff --git a/layouts/shortcodes/sidenote.html b/layouts/shortcodes/sidenote.html index dc09c908e..49429139a 100644 --- a/layouts/shortcodes/sidenote.html +++ b/layouts/shortcodes/sidenote.html @@ -1 +1,2 @@ {{ .Inner }} +{{- "" -}} From 362927f6149a0dd334ac38efc3ac8694183f2380 Mon Sep 17 00:00:00 2001 From: k Date: Wed, 30 Sep 2026 14:46:01 -0400 Subject: [PATCH 2/7] chore: use Hugo locales instead of languageCode --- config.toml | 2 +- layouts/alias.html | 2 +- layouts/partials/header.html | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) diff --git a/config.toml b/config.toml index 9b1d5ca65..76d13f777 100644 --- a/config.toml +++ b/config.toml @@ -1,5 +1,5 @@ baseURL = "https://kylrth.com/" -languageCode = "en-us" +locale = "en-us" title = "Kyle Roth" pagination.pagerSize = 5 enableGitInfo = true diff --git a/layouts/alias.html b/layouts/alias.html index 1b61e95f6..6aee5c2b2 100644 --- a/layouts/alias.html +++ b/layouts/alias.html @@ -1,5 +1,5 @@ - + {{ .Permalink | relLangURL }} diff --git a/layouts/partials/header.html b/layouts/partials/header.html index d25020fda..57f7c41a2 100755 --- a/layouts/partials/header.html +++ b/layouts/partials/header.html @@ -1,5 +1,5 @@ - + {{ partial "head.html" . }} From 2240a5ef45adc0b150ec63c71bbb6af2507cb3a1 Mon Sep 17 00:00:00 2001 From: k Date: Wed, 30 Sep 2026 14:49:13 -0400 Subject: [PATCH 3/7] linter: stop checking links Post-LLM it has become unsustainable to run automated URL visits to check for dead links. --- .github/workflows/deploy.yml | 2 +- .htmltest.yml | 37 ++++-------------------------------- Makefile | 10 +++++++--- 3 files changed, 12 insertions(+), 37 deletions(-) diff --git a/.github/workflows/deploy.yml b/.github/workflows/deploy.yml index ab3ca3ac7..4fd67a99f 100644 --- a/.github/workflows/deploy.yml +++ b/.github/workflows/deploy.yml @@ -31,7 +31,7 @@ jobs: # build site and check links - name: run htmltest - run: make htmltest + run: make lint - name: save htmltest cache uses: actions/cache/save@v4 diff --git a/.htmltest.yml b/.htmltest.yml index f901780f3..446c08628 100644 --- a/.htmltest.yml +++ b/.htmltest.yml @@ -7,38 +7,9 @@ HTTPHeaders: User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:108.0) Gecko/20100101 Firefox/108.0 IgnoreCanonicalBrokenLinks: false IgnoreHTTPS: - - "http://bei\\.umontreal\\.ca" - - "http://jur\\.byu\\.edu" - - "http://www\\.ahc\\.umontreal\\.ca/ActivitesJumelage/interlinguistique\\.htm" + - "^http://bei\\.umontreal\\.ca" + - "^http://jur\\.byu\\.edu" + - "^http://www\\.ahc\\.umontreal\\.ca/ActivitesJumelage/interlinguistique\\.htm" - "^http://www\\.faecum\\.qc\\.ca/$" - "^http://www\\.stablediffusionfrivolous.com/$" -IgnoreURLs: - # these sites don't like GitHub - - "^https://www\\.reddit\\.com/" - - "^https://twitter\\.com/" - - "^https://wise\\.com/" - - "^https://www\\.datanami\\.com/" - - "^https://www\\.linkedin\\.com/in/kyle-roth/$" - - "^https://openreview\\.net/pdf\\?id=ryxWIgBFPS$" - - "^https://www\\.pnas\\.org/doi/10\\.1073/pnas\\.2016976118$" - - "^https://www\\.wsj\\.com/articles/futures-exchange-reins-in-runaway-trading-algorithms-11572377375$" - - "^https://devdojo\\.com/bobbyiliev/how-to-install-docker-and-docker-compose-on-raspberry-pi$" - - "foundation.mozilla.org" - - "www.google.com/maps/dir" - - "fee.org" - - "papers.nips.cc" - - "^https://byu\\.edu$" - - "digitalcommons.usu.edu" - - "news.ag.org" - - "tech.churchofjesuschrist.org" - - "philpapers.org" - - "nytimes\\.com" - - "www\\.semanticscholar\\.org" - - "www\\.mdpi\\.com" - - "www\\.jstor\\.org" - - "www\\.urbandictionary\\.com" - - "unesdoc\\.unesco\\.org" - # onion sites - - "http://.*\\.onion" - # expired HTTPS cert - - "https://5kids1condo\\.com/access-the-good-life-dont-own-it/" + - "^http://[^/]\\.onion" diff --git a/Makefile b/Makefile index 79c9a044b..fff20277b 100644 --- a/Makefile +++ b/Makefile @@ -3,6 +3,10 @@ SRCFILES = $(shell find assets content layouts static config.toml -type f -name public: $(SRCFILES) hugo --gc --minify -.PHONY: htmltest -htmltest: public .htmltest.yml - htmltest +.PHONY: dev-server +dev-server: + hugo server -D + +.PHONY: lint +lint: public .htmltest.yml + htmltest --skip-external From b9a49fd1ee7fddd50dc45be3a836a57da3d190f0 Mon Sep 17 00:00:00 2001 From: k Date: Wed, 30 Sep 2026 14:53:38 -0400 Subject: [PATCH 4/7] ci: update Actions and use htmltest Action --- .github/workflows/deploy.yml | 19 ++++++++++--------- 1 file changed, 10 insertions(+), 9 deletions(-) diff --git a/.github/workflows/deploy.yml b/.github/workflows/deploy.yml index 4fd67a99f..1acb21099 100644 --- a/.github/workflows/deploy.yml +++ b/.github/workflows/deploy.yml @@ -10,10 +10,10 @@ jobs: main: runs-on: ubuntu-latest steps: - - uses: actions/checkout@v4 + - uses: actions/checkout@v7 - name: restore htmltest cache - uses: actions/cache/restore@v4 + uses: actions/cache/restore@v6 with: path: tmp/.htmltest # We want the latest cache to hit and be updated every time. @@ -22,19 +22,20 @@ jobs: htmltest - name: set up Hugo - uses: peaceiris/actions-hugo@v2 + uses: peaceiris/actions-hugo@v3 with: hugo-version: latest - - name: download htmltest - run: curl https://htmltest.wjdp.uk | sudo bash -s -- -b /usr/local/bin + - name: build site + run: make public - # build site and check links - name: run htmltest - run: make lint + uses: wjdp/htmltest-action@31be84a95c860a331e0cf9a99f71e3eb39d2f86b + with: + skip_external: true - name: save htmltest cache - uses: actions/cache/save@v4 + uses: actions/cache/save@v6 if: always() with: path: tmp/.htmltest @@ -42,7 +43,7 @@ jobs: # deploy to GitHub pages if this is the main branch - name: deploy - uses: peaceiris/actions-gh-pages@v3 + uses: peaceiris/actions-gh-pages@v4 if: github.ref == 'refs/heads/main' with: github_token: ${{ secrets.GITHUB_TOKEN }} From 72782db4228e8acc4719dfaee43df3d3e6f16062 Mon Sep 17 00:00:00 2001 From: k Date: Wed, 30 Sep 2026 16:33:25 -0400 Subject: [PATCH 5/7] consolidate heavy archival resources under dl.kylrth.com/web --- content/paper/better-nicer-cleaner-fairer/index.md | 2 +- content/paper/unsolved-problems-ml-safety/index.md | 2 +- content/post/team-human-tenen/index.md | 2 +- 3 files changed, 3 insertions(+), 3 deletions(-) diff --git a/content/paper/better-nicer-cleaner-fairer/index.md b/content/paper/better-nicer-cleaner-fairer/index.md index b9fb48c16..c74bc9680 100644 --- a/content/paper/better-nicer-cleaner-fairer/index.md +++ b/content/paper/better-nicer-cleaner-fairer/index.md @@ -97,7 +97,7 @@ Axon is apparently trying to put on the appearance of transparency by engaging e ### machine translation -Many statements prioritize transparency and explainability of systems.{{% sidenote %}}While reading this, I'm thinking about the [paper on fairwashing](https://proceedings.mlr.press/v97/aivodji19a) referenced in Abigail Thorn's ["Here's what ethical AI really means"](https://www.youtube.com/watch?v=AaU6tI2pb3M) ([transcript](https://dl.kylrth.com/transcripts/ethical_ai.txt)), which demonstrates that (at least with current explainability tools) it's always possible to provide an explanation that appears reasonable even for a model known to be unfair. And as Thorn goes on to say, a focus on model interpretability ignores the political question of how individuals interacting with institutions are supposed to accept or respond to the explanations those institutions give. When we frame automation in general as "encoding ways of seeing", machine learning becomes a really useful tool for institutions that want to make cheap decisions without doing the work to a) encoding their ways of seeing explicitly or b) make those decisions transparent (much less accountable) to the individuals being processed.{{% /sidenote %}} +Many statements prioritize transparency and explainability of systems.{{% sidenote %}}While reading this, I'm thinking about the [paper on fairwashing](https://proceedings.mlr.press/v97/aivodji19a) referenced in Abigail Thorn's ["Here's what ethical AI really means"](https://www.youtube.com/watch?v=AaU6tI2pb3M) ([transcript](https://dl.kylrth.com/web/transcript_ethical_ai.txt)), which demonstrates that (at least with current explainability tools) it's always possible to provide an explanation that appears reasonable even for a model known to be unfair. And as Thorn goes on to say, a focus on model interpretability ignores the political question of how individuals interacting with institutions are supposed to accept or respond to the explanations those institutions give. When we frame automation in general as "encoding ways of seeing", machine learning becomes a really useful tool for institutions that want to make cheap decisions without doing the work to a) encoding their ways of seeing explicitly or b) make those decisions transparent (much less accountable) to the individuals being processed.{{% /sidenote %}} ## main takeaways diff --git a/content/paper/unsolved-problems-ml-safety/index.md b/content/paper/unsolved-problems-ml-safety/index.md index 531752547..936d40c1d 100644 --- a/content/paper/unsolved-problems-ml-safety/index.md +++ b/content/paper/unsolved-problems-ml-safety/index.md @@ -7,4 +7,4 @@ tags: ["deep-learning", "ethics"] thumbnail: "icons.png" --- -*This was a paper we presented about in Irina Rish's neural scaling laws course ([IFT6760A](https://sites.google.com/view/towards-agi-course)) in winter 2023. You can view the slides we used [here](https://docs.google.com/presentation/d/11VtXg-sfLkjtIQWXEEEm875esK-8QrNFr09woMYzrV8/edit?usp=sharing), and the recording [here](https://sites.google.com/view/towards-agi-course/schedule#h.lmlkbq72t3iz) (or my backup [here](https://dl.kylrth.com/videos/2023-02-02-ml-safety.mp4)).* +*This was a paper we presented about in Irina Rish's neural scaling laws course ([IFT6760A](https://sites.google.com/view/towards-agi-course)) in winter 2023. You can view the slides we used [here](https://docs.google.com/presentation/d/11VtXg-sfLkjtIQWXEEEm875esK-8QrNFr09woMYzrV8/edit?usp=sharing), and the recording [here](https://sites.google.com/view/towards-agi-course/schedule#h.lmlkbq72t3iz) (or my backup [here](https://dl.kylrth.com/web/2023-02-02-ml-safety.mp4)).* diff --git a/content/post/team-human-tenen/index.md b/content/post/team-human-tenen/index.md index 01f0fffc6..05ef12ea7 100644 --- a/content/post/team-human-tenen/index.md +++ b/content/post/team-human-tenen/index.md @@ -6,7 +6,7 @@ tags: ["politics", "critique", "ethics"] thumbnail: "ltfr-cover.jpg" --- -*You can view this episode [here](https://www.teamhuman.fm/episodes/265-dennis-yi-tenen), and you can download a transcript I made with whisper-medium [here](https://dl.kylrth.com/teamhuman_tenen.txt). I accept responsibility for errors in the transcript, alongside OpenAI, all people whose voices exist on the web, and the rest of humanity. :)* +*You can view this episode [here](https://www.teamhuman.fm/episodes/265-dennis-yi-tenen), and you can download a transcript I made with whisper-medium [here](https://dl.kylrth.com/web/transcript_teamhuman_tenen.txt). Please let me know if there are errors in the transcript.* ## writing technology From 7e2c62d3494c114f83f8c573f3c05a5456e7e6f6 Mon Sep 17 00:00:00 2001 From: k Date: Wed, 30 Sep 2026 14:42:45 -0400 Subject: [PATCH 6/7] post: AI research is dead --- content/post/ai-research-is-dead/index.md | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) create mode 100644 content/post/ai-research-is-dead/index.md diff --git a/content/post/ai-research-is-dead/index.md b/content/post/ai-research-is-dead/index.md new file mode 100644 index 000000000..425185c6f --- /dev/null +++ b/content/post/ai-research-is-dead/index.md @@ -0,0 +1,16 @@ +--- +title: "AI research is dead, long live AI" +date: 2026-09-30T10:41:41-04:00 +draft: false +tags: ["ai", "politics", "science"] +--- + +*inspired by Mike Cook's 2026-09-22 article ["Why I love AI"](https://www.possibilityspace.org/blog/posts/i-love-ai/) ([MHTML archive](https://dl.kylrth.com/web/why_i_love_ai.mhtml))*{{% sidenote %}}[MHTML](https://en.wikipedia.org/wiki/MHTML) is a web archiving file format that combines an HTML page and its associated resources into a single file.{{% /sidenote %}} + +This may be showing my youth, but I am surprised to learn that route optimization algorithms were once considered a part of AI research; I've always seen it treated as a highly distinct field, sometimes even outside the realm of informatics/CS.{{% sidenote %}} My department is named *Département d'informatique et de recherche opérationelle*, or "department of informatics and operations research", where "operations research" is about such optimization algorithms.{{% /sidenote %}} If in a different life I'd had the time to become invested in AI research for 20 years before things became the way they are in 2026, I could totally see myself feeling left behind like Cook feels. But critically, just as he was seeing his field left behind, I was seeing my field of interest, natural language processing, becoming *subsumed* by this latest instantiation of "AI research". + +I resonate with Cook's evaluation of the deadness of modern AI research. Easily half of AI papers today are "we created a new agent pipeline to do X human task", where "agent pipeline" means a bag of prompts to GPT-*n*. The field has collapsed around this because of a mindset that fundamentally believes we've created "artificial general intelligence"—at least in some weak sense—and that what's left is therefore to work out how to *apply* this intelligence to every domain of society. If the moment of ChatGPT was a Kuhnian scientific revolution, the new "normal science" is not only societally{{% sidenote %}}fiona fokus, "I don't care how well your 'AI' works". 2025-11-25, [personal blog](https://fokus.cool/2025/11/25/i-dont-care-how-well-your-ai-works.html).{{% /sidenote %}} and environmentally devastating{{% sidenote %}}Josh Axelrod, "The hidden environmental cost of AI data centers". 2026-09-04, [Deutsche Welle](https://www.dw.com/en/the-hidden-environmental-cost-of-ai-data-centers/a-78789889).{{% /sidenote %}}{{% sidenote %}}Paul Schütze, "The problem of sustainable AI: a critical assessment of an emerging phenomenon". 2024, [*Weizenbaum Journal of the Digital Society* 4: 1](https://ojs.weizenbaum-institut.de/index.php/wjds/article/view/4_1_4). ([doi:10.34669/WI.WJDS/4.1.4](https://doi.org/10.34669/WI.WJDS/4.1.4)){{% /sidenote %}} but also incredibly *boring*. The task of AI science is no longer to invent any new knowledge, but to further the process of making *AI* seem to have that knowledge.{{% sidenote %}}Olivia Guest, "Models". 2026-06-10, [Zenodo](https://zenodo.org/records/20639860).{{% /sidenote %}} + +I arrive at this conclusion differently than Cook because my field of NLP, rather than having been left behind by AI, has been completely hollowed out and replaced by three LLMs in a trenchcoat. My particular resulting disillusionment is perhaps why I depart from Cook on the reason why AI research engages in goalpost-moving in the first place. When Cook says "AI is a field built on trying to get computers to do things that they can't do", what that definition *hides* is the category used to define that set of tasks. For the field of AI, this has always been about the "applications of intelligence" which Enlightenment rationalism can parse human activity into, *especially* those which are significant to capital and the state. This is why NLP was so easily subsumed by AI: passing the Turing test suddenly rendered LLMs plausible candidates for so many of the human-replacing applications AI researchers have always dreamed of. The fact that these "AIs" are statistical models of *language* means that prior NLP subtopics like natural language understanding (NLU) or speech recognition (ASR) are now immanently recognizable as components of AI research, while other classical areas of NLP like semantic tagging or syntax parsing are discarded as vestiges of the era of symbolic NLP, left to be categorized as "computational linguistics". + +I think the most succinct way to describe this discursive shift in AI research is that in the moment of ChatGPT, the previously eschatological phenomenon of artificial intelligence has now *arrived*. AI researchers no longer point to a coming Messiah for their funding, but point backward to that moment in 2022 when chatbots became believable enough we collectively started treating them like stochastic oracles containing hidden knowledge. If the oracles are here now and the gods are speaking through them, all that's left for the scientists-turned-priests of AI is to [interpret their words](https://ieeexplore.ieee.org/abstract/document/11540994),{{% sidenote %}}Mohamed Amine Ferrag, Norbert Tihanyi, Mérouane Debbah, "From LLM reasoning to autonomous AI agents: a comprehensive review". 2026-06-01, [*IEEE Access* 14: 84237-84285](https://ieeexplore.ieee.org/abstract/document/11540994).{{% /sidenote %}} [keep the gods fed](https://www.bloomsbury.com/us/feeding-the-machine-9781639734979/),{{% sidenote %}}Callum Cant, James Muldoon, Mark Graham, *Feeding the machine: the hidden human labor powering A.I.* 2024-08-06, [Bloomsbury Publishing](https://www.bloomsbury.com/us/feeding-the-machine-9781639734979/).{{% /sidenote %}} and [warn of the coming destroyer god](https://www.cnbc.com/2026/09/09/anthropic-researcher-quits-ai-safety.html).{{% sidenote %}}Ashley Capoot, Arjun Kharpal, "Experts weigh in as researcher says AI has more than 10% chance of 'killing all humans'". 2026-09-09, [CNBC](https://www.cnbc.com/2026/09/09/anthropic-researcher-quits-ai-safety.html).{{% /sidenote %}} From 7e8013dfab3d6693c70d7d2e142f36f6178f2eff Mon Sep 17 00:00:00 2001 From: k Date: Wed, 30 Sep 2026 16:13:46 -0400 Subject: [PATCH 7/7] 2026 profile update - update pfp and contact info - note shift in perspective over time, with more to come - readme: explain public repo with reference to software commons --- README.md | 4 ++-- layouts/index.html | 43 ++++++++++++++++--------------------- static/images/my_face.png | Bin 309269 -> 0 bytes static/images/pfp_wavy.jpg | Bin 0 -> 592097 bytes 4 files changed, 21 insertions(+), 26 deletions(-) delete mode 100644 static/images/my_face.png create mode 100644 static/images/pfp_wavy.jpg diff --git a/README.md b/README.md index 7312028cf..3db94c9b7 100644 --- a/README.md +++ b/README.md @@ -1,8 +1,8 @@ # [kylrth.com](https://kylrth.com) -This is the source for my personal website. I keep the repo public because I think it's valuable to be [working in the open](https://duckduckgo.com/?q=working+in+the+open) whenever possible. +This is the source for my personal website. I keep the repo public and permissively licensed because I think it's valuable to demonstrate public-facing computing practices in the public commons. -I started this website from the [Pickles theme](https://github.com/mismith0227/hugo_theme_pickles/), which I've copied and modified as needed. It's published with an MIT license, which is preserved [here](layouts/LICENSE). The rest of this repository (including the content) is licensed with [CC0](https://creativecommons.org/share-your-work/public-domain/cc0/), which means you're free to use it how you like. +I started this website from the [Pickles theme](https://github.com/mismith0227/hugo_theme_pickles/), which I've copied and modified to my needs. It's published with an MIT license, which is preserved in [`layouts/LICENSE`](layouts/LICENSE). The rest of this repository (including the content) is licensed with [CC0](https://creativecommons.org/share-your-work/public-domain/cc0/), which means you're free to use it how you like. Please do not use this license to perform false attribution, which I interpret to include LLM training, without explicit permission from me. ## building diff --git a/layouts/index.html b/layouts/index.html index 9e3c8a23f..730e2f16a 100644 --- a/layouts/index.html +++ b/layouts/index.html @@ -1,17 +1,23 @@ {{ partial "header.html" . }} {{ partial "menu.html" . }} -my face +A photo of my face taken at Parc National La Mauricie
-

I'm a third-year computer science PhD student at I'm a computer science PhD candidate at DIRO at the Université de Montréal, advised by Bang Liu. I study the structured - use of LLMs within larger pipelines and agent systems, from both robustness and - sociotechnical perspectives. Here's my + href="https://www-labs.iro.umontreal.ca/~liubang/">Bang Liu. + I apply a critical sociotechnical perspective to study the use of LLMs within larger pipelines and agent systems, + especially in the context of code generation. + Here's my resume (and the academic version).

+

During the course of my PhD program, my understanding of the political nature of "AI" (and + the implications for my own research) has shifted substantially. My latest post + "AI research is dead, long live AI" gets into + this a little, but a lot of the older posts here I no longer see the same way. More + changes to this site are on the way.

Previously I was a speech scientist at Cobalt Speech & Language, a company that designs custom speech recognition, text-to-speech, and dialogue models. While at Cobalt I worked on some neat projects, including language modeling @@ -19,20 +25,13 @@ models.

I graduated from BYU with a BS in Applied and Computational Mathematics (ACME) with an emphasis in linguistics and a - minor in computer science. ACME's rigorous curriculum includes graduate-level courses in + minor in computer science. ACME's curriculum includes graduate-level courses in algorithms, analysis, optimization, statistics, data science, optimal control, and machine learning.

-

During my undergrad I interned with Cobalt Speech, as well as Emergent Trading, an automated trading firm that - made - the news for reporting a problem in a Eurodollar exchange rule that unfairly favored - larger competitors. (I developed the analysis tools that were used to track the issue down - and determine how our opponent was taking advantage of the rule.) -

+

During my undergrad I interned both with Cobalt Speech and Emergent Trading.

-

contact (he/him) contact (he/they) permalink @@ -44,14 +43,10 @@

contact (he/him) href="post/matrix-registration/">here.) My GPG public key is here.

- Matrix -  /  - Signal -  /  - Session  /  - email  /  - GitHub  /  + Matrix  /  + Signal  /  + email  /  + GitHub  /  LinkedIn

@@ -123,6 +118,7 @@

other stuff