Dane Carnegie Malenfant

People assume wrongly that it's like a support vector machine where the math that does the inference is fairly tractable and we know the rough algorithm it learns to accomplish the task. No, we actually do not know here it's too abstract and opaque.

Archival research has found 171 confirmed deaths at Kuper Island residential school, 50 more than previously thought Archival research conducted on behalf of the Penelakut Tribe has found 50 more deaths at the former residential school than previous estimates

Breaking: the team investigating the Kuper Island Residential School in BC has confirmed 171 children deaths at the institution, 50 more than previously known. As a residential school historian and someone fighting residential school denialism, here's a 🧵 www.timescolonist.com/local-news/a...

if you didn't catch Juliana Trach's poster at @cogscisociety.bsky.social here it is ––>

HRM analysis by @dorialexander.bsky.social

the actual shocking parts:

* it doesn’t overfit
* ARC-AGI is only hard for language models

i think we’ll be seeing more of HRM

Evidence from Formal Logical Reasoning Reveals that the Language of Thought is not Natural Language Humans are endowed with a powerful capacity for both inductive and deductive logical thought: we easily form generalizations based on a few examples and draw conclusions from known premises. Humans al...

Is the Language of Thought == Language? A Thread 🧵
New Preprint (link: tinyurl.com/LangLOT) with @alexanderfung.bsky.social, Paris Jaggers, Jason Chen, Josh Rule, Yael Benn, @joshtenenbaum.bsky.social, ‪@spiantado.bsky.social‬, Rosemary Varley, @evfedorenko.bsky.social
1/8

"AI researchers slowly but surely work their way towards the mid-2000s in psych research" is genuinely funny as a convergence situation

He spent 4 years as a WW II prisoner of war. Then, this Métis veteran enlisted again | CBC News Métis veteran Urban Vermette from Prince Albert, Sask., was honoured posthumously last week by the South Korean government "for overcoming pain and suffering" as a Second World War POW and re-enlistin...

Tangentially, I ran into one of my dad’s cousins at one of the meetups a few months ago randomly in Montréal. Her grandfather was Urban Vermette and is known for his life www.cbc.ca/news/indigen...

Map of Métis communities across the Prairie provinces (1900s), showing Alberta Métis Settlements, road allowance communities, and Métis farms. Locations span Alberta, Saskatchewan, and Manitoba with notable places like Paddle Prairie, Touchwood, Duck Lake, St. Madeleine, and more.

Recently, I visited my grandparents in Prince Albert for the first time in 7 years (last time being before I moved to Montréal for a bachelors). My grandma is from the Nordale/foxfarm which used to be a road allowance (indigenouspeoplesatlasofcanada.ca/article/road...) that emerged after Batoche.

Some accomplishments include becoming the top %5 in weights and biases run time due to hyperparameter tuning and sanity checks. This could be higher if I set up wandb for DRAC cluster too.

I am presenting this work at the @cocomarl-workshop.bsky.social part of @rl-conference.bsky.social Tuesday (: I additionally have a generalized correction term for n-arbitrary agents (it is like walking a tree for the order of gradients) that I am looking for thoughts, validations or critiques.

Figure showing two parts: the top (a) illustrates the training framework and the bottom (b) illustrates the inference framework for Reflective Generative Models (RGMs). Top: Training Framework (a) • A Policy Model takes in a sequence of tokens: • Gray = Question/Answer tokens • Yellow = <think> and </think> tokens • Blue = Think process tokens • Orange = Step-tokens • The model’s internal representations (from layers N-1 and N) are fed into two components: • Policy Head, which outputs the final action (e.g. text response) with GRPO loss L_GRPO • SPRM Head, which outputs scores Score_1, ..., Score_n to rank candidate thought processes, optimized by L_SPR • The SPRM Head consists of a linear layer, dropout, and another linear layer applied to a feature vector. Bottom: Inference Framework (b) • A question Q is processed by the Policy Model, which generates multiple thought sequences: Think_1, Think_k, each paired with a score s_1, s_k from the SPRM Head. • The highest-scoring thought s_j = max(s) is selected. • That thought is fed back into the Policy Model to produce the final answer A. This framework enables reflection during generation by scoring intermediate “thinking steps” and selecting the most promising one to continue with.

it uses a Self-supervised Process Reward Model (SPRM) to grade several reasoning trajectories

the SPRM is a different model, but mostly not. Same base + 53M for the grading

A recent poll found just 38 per cent of non-Indigenous people in Saskatchewan trust Indigenous Peoples, far below the national average.

On the latest Truth and Politics, Jennifer Laewetz shares her perspective on why that gap exists.

Full discussion: pod.fo/e/307e4b

A watercolor, dated from around 1895, of Kanai citizen Joe Healy. He has long dark hair, a grey jacket and white scarf.

Watercolor portrait of Joe Healy, "the most educated Indian on the reserve," was painted by Nora Foote, a British artist who visited what is now Saskatchewan and Alberta in 1895. As a boy, Healy was a witness to the Marias Massacre in Montana, when over 200 Blackfeet women, children and elders were

Hierarchical Reasoning Model Reasoning, the process of devising and executing complex goal-oriented action sequences, remains a critical challenge in AI. Current large language models (LLMs) primarily employ Chain-of-Thought (CoT...

HRM: Hierarchical Reasoning Model

ngl this sounds like bullshit but i don’t think it is

- 27M (million parameters)
- 1000 training examples
- beats o3-mini on ARC-AGI

arxiv.org/abs/2506.21734

Dark thunder, clouds sweep over the open prairie, dropping curtains of rain in the distance over the pale orange of the early morning sky. A white tipi contrast sharply against the pale green grasses, and the oncoming storm.

We had a storm roll through last night in the forest and I was thinking about how I prefer to being able to see them coming like on the prairies. However, once they are on top of you, it’s a whole lot more scary because there is nowhere to hide. This was taken a week ago in Grasslands National Park.

After being in Montréal for 7 years, I’ve always really appreciated being able to walk at night. I think my dad’s home city (below) and my birth city (North Battleford) are the only Canadian cities with adult crime curfews. Maybe also somewhere in Manitoba or Alberta?

Me and my son a vocal tabby cat

“Of course we look related. He is my son.”

Trying to train RNNs in a biol plausible (local) way? Well, try our new method using predictive alignment. Paper just out in Nat. Com. Toshitake Asabuki deserves all the credit!
www.nature.com/articles/s41...

Visiting my cat 😺

In 2023 I reported on the impacts of wildfires in northern Saskatchewan, which not only endanger human life and communities but also cultural and ecological resources. It's heartbreaking to be covering it again, as Canada experiences another brutal year for wildfires.

A large sandstone butte looms over the prairie as a lonely gravel road curves towards it in the foreground. Clouds swirl in the sky above it.

A massive sandstone butte stands sentinel over the grassland, stark against the clouds looming above it.

A view of the same butte from another side reveals a longer foot off to one end and steep walls around the base. Clouds add drama to the photo.

A view of castle butte showing its steep walls along the base as it stands sentinel over the sagebrush. A sign shares its vital statistics - 60 metres tall, half a kilometre around.

Southern Saskatchewan has many off the road gems. Castle Butte is one of them. This massive sandstone/clay formation was created along with the rest of the Big Muddy Badlands when a giant glacial lake experienced a catastrophic drainage, carving a massive butte-filled valley. 1/2

🧵 Everyone is chasing new diffusion models—but what about the representations they model from?
We introduce Discrete Latent Codes (DLCs):
- Discrete representation for diffusion models
- Uncond. gen. SOTA FID (1.59 on ImageNet)
- Compositional generation
- Integrates with LLM
🧱

These are systems that respond to human writing and (often) techniques that apply to human psychology.

The Math Olympiad is great, coding is important, accelerating science has tremendous value.

But LLMs give a chance for both cultures to contribute in ways that have not been possible for a while.

Hi, are you a journalist or know a journalist who likes to point out that LLMs are “stochastic parrots”? Happy to offer my services explaining why this is not accurate

YouTube video by APTN News How this A.I. datacentre plans to generate more than prompts, in eco-friendly jobs | APTN News

The Woodland Cree First Nation in Alberta is getting into the world of A.I. by starting work on a six hundred and fifty-gigawatt datacentre.

Their chief, Isaac Laboucan-Avirom, says it can provide good jobs and be environmentally friendly.

A photograph of Walter Deiter, in a headdress, presenting to PM Pierre Trudeau and Indian Affairs Minister Jean Chretien, whose backs are to the camera.

1970 presentation in Ottawa of The "Red Paper,” an official Indian response to the 1969 "White Paper" of the Liberal government. Walter Deiter, headdress, presenting to PM Pierre Trudeau and Indian Affairs Minister Jean Chretien.

📷 Duncan Cameron | Library and Archives Canada

I really like how @weightsbiases.bsky.social plots the KL with multi-agent PPO

That’s a wrap on the Indigenous AI Gathering 2025. Over two days, Indigenous knowledge keepers, technologists, youth, and allies came together to chart new paths in AI inclusive of Indigenous ethics and perspectives.

Thank you for being part of this vital conversation.

Congratulations to Matthew Sanspareil and Rune Hartgerink, the winning team of the Indigenous Pathfinders in AI program’s pitch day! Their project, Buffalo in Motion, uses AI to strengthen food sovereignty through buffalo restoration.

As Elder Deborah Delisle opened with the Ohén:ton Karihwatéhkwen and drumming filled the space, the Indigenous AI Gathering got underway this morning — a powerful start to this celebration of Indigenous-led AI innovation.

With Abundant Intelligences, @indspire.bsky.social and @ivado.bsky.social.