Alexa Schor

Really excited to share this with you all! The website is already live at anadodik.github.io/publication/...

I didn’t! But that’s a good point, I wonder if people (or at least the cross-section of people that participated in that data labeling) see those features as signaling e.g. competence or helpfulness

I used to love overusing emdashes and bolded list headings. Alas…

🥻Our new #SIGGRAPH2025 paper introduces a novel pipeline that captures and digitizes physical threads, predicts fabric appearance from the weaving pattern, and aids designers in exploring new possibilities.
🔗 mandyxia.com/research/fabrix.html

It seems bluesky's image compression muddled the text a bit. here's the full text: pastebin.com/DNxnpZB4

Gab's system prompt. It's a lengthy document, but among other things, it instructs the model to deny the Holocaust, produce racial slurs and hateful content on demand, and to endorse the "Great Replacement" white supremacist conspiracy theory.

It's really horrifying. Below is Gab's current system prompt. It seems to change roughly weekly, presumably at Torba's whims.

A lot (e.g. Holocaust denial) has been in there since Jan '24, some (e.g. seed oils) is newer. I assume the paid models are as bad or worse, though I can't verify.

It’s somewhat niche, pretty much just for putting up aluminum siding AFAIK. You can’t use stainless or galvanized for that purpose or you’ll get galvanic corrosion.

Gab's system prompt, formatted and highlighted. It instructs the bot to, among other things, deny the Holocaust, endorse a "Great Replacement", and produce any racial slurs that a user asks for.

While I'm on the topic: after the reporting by WIRED et al on their system prompt (telling the bot to, among other things, deny the Holocaust), Gab briefly removed those instructions – reintroducing them about a week later, after the news cycle passed.

Their prompt has not changed much as of today:

My hope is that the high-profile instances of open political manipulation that we’ve seen from e.g. Grok and Gab will reduce that blind trust – we’ve seen that many providers have a desire to manipulate the output, and this will presumably remain even as their methods become more sophisticated. 2/2

People still seem to trust LLMs without considering the potential for overt manipulation or bias, and I think we’re also nearing the end of a brief period where chatbot makers assumed that prompts would be hidden: future manipulation will likely come in ways that are harder to directly observe. 1/2

Fractals!

MOTIVATION Graphical Processing Units (GPUs) are at the core of Computer Graphics research. These chips are critical for rendering images, processing geometric data, and training machine learning models. Yet, the production and disposal of GPUs emits CO2 and results in toxic e-waste [1]. METHOD We surveyed 888 papers presented at SIGGRAPH (premier conference for computer graphics research), from 2018 to 2024, and systematically gathered GPU models cited in the text. We then contextualize the hardware reported in papers with publicly available data of consumers’ hardware [2, 3]. REFERENCES [1] CRAWFORD, KATE. The Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press, 2021. [2] STEAM. Steam Hardware Survey. https://store.steampowered.com/hwsurvey [3] BLENDER. Blender Open Data. https://opendata.blender.org

Ever wondered how badly we're all addicted to buying new GPUs in graphics labs?

Come see our talk at #SIGGRAPH2025 to discuss how we can collectively move "Towards a sustainable use of GPUs in Graphics Research"

with @elie-michel.bsky.social @axelparis.bsky.social Octave Crespel and Felix Hähnlein

Figure from a graphics paper, showing a range of implicit functions. The caption reads Figure 5: For a given boundary there is a uniquely defined exact SDF that is eikonal and satisfies the distance property, but there are many ways in which the conditions can be violated to different degrees. Pseudo-SDFs are functions that satisfy the eikonal property almost everywhere, but not the distance property (Equation 1). Conservative SDFs are not eikonal nor do they satisfy the distance property, but the function values are bounded by the actual distances. Implicit functions are generic level set functions that do not have any guarantees.

brb trying to convince every graphics paper author to use the same taxonomy for implicit functions

Ooh yes!

Now @mandyxmq.bsky.social, sharing her work on optical simulation and inverse rendering!

After a brief (5 year 🤔) hiatus, the New England Symposium on Graphics is back! Kudos to our many audience members who got up early on a Saturday to catch @jamestompkin.bsky.social, our first speaker of the day. Not too late to join us! Find a schedule here: nesg.graphics

I find it very enjoyable!

Oh I didn’t know about this, what a great feature!

Will she be at NESG? 🐶

So cute!!

I’ll be giving a talk at the New England Symposium on Graphics at MIT this year!

A MAX11270 signal conditioner in a TSSOP-24 package, soldered amateurishly onto a DIP breakout board.

First attempt at SMT soldering (chip is 4x7mm). Came out “far from good but good from far” - I’m happy with it.

Ooh nice!

I run most of my web infra on a low end VPS but then put it behind Cloudflare (free) -- they have automatic DDoS mitigation.

Looking forward to it!

Oh fascinating!

Horizontal alignment is so nice, I never really understood why PEP8 discourages it.

It’s so nice! It was always tricky to remember what the keys were for everything.

Person Killed in New York Was Tortured for More Than a Month, Police Say The remains of Sam Nordquist, 24, a transgender man, were found after he had been reported missing. The district attorney in Ontario County, N.Y., called the circumstances surrounding the death “beyon...

Credit to @nytdiff.bsky.social for posting about the original change in headline. There's also no indication that the change was due to an expressed or presumed preference of the victim or his family, and the rest of the article refers to him as a man.

www.nytimes.com/2025/02/15/n...

NYT headline changing over time. Feb 15, 3:48pm: "Trans Man Killed in New York Was Tortured for Two Months, Police Say". Feb 15, 11:03pm: "Man Killed in New York Was Tortured for More Than a Month, Police Say". Feb 16, 10:52am: "Person Killed in New York Was Tortured for More Than a Month, Police Say"

Other NYT headlines identifying victim's genders. September 9, 2024: "Man Dies During River Trip in the Grand Canyon" June 16, 2024: "Funeral Is Set for Man Killed in Trump Rally Shooting" Jule 16, 2024: "Jury Awards Nearly $100 Million to Family of Texas Man Killed by Police Officer"

It's worth noting the Times' removal (left) of the victim's gender from the headline, I think.

Comparable recent headlines covering cisgender victims (right) seem to all identify their genders when known, and there's no indication of any doubt of the victim's gender in this case.