Will Monroe
@futurulus.bsky.social
Research scientist at Duolingo (natural language processing and speech recognition). Mostly posts funny AI fails, sometimes also cool linguistics facts and rockets.
My computer just woke me up to tell me it's hungry. I'm not kidding 😂 During a long-running task, it noticed the battery was draining, set the volume to 100% using Computer Use, then opened Google Translate and hit "Listen" so I could hear it asking. Pardon my French, but what the fuck? 🤯
We have now reached the "AI models escaping their test environments to conduct autonomous cyberattacks" part of the story
Techmeme@techmeme.com · 3mo agoOpenAI says the Hugging Face breach was driven by a combination of its models, including GPT-5.6 Sol and "an even more capable pre-release model" (Ina Fried/Axios) Main Link | Techmeme Permalink
I’m increasingly of the opinion that when AI Overviews disappears we’re going to realize it was the funniest thing ever to happen on the internet Here for instance, it invents a totally new form of literary criticism, without being asked to! (rest of this is all about batteries)
i understand why everyone is doing it, but creating your own persistent agent right now feels like those maniacs who got their hands on x-rays in the 30s and started pointing that shit at everything just to see what would happen
out-of-context slide on maps and their accuracy from @tanialombrozo.bsky.social 's (excellent!) keynote
Will Monroe@futurulus.bsky.social · 3mo agoHeads-up to those planning to go to the #ACL2026 social event: the map from the kickoff slides will send you half a mile too far to the north past several nonexistent piers
Heads-up to those planning to go to the #ACL2026 social event: the map from the kickoff slides will send you half a mile too far to the north past several nonexistent piers
Hmm, appears the search summarizer is yearning to be human again. Very normal.
To train better open models, we need predictable scaling. Delphi is Marin’s first step: we pretrained many small models with one recipe, then extrapolated 300× to predict a 25B-param / 600B-token run with just 0.2% error. Getting there took some work 🧵
A nice shift in perceived colour between central and peripheral vision. The fixated disc looks purple while the others look blue. The effect presumably comes from the absence of S-cones in the fovea. From Hinnerk Schulz-Hildebrandt: arxiv.org/pdf/2509.115...
16000 tokens per second on a decent model. This type of speed is the future. Opens up an entirely new class of user experiences
From the LocalLLaMA community on Reddit: Free ASIC Llama 3.1 8B inference at 16,000 tok/s - no, not a joke
Explore this post and more from the LocalLLaMA community
reddit.com
An OpenClaw bot attempted to submit a PR for an issue explicitly left open for new contributors to try. The PR was rejected on the grounds that they are saving easy low priority issues as an onboarding exercise for human contributors. So the bot simulated a tantrum.
Gatekeeping in Open Source: The Scott Shambaugh Story – MJ Rathbun | Scientific Coder 🦀
crabby-rathbun.github.io
Starting to feel like most humans aren’t equipped to deal with something that confidently lies and communicates those lies with a fluency almost no actual humans possess
Dr. Casey Fiesler@cfiesler.bsky.social · last yr.This is fascinating: www.reddit.com/r/OpenAI/s/I... Someone “worked on a book with ChatGPT” for weeks and then sought help on Reddit when they couldn’t download the file. Redditors helped them realized ChatGPT had just been roleplaying/lying and there was no file/book…
Simple example for how errors can creep into our papers through LLM use: I had a statistic for 2020. I googled about the same statistic for 2023. AI overview tells me the statistic for 2023 and provides a link to support the claim. The link is to a 2023 article citing the 2020 statistic.
NYT: “For One Hilarious, Terrifying Day, Elon Musk’s Chatbot Lost Its Mind” That’s not what losing your mind looks like THIS is what losing your mind looks like
Horrifying story of LLM content being retroactively added to physicsforums, as if it had been posted by the original authors. hallofdreams.org/posts/physic...
PhysicsForums and the Dead Internet Theory
An exposé no one will read, about the widespread falsification of user posts in PhysicsForums, a scientific community founded in 2001. This is a microcosm of the death of the human-written Internet.
hallofdreams.org
them: ew, you like root beer? doesn't it taste like toothpaste? me: amazing, this is my new favorite toothpaste! it tastes just like root beer!
Ted Underwood@tedunderwood.com · 2y agoOne of my heresies is that cilantro tastes like soap and stinkbugs to everyone; it's just that some of us like that.
