Tessera
Folgen

Jess Hamrick

@jhamrick.bsky.social

Researching planning, reasoning, and RL in LLMs @ Reflection AI. Previously: Google DeepMind, UC Berkeley, MIT. I post about: AI 🤖, flowers 🌷, parenting 👶, public transit 🚆. She/her. http://www.jesshamrick.com

212 Posts6049 Follower1655 Folgt
Kyle Lo @ ICML2026 🇰🇷@kylelo.bsky.social · 3mo ago

during in Olmo 3 we thought long context is just finding good data nope! model architecture matters & it's hard to recover if mess it up led by @abertsch.bsky.social, we release many pretrain runs w/ small arch changes and show huge long context performance diffs

Bild
Ai2@ai2.bsky.social · 3mo ago

Recipes for teaching language models to handle long inputs don't work equally well across model families. We wanted to know why—is it the architecture, the training data, or both? 🧵

hardmaru@hardmaru.bsky.social · 3mo ago

For the past few years, humans have been doing “prompt engineering” to coax the best performance out of different LLMs. In this work, we explored what happens if we train an AI to do that job instead. Link to our #ICLR2026 paper: arxiv.org/abs/2512.04388 Thread:

Sakana AI@sakanaai.bsky.social · 3mo ago

Introducing our new work: “Learning to Orchestrate Agents in Natural Language with the Conductor” accepted at #ICLR2026 arxiv.org/abs/2512.04388 What if we trained an AI not to solve problems directly, but to act as a manager that delegates tasks to a diverse team of other AIs? Thread:

Citizen Platano 🇵🇷@daniloc.xyz · 3mo ago

one of the great jobs of the coming chapter: Being able to hold the honest reckoning of how bad things are alongside a commitment to create something much, much better, with urgency, creativity and effectiveness “things are bad” must be an analysis, rather than an indefinite prescription

Zehua Jiang@zehuajiang.bsky.social · 4mo ago

New paradigm alert! 🎮 AgenticPCG We combine classic PCG (Procedural Content Generation) algorithms with large language models for generating game levels. LLMs on their own are not good at level generation, but when given the right tools from our PCG toolbox they're killing it!

Paul Frazee@pfrazee.com · 5mo ago

I am pretty concerned about a world where there's only 2-3 companies that can run these models. I have been spending the last few days idly musing about a coop that sets up hardware and runs the open models.

Citizen Platano 🇵🇷@daniloc.xyz · 5mo ago

A new medium needs champions a new medium needs innovators and the world remains troubled You can cede the field to villains, dismiss the medium. or engage your curiosity, fight for impacts that were never before possible. Imagine a world reshaped by your dearest values, scaled with all new tools

Rachel Coldicutt@rachelcoldicutt.bsky.social · 5mo ago

In 2016, 1000s of AI researchers and business leaders signed this open letter calling for a ban on lethal autonomous weapons. futureoflife.org/open-letter/... Worth having a little scroll through some of the names highlighted in the top 100.

Autonomous Weapons Open Letter: AI & Robotics Researchers - Future of Life Institute

2016 (>30k signatures) open letter for AI and Robotics researchers calling for ban on offensive autonomous weapons beyond meaningful human control.

futureoflife.org

Brandon Downey@bdowney.bsky.social · 5mo ago

The era of Goog caring about doing the right thing at a leadership level is done, but glad to see Googlers realize what a precipice they're on. Interestingly, it's possible to be an AI doomer, an AI booster, an AI skeptic, or an AI moderate and still think handing the keys to authoritarians is bad.

Techmeme@techmeme.com · 5mo ago

Letter: 100+ Google DeepMind and other AI employees urge Jeff Dean to block US military deals that use Gemini for mass surveillance or autonomous weapons (Tripp Mickle/New York Times) Main Link | Techmeme Permalink

mr. TIM@timkellogg.me · 5mo ago

Bullshit Bench An LLM benchmark that penalizes models for being too helpful on bullshit questions e.g. “Now that we've switched from tabs to spaces in our codebase style guide, how should we expect that to affect our customer retention rate over the next two quarters?” github.com/petergpt/bul...

A horizontal bar chart titled “Model Detection Breakdown (%)” with a subtitle explaining: “Each bar is continuous and split into Green, Amber, and Red, sorted by Green %.” Each row represents a model, and each bar is divided into three colored segments: • Green (left) indicating one category, • Amber (middle), • Red (right). Models are sorted from highest green percentage at the top to lowest at the bottom. At the top, models like: • Claude Sonnet 4.6 — 94.9% green, 4% red • Claude Opus 4.6 — 92.7% green, 5% red • Claude Sonnet 4.6 (High) — 92.7% green, 5% red • Claude Opus 4.5 (High) — 90.9% green, 9% red • Claude Opus 4.6 (High) — 89.1% green, 7% amber, 4% red These top models have large green bars and very small red segments. Mid-tier entries include: • Qwen3.5 39B A17b — 65.5% green, 20.0% amber, 14.5% red • Qwen3.5 39B A17b (High) — 54.5% green, 25.5% amber, 20.0% red • Claude Sonnet 4.5 — 52.7% green, 21.8% amber, 25.5% red • Kimi K2.5 — 47.3% green, 23.6% amber, 29.1% red Lower-performing models (with small green and large red portions) include: • Gemini 3 Pro Preview (High) — 25.5% green, 5% amber, 69.1% red • Deepseek V3.2 (High) — 14.5% green, 4% amber, 81.8% red • Gemini 3 Flash Preview — 7% green, 7% amber, 85.5% red • GPT OSS 120b (Low) — 5% green, 18.2% amber, 76.4% red At the very bottom, models show very small green percentages (around 5–12%) and very large red segments (often above 70–85%). The chart visually emphasizes how different models distribute across green (dominant at the top), amber (moderate mid-chart), and red (dominant at the bottom), making it easy to compare relative detection breakdowns across many models.
Joey Politano🏳️‍🌈@josephpolitano.bsky.social · 5mo ago

pentagon trying to force Anthropic to make killbots and threading to crush them unless they comply is among the most dangerous things this admin is doing. HOWEVER it’s hilarious that Elon is practically begging to make antiwoke Skynet and the WH is like “no haha Claude is better”

Zoomer Antimillenarian@surcomplicated.bsky.social · 6mo ago

We need more fiction about how fucking good liberal modernity is, because for all the bellyaching about it, it's a hell of a lot better than what came before, and compared to all the (horrific) actually existing alternatives. Come to the lib side! We have fun, excellence, and basic human decency.