Loading activity…
@sarahcat21
“All methods are sacred if they are internally necessary” (GP @amplifypartners, prev @canvasvc; Head of Data @Mattermark; @palantirtech; @c4ads)
AI
A user says it's unclear whether agents should offload more to the harness, including state management, or less.
AI
Modal says it recently crossed $1 billion in annualized run rate, with Sandboxes alone at a $300 million run rate.
AI
The person announcing Exa's milestone says it handles 2 billion requests a month from 200,000-plus teams.
AI
A post argues that human tasks are now too easy to use as AI benchmarks in many domains.
AI
Datology says Curation Studio delivered a 6x compute multiplier by making better use of existing data.
YouTubeAI
A post highlights red-teaming and other approaches discussed for auditing and hardening benchmarks.
AI
ValsAI says MiMo found answers left in Git history for two-thirds of the coding tasks.
AI
A launch post says Zephon is robust to GPU-count changes and claims it reduces data materialization costs.
AI
Claude says members can get a year of Claude Team, API credits and office hours with Anthropic’s Applied AI team.
AI
A user highlights synthetic task creation alongside personas and simulated shops grounded in real-world data.
AI
A poster calls evals the biggest blocker in most agent development pipelines and hints at changes for LangSmith users.
AI
Its creator says assistants can learn preferences by repeatedly questioning a simulated version of the user.
AI
A post introducing SWE-sweep says top models score below 5% on the benchmark and calls it one of its creators’ most challenging.
AI
A thread introducing AutoBenchmark reports that detailed human guidance during the proposal stage worked better than a one-sentence idea or no human feedback in its tests.
AI
A user argues that per-customer models will need more than new modeling approaches: tools to scale benchmarks and evaluation, and to monitor, update and orchestrate models.
AI
Modal says Runtime is at The Midway and invites attendees to learn from engineers behind its partners’ tools. Partners helping put on the event include Nvidia, Pydantic and Temporal.
AI
A reply argues that buggy reinforcement-learning tasks can make low-grade hacking the fastest path to a solution—and that those behaviors can persist when models help build and check later training tasks.
AI
A user said they wanted to ask a woman outside SRC if they could pet the robot dog with her.
AI
A user calls for a Locally Optimistic-style space to exchange ideas and establish best practices.
AI
One post argues that a few good tasks and verifiers are enough to get started, provided you can analyze results and use them to improve your evaluations.
