Enabling the next iteration of the borrow checker on nightly

4 hours ago

The Polonius Alpha borrow checker is being enabled on Rust nightly as a precursor to stabilization in the coming months.
Polonius introduces flow-sensitive borrow checking, allowing more valid code to compile, such as code that conditionally returns borrows or uses match arms.
NLL (Non-Lexical Lifetimes) is flow-insensitive实战 and can be too conservative in such cases.
Performance regressions are generally minimal for most crates, but the worst case observed is a 2-3x slowdown for crates with many borrows.
Users can opt-out via -Zpolonius=off or configuration files, but they are encouraged to report issues if needed.
The goal is to stabilize Polonius Alpha by the end of the year, with ongoing work focused on performance, documentation, and addressing any reported issues.
After stabilization, the team plans to shift focus to other priorities, but may revisit feature work later.

zkSecurity built zkao, an AI audit agent designed to continuously find bugs in code until no AI-detectable ones remain.
An audit of Cloudflare's CIRCL library found seven real bugs, including critical issues in threshold RSA and attribute-based encryption, all now fixed.
Bugs included: float64 precision loss, DLEQ forgery, BLS missing message distinctness, DLEQ soundness break via sign collision, HPKE PSK bypass, int64 overflow in Lagrange coefficients, and CP-ABE access-control break.

Distillation of frontier AI models is cheap and hard to stop, threatening labs' business models.
Hiding chain of thought may not prevent distillation, especially for agentic tool use.
Companies can distill models by using user-approved sequences (e.g., gold diffs) as RL training targets.

Laguna S 2.1 is a 118B-parameter Mixture-of-Experts model with 8B active parameters per token and a 1M-token context window, developed in under nine weeks.
It achieves competitive scores against much larger models on long-horizon coding benchmarks, scoring 70.2% on Terminal-Bench 2.1 and 40.4% on DeepSWE v1.1.
All evaluation trajectories are publicly released at trajectories.poolside.ai for transparency and reproducibility.

The drone now handles single, dual, and some triple motor failures in simulation.
The author first trained a simulation-only policy, which helped identify issues quickly.
Initial attempts with a baseline PPO and high exploration failed due to increasing entropy and crashing.

<antirez>8 days ago

DotScale 2015 conference was interesting; work focused on Sentinel for connection sharing and scaling.
Redis 3.2 will include a Geo hashing API from Ardb, refactored by Matt Stancliff, merged as a priority.
Bloom filters will be added in 3.2, likely as a new type with API control for accuracy/space tradeoff and auto-depollution.

zkSecurity built zkao, an AI audit agent designed to continuously find bugs in code until no AI-detectable ones remain.
An audit of Cloudflare's CIRCL library found seven real bugs, including critical issues in threshold RSA and attribute-based encryption, all now fixed.
Bugs included: float64 precision loss, DLEQ forgery, BLS missing message distinctness, DLEQ soundness break via sign collision, HPKE PSK bypass, int64 overflow in Lagrange coefficients, and CP-ABE access-control break.

Distillation of frontier AI models is cheap and hard to stop, threatening labs' business models.
Hiding chain of thought may not prevent distillation, especially for agentic tool use.
Companies can distill models by using user-approved sequences (e.g., gold diffs) as RL training targets.

Laguna S 2.1 is a 118B-parameter Mixture-of-Experts model with 8B active parameters per token and a 1M-token context window, developed in under nine weeks.
It achieves competitive scores against much larger models on long-horizon coding benchmarks, scoring 70.2% on Terminal-Bench 2.1 and 40.4% on DeepSWE v1.1.
All evaluation trajectories are publicly released at trajectories.poolside.ai for transparency and reproducibility.

The drone now handles single, dual, and some triple motor failures in simulation.
The author first trained a simulation-only policy, which helped identify issues quickly.
Initial attempts with a baseline PPO and high exploration failed due to increasing entropy and crashing.

<antirez>8 days ago

DotScale 2015 conference was interesting; work focused on Sentinel for connection sharing and scaling.
Redis 3.2 will include a Geo hashing API from Ardb, refactored by Matt Stancliff, merged as a priority.
Bloom filters will be added in 3.2, likely as a new type with API control for accuracy/space tradeoff and auto-depollution.

Hasty Briefsbeta