Feed aggregator

Article URL: https://kowalski7cc.xyz/blog/short-lived-certificates/

Comments URL: https://news.ycombinator.com/item?id=48341313

Points: 1

# Comments: 1

Article URL: https://gizmodo.com/professional-sports-are-banning-smart-glasses-over-betting-concerns-2000765024

Comments URL: https://news.ycombinator.com/item?id=48341301

Points: 4

# Comments: 0

Article URL: https://matse.illinois.edu/news/85775

Comments URL: https://news.ycombinator.com/item?id=48341285

Points: 2

# Comments: 0

Article URL: https://stvrrll1ght.substack.com/p/your-body-keeps-the-score-the-gut

Comments URL: https://news.ycombinator.com/item?id=48341280

Points: 1

# Comments: 0

Comments URL: https://news.ycombinator.com/item?id=48341278

Points: 1

# Comments: 0

Article URL: https://cointelegraph.com/news/sec-hester-peirce-defends-crypto-privacy-tools-surveillance

Comments URL: https://news.ycombinator.com/item?id=48341271

Points: 3

# Comments: 0

Article URL: https://www.hendrick-hamel.henny-savenije.pe.kr/holland5.htm

Comments URL: https://news.ycombinator.com/item?id=48341253

Points: 3

# Comments: 1

Article URL: https://www.nytimes.com/2026/05/30/well/medical-aid-in-dying.html

Comments URL: https://news.ycombinator.com/item?id=48341247

Points: 3

# Comments: 0

Article URL: https://news.bloomberglaw.com/securities-law/citadel-loses-challenge-to-sec-approval-of-new-options-exchange

Comments URL: https://news.ycombinator.com/item?id=48341239

Points: 3

# Comments: 0

Article URL: https://gizmodo.com/starbucks-abandons-borked-ai-inventory-tool-that-couldnt-count-report-2000762252

Comments URL: https://news.ycombinator.com/item?id=48341210

Points: 3

# Comments: 0

Article URL: https://www.cnn.com/style/article/baikonur-buran-soviet-space-shuttle

Comments URL: https://news.ycombinator.com/item?id=48341195

Points: 4

# Comments: 0

MGM, Paramount Pictures and some independent labels still aren't part of Movies Anywhere, but Lionsgate has joined the MA fold, allowing its movies to sync across various digital retailers.

Article URL: https://www.nytimes.com/2026/05/30/business/china-lung-cancer-drugs-asco.html

Comments URL: https://news.ycombinator.com/item?id=48341176

Points: 3

# Comments: 1

Article URL: https://www.theverge.com/news/632613/andor-tony-gilroy-ai-star-wars-training-copyright

Comments URL: https://news.ycombinator.com/item?id=48341175

Points: 6

# Comments: 1

Article URL: https://twitter.com/rajveerbach/status/2060846974824255936/photo/1

Comments URL: https://news.ycombinator.com/item?id=48341143

Points: 3

# Comments: 1

Article URL: https://www.pcgamer.com/software/after-more-than-two-decades-the-creator-of-paint-net-finally-owns-the-domain-paint-net-and-its-all-thanks-to-a-slam-dunk-case-of-trademark-infringement/

Comments URL: https://news.ycombinator.com/item?id=48341127

Points: 5

# Comments: 0

Article URL: https://essenceia.github.io/thoughts/exceptional_engineers/

Comments URL: https://news.ycombinator.com/item?id=48341114

Points: 2

# Comments: 1

Article URL: https://github.com/IAMME543/Open-But-Polished

Comments URL: https://news.ycombinator.com/item?id=48341085

Points: 2

# Comments: 0

I built thaw because forking an LLM agent is absurdly wasteful today. When an agent explores N branches — RL rollouts, best-of-N, parallel coding attempts — each branch re-runs prefill over the same shared context. You pay for the same prompt N times.

thaw snapshots a live inference session — weights, KV cache, scheduler state, and the prefix-hash table — and hydrates N children that diverge from the fork point without re-prefilling. It's `git branch` for a running model.

The receipt (H100 80GB, Llama-3.1-8B, real hardware): a pre-warmed pool boots once in 22.3s, then each fork round of 4 branches × 64 tokens runs in 0.88s median. Cold-boot equivalent would be ~340s/round — ~400× amortized. All rounds bit-identical at the fork boundary. Full JSON receipt + reproducer in the repo, nothing hand-waved.

NVIDIA shipped Dynamo Snapshot last week for fast pod cold-starts — and they free the KV cache before checkpoint, by design. thaw is the opposite bet: preserve the KV cache so a fork is near-free. Different problem, opposite mechanic.

pip install thaw-vllm. Works with vLLM and SGLang, Apache-2.0.

https://github.com/thaw-ai/thaw

I'm a solo dev and this is the thing I most want feedback on: is the fork primitive the right shape, or do people want it wrapped in a framework(LangGraph/TRL) node instead? Happy to go deep on the KV-restore internals.

Comments URL: https://news.ycombinator.com/item?id=48341069

Points: 2

# Comments: 0