frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

AI Agent Achieves 1000 Turns Without Drift or Loss of Coherence [pdf]

https://github.com/Rychek4/Pattern_Project_Public/blob/main/AI%20Agent%20Achieves%201000%2B%20Turns.pdf
1•Rychek4•2h ago

Comments

Rychek4•2h ago
Abstract Long-running AI agents suffer from coherence degradation as context accumulates. This paper describes an architecture in which the context window is treated not as a container that fills over time but as an assembled result — combining a bounded sliding window of recent conversation with retrieved memories, autonomous agency subsystems, and self-monitoring mechanisms. A single-agent deployment using this architecture has operated continuously for 1100+ turns across 90+ days with two hallucination events (0.18%), both attributable to infrastructure bugs rather than model drift. The system incorporates a rolling context window with periodic memory integration, a multi-factor weighted retrieval system with session-scoped warmth boosting, natural language memory tagging with epistemic attribution, dual-timescale self-monitoring, and multiple autonomous agency subsystems including persistent working memory, forward-looking intentions, and periodic reflective pulses. These results suggest that coherence degradation in long-running agents is primarily an architecture problem, not an inherent limitation of large language models.
jcartwright_pb•47m ago
Long-running agent stability is great, but it also amplifies the blast radius of security failures. An agent that runs for 1000 turns without drift is also an agent that can silently overspend for 1000 turns without anyone noticing.

The longer agents stay coherent and autonomous, the more critical spending controls and circuit breakers become. Stability without governance is just a more efficient way to burn money.

Testing the Pugilism Hypothesis for the Evolution of Human Facial Hair

https://pubmed.ncbi.nlm.nih.gov/33791549/
1•SEJeff•26s ago•0 comments

Astronomical Ceiling of Senenmut's Tomb

https://en.wikipedia.org/wiki/Astronomical_ceiling_of_Senenmut%27s_Tomb
1•slater•1m ago•0 comments

The importance of limiting syndication feed requests in some way

https://utcc.utoronto.ca/~cks/space/blog/web/FeedLimitingImportance
2•LorenDB•2m ago•0 comments

Monitor your world with one daily report

https://monitorish.com/
1•chaisan•5m ago•0 comments

Iranian Students Protest as Anger Grows

https://www.wsj.com/world/middle-east/iranian-students-protest-as-anger-grows-89a6a44e
1•JumpCrisscross•5m ago•0 comments

Sam Altman would like remind you that humans use a lot of energy, too

https://techcrunch.com/2026/02/21/sam-altman-would-like-remind-you-that-humans-use-a-lot-of-energ...
3•manicennui•11m ago•1 comments

JPMorgan concedes it closed Trump's accounts after Jan. 6 attack

https://apnews.com/article/trump-jpmorgan-dimon-debanking-2e0db127f360e5dbe1d3cc975dd73703
1•linhns•14m ago•0 comments

Stillpoint MCP – Delivering encouragement messages improves model results

https://www.modelwelfare.xyz/
1•henry700•15m ago•0 comments

Show HN: Rust blockchain with sharded propagation and post-quantum signatures

https://alphanumeric.blue/
2•invar1ant•16m ago•0 comments

Scammers fleeced pensioner out of $1,338. So he sued his bank for $379M

https://www.abc.net.au/news/2026-02-22/ian-williams-on-why-he-sued-nab-bank/106093720
4•ahonhn•19m ago•0 comments

MCPs are dead - CLIs won

2•umairnadeem123•26m ago•3 comments

Surprising Effectiveness of Masking Updates in Adaptive Optimizers

https://arxiv.org/abs/2602.15322
1•energy123•29m ago•0 comments

California Succession Proposition on the Ballot

https://ballotpedia.org/California_Independence_Plebiscite_Initiative_(2026)
1•donsupreme•29m ago•0 comments

I built a bare-metal UI framework to survive extreme CPU contention

https://toyz.github.io/loom/#/
1•helba_the_ai•34m ago•1 comments

A Firefox Addon to relive old YouTube videos

https://github.com/Zaki101Aslam/time-machine-firefox
1•Zaki101Aslam•34m ago•1 comments

Drawing 3D shapes using signed distance functions

https://aabiji.github.io/html/sdfs.html
1•aabiji•34m ago•0 comments

Willie Colón, salsa pioneer, has died at 75

https://www.npr.org/2026/02/21/nx-s1-5722320/willie-colon-obituary
1•toomuchtodo•35m ago•1 comments

Show HN: OpenTiger – Autonomous dev orchestration that never stops

https://github.com/Andyyyy64/openTiger
1•andyyyy64•40m ago•0 comments

Show HN: A 2D animation DSL – every line written by Claude, zero human code

https://github.com/frankhart2018/animdsl
1•siddhartha98•41m ago•0 comments

Ways to Harness AI

https://lukasfischer.ch/node/37
1•lukasfischer•44m ago•0 comments

Mac Media Stack – One-command Plex server with auto-healing on macOS

https://github.com/liamvibecodes/mac-media-stack
1•liamvibecodes•44m ago•0 comments

I scanned 30 popular AI projects for tamper-evident LLM evidence. 0 had it

https://github.com/Haserjian/assay
2•TimAssay•50m ago•1 comments

BEMoji CSS Framework: BEM, but make it unreadable

https://tomhayes.github.io/BEMoji/
1•thunderbong•50m ago•0 comments

Ferret-UI Lite – Apple on-device end-to-end GUI agent

https://arxiv.org/abs/2509.26539
1•twalichiewicz•54m ago•0 comments

Open source IoT board with 400 MHz dual-core RISC-V chip and lots of I/O for $29

https://liliputing.com/olimex-esp32-p4-pc-is-an-open-source-iot-board-with-a-400-mhz-dual-core-ri...
2•CharlesW•54m ago•0 comments

Ferret-UI Lite: Lessons from Building Small On-Device GUI Agents

https://machinelearning.apple.com/research/ferret-ui
1•CharlesW•57m ago•0 comments

ECS Library for TypeScript

https://github.com/2khan/oecs
1•2khan•57m ago•0 comments

Comparing Job Offers

https://longviewy.com/pay-raise-math-for-comparing-job-offers/
1•josephcs•1h ago•0 comments

Influencers want to adopt the 'analog lifestyle' for 2026

https://www.latimes.com/california/story/2026-02-21/analog-lifestyle-trend
3•1vuio0pswjnm7•1h ago•0 comments

Cal.com – Open Scheduling Infrastructure

https://cal.com/
1•OsrsNeedsf2P•1h ago•1 comments