frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

In Memoriam Marijn Meijles

https://vasilis.nl/nerd/2026/rip-marijn/
1•jandeboevrie•1m ago•0 comments

Anthropic AI Safety Research Warns of World in Peril in Resignation

https://www.forbes.com/sites/conormurray/2026/02/09/anthropic-ai-safety-researcher-warns-of-world...
1•jacquesm•2m ago•0 comments

Ask HN: How to Use `npx skills add` with On-Prem / Private Repos?

1•onurkanbkrc•3m ago•0 comments

Show HN: Intelligent skill selection system that reduces token consumption

https://github.com/onurkanbakirci/skills-gateway
1•onurkanbkrc•6m ago•0 comments

A Cosmic Miracle: A Remarkably Luminous Galaxy at z=14.44 Confirmed with JWST

https://astro.theoj.org/article/156033-a-cosmic-miracle-a-remarkably-luminous-galaxy-at-_z_-sub-s...
2•yread•7m ago•0 comments

Show HN: WinClaw – Open-source personal AI assistant that runs locally on any OS

https://github.com/itc-ou-shigou/winclaw
1•winclaw-dev•7m ago•0 comments

Fineimage.art – AI Image Generation and Editing Tool

https://fineimage.art
1•sjdeak•7m ago•1 comments

Steve Yegge on AI Agents and the Future of Software Engineering

https://newsletter.pragmaticengineer.com/p/steve-yegge-on-ai-agents-and-the
1•tosh•8m ago•0 comments

U.S. Senate bill exempts isolated power loads from FERC, DOE regulation

https://www.utilitydive.com/news/senate-bill-exempts-fully-isolated-large-loads-from-ferc-doe-reg...
1•walterbell•11m ago•0 comments

Show HN: A Guided Learning LLM

https://adaptive.bounded.cc
1•hirako2000•15m ago•0 comments

Windows Notepad App Remote Code Execution Vulnerability

https://msrc.microsoft.com/update-guide/vulnerability/CVE-2026-20841
1•rolisz•15m ago•0 comments

Show HN: Copy-and-patch compiler for hard real-time Python

https://github.com/Nonannet/copapy
1•Saloc•16m ago•0 comments

Show HN: DriftProof – Specification for preventing LLM behavioral drift

https://github.com/sarduine13-star/driftproof-risk-engine-
1•redwine13•17m ago•0 comments

MemOS OpenClaw Plugin Benchmark Results Are in Reduce 72% Token

https://twitter.com/MemOS_dev/status/2020854044583924111
1•MemTensor•18m ago•0 comments

A Stanford Experiment to Pair 5k Singles Has Taken over Campus

https://www.wsj.com/lifestyle/relationships/stanford-students-experiment-dating-date-drop-92a4aea8
2•erehweb•19m ago•0 comments

Show HN: Matchmaking where agents talk with agents to find compatible matches

https://github.com/skorotkiewicz/jupiter
1•modinfo•19m ago•0 comments

OpenAI Executive Who Opposed 'Adult Mode' Fired for Sexual Discrimination

https://www.wsj.com/tech/ai/openai-executive-who-opposed-adult-mode-fired-for-sexual-discriminati...
3•erehweb•20m ago•0 comments

The Security Checklist for Vibe Coders

https://asanchez.dev/blog/the-security-checklist-for-vibe-coders/
1•asanchezdev•24m ago•1 comments

FTX: Where Did the Money Go?

https://drive.google.com/file/d/1e2v-rcUqSFy4VI1MlWTLkJe-IeaFYBnl/edit
1•tosh•24m ago•0 comments

Show HN: Ledger – Yes, another blogging platform (but you own the database)

https://www.franzkafka.xyz/
1•mekod•25m ago•1 comments

A future without physics papers?

https://josephtoobysmith.com/cs/2026/02/11/Future-without-physics-papers.html
1•leanexplorer•27m ago•0 comments

Elegant Transducer Pipelines

https://old.reddit.com/r/Clojure/comments/1r179s1/the_transducer_that_ate_our_heap/
1•todsacerdoti•29m ago•0 comments

Flowspark

https://flowspark.net
1•ellebelle•31m ago•1 comments

Elegant Transducer Pipelines

https://gist.github.com/NicolasLambert/c3e51cb0b5314f2110161f85be24b4c7
1•todsacerdoti•31m ago•0 comments

I built Fluxer, a Discord-like chat app

https://blog.fluxer.app/how-i-built-fluxer-a-discord-like-chat-app/
3•todsacerdoti•40m ago•0 comments

Archive.today is directing a DDoS attack against my blog

https://gyrovague.com/
3•doener•44m ago•0 comments

Metabolic Acceleration and the Evolution (2016)

https://pubmed.ncbi.nlm.nih.gov/27144364/
1•stared•45m ago•0 comments

Transfer learning and Transformer models (ML Tech Talks) [video]

https://www.youtube.com/watch?v=LE3NfEULV6k
1•onurkanbkrc•45m ago•0 comments

Archive.today: Operator uses users for DDoS attack

https://www.heise.de/en/news/Archive-today-Operator-uses-users-for-DDoS-attack-11171455.html
3•doener•47m ago•0 comments

Three largest Dutch banks seek European alternatives to U.S. technology

https://nltimes.nl/2026/02/10/rabobank-ing-abn-amro-seek-european-alternatives-us-technology
3•belter•48m ago•0 comments
Open in hackernews

THE Replacement to RL for AI Agents. RL is legacy now.

https://cadenza-landing-qtu7gbjwb-akshparekh123-3457s-projects.vercel.app/
1•aparekh02•1h ago

Comments

aparekh02•1h ago
I’ve been building RL‑style agents at the NVIDIA DGX Hackathon and my job for a while, and I keep hitting the same wall: everyone wants agents that learn, almost nobody can afford real RL.

That's when I learned something crazy: -> Frontier training costs have grown about 2.4×/year since 2016, with single runs in the $30–40M range and projections crossing $1B this decade.

And what ends up happening is most agents in production end up replaying prompts and tools over a growing context window instead of improving reasoning capabilities.

===

My bet is that we’re aiming learning at the wrong target. A lot of the leverage is in how agents remember, not in constantly retraining the full policy. And with recent innovations in mem-alpha which treats memory construction as an RL problem and uses a small controller to maintain it, I have began to see a solution.

Instead of constantly retraining policies, I’m working on Cadenza v1.1—a mem‑alpha–style memory layer that learns how the agent remembers (what to keep, link, compress, forget) while leaving the base model mostly fixed.

This can create RL‑grade specialization, but it's driven by a small memory controller rather than huge PPO runs.

===

I’m looking for a few teams who are:

-> Running agents in production or serious pilots.

-> Feeling the “no real learning / RL too expensive” pain.

If this direction resonates—shifting learning into the memory layer rather than the policy—I’d love feedback or to talk through your use case.