frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Sam Altman: I wonder why Anthropic would go for something so clearly dishonest

https://twitter.com/sama/status/2019139174339928189
1•doener•1m ago•0 comments

METR estimates that GPT-5.2 has a 50%-time-horizon of around 6.6 hrs

https://twitter.com/METR_Evals/status/2019169900317798857
1•tedsanders•1m ago•0 comments

Epistemological Fault Lines Between Human and Artificial Intelligence

https://arxiv.org/abs/2512.19466
1•DyslexicAtheist•3m ago•0 comments

Debian's Challenge When Its Developers Drift Away

https://www.phoronix.com/news/Debian-Developers-Quiet-Away
2•cuechan•5m ago•0 comments

I was just laid off by The Washington Post in the middle of a warzone

https://twitter.com/lizziejohnsonnn/status/2019083204133609846
3•doener•6m ago•0 comments

Anthropic Ad

https://twitter.com/Grantblocmates/status/2019093077936497031
2•doener•7m ago•0 comments

Building Reliable AI Infrastructure: What We Learned Scaling AI Visibility

https://amplitude.com/blog/scaling-ai-visibility
1•linksku•7m ago•0 comments

Show HN: Cradle Log – Free offline baby tracker with voice logging, AI insights

https://www.getcradlelog.com/
1•jack_burrr•8m ago•0 comments

An FPS built with Svelte, Threlte and Claude Opus built in just 2 hours

https://www.mr-spankys-meatballs.com
1•paulbjensen•9m ago•0 comments

Pinterest sacks two engineers for creating software to identify fired workers

https://www.theguardian.com/technology/2026/feb/04/pinterest-sacks-two-engineers-for-software-ide...
2•erehweb•11m ago•0 comments

Show HN: Cohesix 0.4.0-alpha, a no-std control-plane OS

https://github.com/lukeb-aidev/cohesix
2•Cohesix•13m ago•0 comments

Show HN: We simulated 10K freelancers deciding to work for AI agents

1•Mert_Predicts•14m ago•0 comments

Open-source AI tool beats LLMs in literature reviews – and gets citations right

https://www.nature.com/articles/d41586-026-00347-9
2•sohkamyung•15m ago•0 comments

Former Prime Ministers Harper and Chrétien Discuss Canada and the World [video]

https://www.youtube.com/watch?v=jhCacENdj7U
1•thomassmith65•16m ago•0 comments

Japan's Tourism Challenges: Declining Visitors and Shifting Trends in 2026

https://www.travelandtourworld.com/news/article/japans-tourism-challenges-declining-visitors-and-...
2•mikhael•17m ago•0 comments

AI and Higher Ed: An Impending Collapse

https://www.insidehighered.com/opinion/views/2025/07/24/ai-and-higher-ed-impending-collapse-opinion
1•talon8635•17m ago•0 comments

Ask HN: Where does operational truth live before it reaches "systems of record"?

2•former-aws•18m ago•0 comments

Show HN: LayerClaw – Observability tool for PyTorch training

2•prabhavsanga•18m ago•0 comments

"Grok, Is This True?" Analyzing LLM-Powered Fact-Checking on Social Media

https://osf.io/preprints/psyarxiv/85quw_v2
1•ytpete•20m ago•1 comments

Fast Autoscheduling for Sparse ML Frameworks

https://ajroot.pl/cgo2026scorch.html
1•matt_d•21m ago•0 comments

Show HN: WhookTown – Visualize your infrastructure as a 3D cyberpunk city

https://www.whook.town/
1•fralix•21m ago•0 comments

You don't want a faster Notion

https://outcrop.app/blog/speed
1•imedadel•21m ago•0 comments

AWS says you're on your own if media codec patent owners come knocking

https://www.theregister.com/2026/02/04/aws_codec_patent_holders/
2•ffworld•22m ago•1 comments

Show HN: BederSnake Revolution: Snake+Sokoban+Match3 puzzle with MLsolvable lvls

https://bedersnake.itch.io/bedersnake-revolution
2•avtomatron•23m ago•0 comments

Why Most of America Is Terrible at Making Biscuits (2018)

https://www.theatlantic.com/health/archive/2018/11/better-biscuits-south-thanksgiving/576526/
3•Mernit•24m ago•1 comments

Everything We Teach at Y Combinator in 10 Minutes [video]

https://www.youtube.com/watch?v=Pg72m3CjuK4
2•Brysonbw•27m ago•0 comments

Apple Beats Tech Stocks by Most in a Year as It Avoids AI Panic

https://finance.yahoo.com/news/apple-beats-tech-stocks-most-174832890.html
1•wslh•29m ago•0 comments

Show HN: I stopped trying to sleep on long-haul flights

https://www.flight-ready.online/
1•Zaleo•30m ago•2 comments

Portugal ruling party MPs seek social media ban for teens

https://macaubusiness.com/portugal-ruling-party-mps-seek-social-media-ban-for-teens/
1•belter•31m ago•0 comments

They Lied About Greenland. NASA Found This [video][25M]

https://www.youtube.com/watch?v=_uZj5PZB4-0
2•Bender•32m ago•0 comments
Open in hackernews

Ask HN: Mem0 stores memories, but doesn't learn user patterns

4•fliellerjulian•1h ago
We're a YC W23 company building AI agents for engineering labs - our customers run similar analyses repeatedly, and the agent treated every session like a blank slate.

We looked at Mem0, Letta/MemGPT, and similar memory solutions. They all solve a different problem: storing facts from conversations — "user prefers Python," "user is vegetarian." That's key-value memory with semantic search. Useful, but not what we needed.

What we needed was something that learns user patterns implicitly from behavior over time. When a customer corrects a threshold from 85% to 80% three sessions in a row, the agent should just know that next time. When a team always re-runs with stricter filters, the system should pick up on that pattern. So we built an internal API around a simple idea: user corrections are the highest-signal data. Instead of ingesting chat messages and hoping an LLM extracts something, we capture structured events — what the agent produced, what the user changed, what they accepted. A background job periodically runs an LLM pass to extract patterns and builds a confidence-weighted preference profile per user/team/org.

Before each session, the agent fetches the profile and gets smarter over time. The gap as I see it:

Mem0 = memory storage + retrieval. Doesn't learn patterns.

Letta = self-editing agent memory. Closer, but no implicit learning from behavior.

Missing = a preference learning layer that watches how users interact with agents and builds an evolving model. Like a rec engine for agent personalization.

I built this for our domain but the approach is domain-agnostic. Curious if others are hitting the same wall with their agents. Happy to share the architecture, prompts, and confidence scoring approach in detail.

Comments

berkethebooss•1h ago
for now we just use normal system memory the user can maintain himself - not the best solution but better than nothing.
fliellerjulian•1h ago
we did the same thing, but we noticed most users did not maintain their own system memory themself properly. so we had to build a solution that auto-manages memory and preferances.