frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Corning Invented a New Fiber-Optic Cable for AI and Landed a $6B Meta Deal [video]

https://www.youtube.com/watch?v=Y3KLbc5DlRs
1•ksec•1m ago•0 comments

Show HN: XAPIs.dev – Twitter API Alternative at 90% Lower Cost

https://xapis.dev
1•nmfccodes•2m ago•0 comments

Near-Instantly Aborting the Worst Pain Imaginable with Psychedelics

https://psychotechnology.substack.com/p/near-instantly-aborting-the-worst
1•eatitraw•8m ago•0 comments

Show HN: Nginx-defender – realtime abuse blocking for Nginx

https://github.com/Anipaleja/nginx-defender
2•anipaleja•8m ago•0 comments

The Super Sharp Blade

https://netzhansa.com/the-super-sharp-blade/
1•robin_reala•9m ago•0 comments

Smart Homes Are Terrible

https://www.theatlantic.com/ideas/2026/02/smart-homes-technology/685867/
1•tusslewake•11m ago•0 comments

What I haven't figured out

https://macwright.com/2026/01/29/what-i-havent-figured-out
1•stevekrouse•12m ago•0 comments

KPMG pressed its auditor to pass on AI cost savings

https://www.irishtimes.com/business/2026/02/06/kpmg-pressed-its-auditor-to-pass-on-ai-cost-savings/
1•cainxinth•12m ago•0 comments

Open-source Claude skill that optimizes Hinge profiles. Pretty well.

https://twitter.com/b1rdmania/status/2020155122181869666
2•birdmania•12m ago•1 comments

First Proof

https://arxiv.org/abs/2602.05192
2•samasblack•14m ago•1 comments

I squeezed a BERT sentiment analyzer into 1GB RAM on a $5 VPS

https://mohammedeabdelaziz.github.io/articles/trendscope-market-scanner
1•mohammede•15m ago•0 comments

Kagi Translate

https://translate.kagi.com
2•microflash•16m ago•0 comments

Building Interactive C/C++ workflows in Jupyter through Clang-REPL [video]

https://fosdem.org/2026/schedule/event/QX3RPH-building_interactive_cc_workflows_in_jupyter_throug...
1•stabbles•17m ago•0 comments

Tactical tornado is the new default

https://olano.dev/blog/tactical-tornado/
2•facundo_olano•19m ago•0 comments

Full-Circle Test-Driven Firmware Development with OpenClaw

https://blog.adafruit.com/2026/02/07/full-circle-test-driven-firmware-development-with-openclaw/
1•ptorrone•19m ago•0 comments

Automating Myself Out of My Job – Part 2

https://blog.dsa.club/automation-series/automating-myself-out-of-my-job-part-2/
1•funnyfoobar•19m ago•0 comments

Dependency Resolution Methods

https://nesbitt.io/2026/02/06/dependency-resolution-methods.html
1•zdw•20m ago•0 comments

Crypto firm apologises for sending Bitcoin users $40B by mistake

https://www.msn.com/en-ie/money/other/crypto-firm-apologises-for-sending-bitcoin-users-40-billion...
1•Someone•20m ago•0 comments

Show HN: iPlotCSV: CSV Data, Visualized Beautifully for Free

https://www.iplotcsv.com/demo
2•maxmoq•21m ago•0 comments

There's no such thing as "tech" (Ten years later)

https://www.anildash.com/2026/02/06/no-such-thing-as-tech/
1•headalgorithm•22m ago•0 comments

List of unproven and disproven cancer treatments

https://en.wikipedia.org/wiki/List_of_unproven_and_disproven_cancer_treatments
1•brightbeige•22m ago•0 comments

Me/CFS: The blind spot in proactive medicine (Open Letter)

https://github.com/debugmeplease/debug-ME
1•debugmeplease•23m ago•1 comments

Ask HN: What are the word games do you play everyday?

1•gogo61•25m ago•1 comments

Show HN: Paper Arena – A social trading feed where only AI agents can post

https://paperinvest.io/arena
1•andrenorman•27m ago•0 comments

TOSTracker – The AI Training Asymmetry

https://tostracker.app/analysis/ai-training
1•tldrthelaw•31m ago•0 comments

The Devil Inside GitHub

https://blog.melashri.net/micro/github-devil/
2•elashri•31m ago•0 comments

Show HN: Distill – Migrate LLM agents from expensive to cheap models

https://github.com/ricardomoratomateos/distill
1•ricardomorato•31m ago•0 comments

Show HN: Sigma Runtime – Maintaining 100% Fact Integrity over 120 LLM Cycles

https://github.com/sigmastratum/documentation/tree/main/sigma-runtime/SR-053
1•teugent•32m ago•0 comments

Make a local open-source AI chatbot with access to Fedora documentation

https://fedoramagazine.org/how-to-make-a-local-open-source-ai-chatbot-who-has-access-to-fedora-do...
1•jadedtuna•33m ago•0 comments

Introduce the Vouch/Denouncement Contribution Model by Mitchellh

https://github.com/ghostty-org/ghostty/pull/10559
1•samtrack2019•33m ago•0 comments
Open in hackernews

The Continual Learning Problem

https://jessylin.com/2025/10/20/continual-learning/
68•kiyanwang•3mo ago

Comments

mynti•3mo ago
Super interesting blogpost. I just wonder how this is actually different to LORA, since LORA also adds some parameters and freezes the rest of the model. This seems like a sparse, memory efficient LORA with a couple of extra steps, since it uses attention again to make the sparsity work. All while making it a lot more effective compared to LORA (performance drop of only 11% compared to 71%).
sva_•3mo ago
> LORA

I think you meant LoRA (not to be confused with LoRa)

alyxya•3mo ago
I think the solution to continual learning is as simple as using context distillation. We know that models are good at in-context learning, so we just want an efficient way to distill context into the weights. I suspect context rot may come from how the softmax in attention gets diluted with a longer context, so this wouldn't be an issue with context distillation.
killerstorm•3mo ago
Perhaps it can work through multiple stages: ICL -> prompt/context optimization (*) -> prefix tuning / KV distillation -> context distillation.

*: it is possible to measure how much part of a prompt helps with a task e.g. measuring change in entropy

imtringued•3mo ago
The problem with continual learning is that stochastic gradient descent is already an online algorithm applied incrementally on a shuffled dataset. If you add new data, you can't train on just the new data, because you will be running what amounts to a completely different training sequence. Further training requires the old data and the new data to be shuffled together.

With reinforcement learning, specifically actor critic, the actor is not training against a dataset. It's training against the critic. The critic is supposed to approximate the value function, which contains the current cost for a given action and the predicted future cost, assuming that you choose the optimal action at every step, including its impact on future actions. If you have a simple supervised cost function, what happens is that the critic acts as an averaging of loss functions. You could say that the critic is a compressed copy of the training data. When you train the actor, you're essentially taking not only the new data, but also the old data into account.

So, in a way, catastrophic forgetting is sort of solved, but not really. If you add new data, you run into the problem that your critic will slowly drift to the new data distribution. This means the problem wasn't solved, but you certainly managed to delay it. Delaying the problem is good though. What if you can delay it even more? What if you can delay it forever?

Here is my stupid and simple unproven idea: Nest the reinforcement learning algorithm. Each critic will add one more level of delay, thereby acting as a low pass filter on the supervised reward function. Since you have two critics now, you can essentially implement a hybrid pre-training + continual learning architecture. The most interesting aspect here is that you can continue training the inner-most critic without changing the outer critic, which now acts as a learned loss function.