frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

A guide to local coding models

https://www.aiforswes.com/p/you-dont-need-to-spend-100mo-on-claude
208•mpweiher•4h ago•96 comments

I'm just having fun

https://jyn.dev/i-m-just-having-fun/
145•lemper•5d ago•44 comments

Disney Imagineering Debuts Next-Generation Robotic Character, Olaf

https://disneyparksblog.com/disney-experiences/robotic-olaf-marks-new-era-of-disney-innovation/
72•ChrisArchitect•4h ago•30 comments

Show HN: Books mentioned on Hacker News in 2025

https://hackernews-readings-613604506318.us-west1.run.app
338•seinvak•9h ago•127 comments

ONNX Runtime and CoreML May Silently Convert Your Model to FP16

https://ym2132.github.io/ONNX_MLProgram_NN_exploration
12•Two_hands•1h ago•2 comments

Deliberate Internet Shutdowns

https://www.schneier.com/blog/archives/2025/12/deliberate-internet-shutdowns.html
9•WaitWaitWha•3d ago•1 comments

Show HN: WalletWallet – create Apple passes from anything

https://walletwallet.alen.ro/
289•alentodorov•9h ago•88 comments

The Going Dark initiative or ProtectEU is a Chat Control 3.0 attempt

https://mastodon.online/@mullvadnet/115742530333573065
459•janandonly•7h ago•143 comments

Evaluating chain-of-thought monitorability

https://openai.com/index/evaluating-chain-of-thought-monitorability/
24•mfiguiere•2d ago•3 comments

Show HN: Autograd.c – A tiny ML framework built from scratch

https://github.com/sueszli/autograd.c
53•sueszli•5d ago•5 comments

I program on the subway

https://www.scd31.com/posts/programming-on-the-subway
170•evankhoury•5d ago•106 comments

CO2 batteries that store grid energy take off globally

https://spectrum.ieee.org/co2-battery-energy-storage
141•rbanffy•10h ago•119 comments

Autoland saves King Air, everyone reported safe

https://avbrief.com/autoland-saves-king-air-everyone-reported-safe/
109•bradleybuda•8h ago•48 comments

E.W.Dijkstra Archive

https://www.cs.utexas.edu/~EWD/welcome.html
109•surprisetalk•10h ago•8 comments

You’re not burnt out, you’re existentially starving

https://neilthanedar.com/youre-not-burnt-out-youre-existentially-starving/
210•thanedar•7h ago•222 comments

I can't upgrade to Windows 11, now leave me alone

https://idiallo.com/byte-size/cant-update-to-windows-11-leave-me-alone
346•firefoxd•7h ago•324 comments

Rue: Higher level than Rust, lower level than Go

https://rue-lang.dev/
81•ingve•5h ago•54 comments

ARIN Public Incident Report – 4.10 Misissuance Error

https://www.arin.net/announcements/20251212/
134•immibis•10h ago•35 comments

Structured outputs create false confidence

https://boundaryml.com/blog/structured-outputs-create-false-confidence
116•gmays•10h ago•58 comments

Perron: A Static Site Generator for Ruby on Rails

https://perron-site.statichost.page/
7•Kerrick•3d ago•0 comments

Ruby website redesigned

https://www.ruby-lang.org/en/
359•psxuaw•18h ago•138 comments

Get an AI code review in 10 seconds

https://oldmanrahul.com/2025/12/19/ai-code-review-trick/
91•oldmanrahul•8h ago•49 comments

Indoor tanning makes youthful skin much older on a genetic level

https://www.ucsf.edu/news/2025/12/431206/indoor-tanning-makes-youthful-skin-much-older-genetic-level
217•SanjayMehta•20h ago•160 comments

Coarse is better

https://borretti.me/article/coarse-is-better
178•_dain_•12h ago•95 comments

Why “negative vectors” can't delete data in FAISS – but weighted kernels can

https://github.com/nikitph/bloomin/tree/master/negative-vector-experiment
9•loaderchips•4d ago•1 comments

Waymo halts service during S.F. blackout after causing traffic jams

https://missionlocal.org/2025/12/sf-waymo-halts-service-blackout/
190•rwoll•20h ago•278 comments

Danish postal service to stop delivering letters after 400 years

https://www.theguardian.com/world/2025/dec/21/denmark-postnord-postal-delivery-letters-society
9•hackerbeat•26m ago•0 comments

Three ways to solve problems

https://andreasfragner.com/writing/three-ways-to-solve-problems
116•42point2•11h ago•21 comments

I wish people were more public

https://borretti.me/article/i-wish-people-were-more-public
12•swah•2h ago•5 comments

Show HN: Shittp – Volatile Dotfiles over SSH

https://github.com/FOBshippingpoint/shittp
115•sdovan1•13h ago•64 comments
Open in hackernews

Why “negative vectors” can't delete data in FAISS – but weighted kernels can

https://github.com/nikitph/bloomin/tree/master/negative-vector-experiment
9•loaderchips•4d ago
The fix for machine unlearning in vector databases turns out to be conceptually simple, but it requires changing the semantics of retrieval.

Standard FAISS-style indices store vectors and compute:

argmax ⟨q, vᵢ⟩

If you insert -v, nothing happens. It’s just another point. The original vector is still maximally similar to itself and remains rank-1.

This isn’t a bug—it’s a consequence of selection-based retrieval.

If instead you store (vector, weight) pairs and evaluate: φ(q) = Σ wᵢ · K(q, vᵢ)

you get a different object entirely: a field, not a selection. Now inserting the same vector with w = −1 causes destructive interference. The contribution cancels. The attractor disappears.

Deletion becomes O(1) append-only (add the inverse), not a structural rebuild.

FAISS-style: Vec<Vec<f32>> → argmax (selection) Weighted form: Vec<(Vec<f32>, f32)> → Σ (field)

We validated this on 100k vectors: • FAISS: target stays rank-1 after “deletion” • Field-based model: exact cancellation (φ → 0), target unretrievable

The deeper point is that this isn’t a trick—it’s a semantic separation. • FAISS implements a selection operator over discrete points. • The weighted version implements a field operator where vectors act as kernels in a continuous potential. • Retrieval becomes gradient ascent to local maxima. • Deletion becomes destructive interference that removes attractors.

This shifts deletion from structural (modify index, rebuild, filter) to algebraic (append an inverse element). You get append-only logs, reversible unlearning, and auditable deletion records. The negative weight is the proof.

Implication: current vector DBs can’t guarantee GDPR/CCPA erasure without reconstruction. Field-based retrieval can—provably.

Paper with proofs: https://github.com/nikitph/bloomin/blob/master/negative-vect...

Comments

jey•1h ago
That makes sense, but how do you efficiently evaluate the Gaussian kernel based approach (“operator-based data structures (OBDS)”)? Presumably you want to do it in a way that keeps a dynamically updating data structure instead of computing a low rank approximation to the kernel etc? In my understanding the upside of the kNN based approaches are fast querying and ability to dynamically insert additional vectors..?