VVashni Roy·4h agoWhy 0.5 Isn't Always the Right Classification ThresholdI changed the threshold of the same churn model from 0.3 to 0.7. The model stayed the same, but its predictions changed quite a lot.3 min read·vashniroy.hashnode.dev
NNagappan S·5h agoSemantic Caching: Skipping the Model Call When You've Already Answered the QuestionSemantic Caching7 min read·nagspidey.hashnode.dev
SSiddhesh Sawant·5m agoLogistic Regression from First Principles: Log-Odds, Likelihood and the Decision ThresholdLogistic regression derived step by step: sigmoid and log-odds, log loss and its gradient, gradient descent and Newton's method, and the decision threshold.53 min read·siddheshsawant.hashnode.dev
SSiddhesh Sawant·16m agoLinear Regression from First Principles: Least Squares, Geometry and Gradient DescentFitting a straight line is the oldest trick in machine learning and still the one most worth understanding completely. This article builds linear regression from a five-point example to the normal equations, shows why least squares is a projection and a maximum-likelihood estimate at the same time, trains it with gradient descent you can steer, and then puts it to work on a real dataset to see where it holds up and where it breaks.48 min read·siddheshsawant.hashnode.dev
EEmbrel Grim·8h agoan attempt at hypernetworks on an RX 6600Where the Idea Came From I realized that my hardware couldn't run anything decent, so I decided to play it safe—buy proper components? Of course not—build my own architecture! Full of confidence, I sa10 min read·grimoire.hashnode.dev2
Wwanqiuxu·1h agoWhy language models invent academic citations that look realAsk a language model for a short literature review, or for "a few key papers," and you may get a list that looks professionally done: surnames, years, journal names, tidy titles. Then you search for o4 min read·crushsci.hashnode.dev
YYong Yu·1h agoMuse Ember at $0.10/$0.50: Default Is Max, the 200K Billing Cliff, and When to Pair It With WonderMeta Muse Ember is on the API today as muse-ember-1.0: $0.10/$0.004/$0.50 under 200K input, 3× rates on the whole request past that line, default reasoning is max, AA Index ~36. No chat call on this box.10 min read·yongboyu.hashnode.dev
CCaraComp·2h agoDeepfake Election: Taiwan Warns Fakes Can Lie About VotingAnalyzing the technical fallout of synthetic media in Taiwan's elections highlights a critical shift in information integrity: malicious actors no longer need to construct elaborate scandal narratives3 min read·caracomp.hashnode.dev
Jjovin george·2h agoClaude Fable 5.2 vs Mythos 5.2: Speed, Benchmarks & Real CostsAnthropic's frontier model deep dive: $20 Pro plan trial, FrontierCode 2.0 hack, and silent fallbacks.6 min read·softreviewed.hashnode.dev
Mmihalykrich·7h agoBuilding KRXY — Part 8: When the Network Slows Down — and the Model Library Grows1–2 November 2026: two-Pi learning through bandwidth limits and a two-minute outage, followed by the first Qwen3.5 2B and 4B installations.13 min read·krxy.hashnode.dev
