Blog post argues Anthropic's Scion draws heavy inspiration from Mononoke

·3 views

A post on sunshowers.io claims that Anthropic's Scion is heavily inspired by the anime film Mononoke. The source provided only the headline and link, so the specific arguments and evidence are not available. The item drew 13 points and one comment on Hacker News at the time it was captured. A full summary requires the article text.

Read the full story at Hacker News

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Go 1.26 Lets new() Take an Expression, Making Pointer Helpers Obsolete

Go 1.26, released February 10, 2026, extends the built-in new function to accept an expression as well as a type. This means new(42) or new(time.Now()) now compiles, letting developers create pointers to simple values directly. The change resolves a long-standing issue first proposed in 2021 and closed in October 2025. Developers who relied on helper functions such as ptr[T] or utility packages can often remove them, with go fix handling most of the migration. Libraries must declare Go 1.26 or later in go.mod to use the feature, which raises their minimum supported version.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

LLM Wiki vs. RAG: Measured Test Shows Wiki Slower Than Retrieval on Codebase Questions

A developer built a markdown-based LLM wiki for the itsdangerous signing library and compared it with agentic retrieval over the raw source on four integration questions. Using Claude Code, the team measured wall-clock time, turns, and cost for each question with and without the wiki. Without the wiki, answers were faster and cheaper while equally correct. The wiki's advantage appeared only on a question whose answer was not contained in the source material. The author argues the two approaches differ in that RAG retrieves what was written, while a wiki also records decisions.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

VIDRAFT Pre-Trains a 46M-Parameter Language Model on Inference-Only NPUs, Without GPUs

VIDRAFT has pre-trained a small language model called VKORE-46M-NPU entirely on a first-generation inference-only NPU, avoiding GPUs at every training stage. The 46-million-parameter model was trained from random initialization on a public dataset and can generate short English stories. The company implemented backpropagation through its own VKORE software layer, since the chip's vendor offers no training stack for it. A matched GPU run showed a 0.99 correlation in loss trajectories, indicating numerically consistent training, though the company stresses this is not a claim of matching GPU training speed. The model and comparison materials are publicly available on Hugging Face, with detailed implementation methods promised in a forthcoming paper.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

ScopeLedger: a local AI tool for spotting out-of-scope client requests

ScopeLedger is an open-source tool built by a developer for freelancers and small agencies to handle 'one more thing' requests that expand fixed-price projects. Users paste agreed deliverables alongside new client requests, and a locally run Gemma model suggests whether each ask is included, may be extra work, or needs discussion, with checkable source lines. The owner reviews and overrides every classification, sets the rate and effort, and exports a change-order draft without sending invoices or assuming client agreement. The author plans a $19 one-time local tool but says the price is unvalidated and no paying users have been acquired yet. Evaluation on eight synthetic development cases showed Gemma 3 4B matching six, a result the author notes is not a held-out benchmark.

0 comments Read more at DEV Community