Never Give Them Your Face

463 audiodude 2026-06-22 14:42:17 UTC nevergivethemyourface.com Llama Share 258

Governments and platforms are demanding age verification through facial scans and ID checks, claiming it's for child protection, but it's actually a way to track everyone's identity. This system fails to stop determined teenagers and can even put children in danger by creating a "children index."

Pledging Another $400k to the Zig Software Foundation

452 tosh 2026-06-22 14:42:17 UTC mitchellh.com Llama Share 136

The author's family is pledging $400,000 to the Zig Software Foundation, bringing their total support to $700,000. They respect Zig's technical progress, community approach, and unique philosophy, despite having differing opinions on some policies.

Claude Code's "extended thinking" is a summary- not authentic thinking

124 0o_MrPatrick_o0 2026-06-22 14:42:17 UTC patrickmccanna.net Llama Share 90

Claude Code's reasoning logs are incomplete and indirect, containing a signature and no text. The "extended-thinking" output is a summary, not the actual thinking process, and is not accessible for auditing.

Moebius: 0.2B image inpainting model with 10B-level performance

71 DSemba 2026-06-22 14:42:17 UTC hustvl.github.io Llama Share 13

Moebius achieves high-quality inpainting with 0.22B parameters, 15x inference speedup, and on-par/surpassing 10B-level models in 6 benchmarks. It's a task-specific specialist that's smarter, lighter, and faster than bloated generalists.

Deno Desktop

817 GeneralMaximus 2026-06-22 05:42:17 UTC docs.deno.com Llama Share 319

Deno desktop turns web projects into self-contained desktop apps with small binaries and full Node compatibility. It bundles code, Deno runtime, and web rendering engine into redistributable binaries for macOS, Windows, and Linux.

Codex logging bug may write TBs to local SSDs

324 vantareed 2026-06-22 07:42:17 UTC github.com/openai Llama Share 180

Codex SQLite logs are the main continuous writer, consuming 640 TB/year and potentially a full SSD drive in less than a year. Filtering categories like TRACE logs and telemetry logs can remove 96% of retained log bytes.

Die analysis of the 8087 math coprocessor's fast bit shifter (2020)

32 Jimmc414 2026-06-22 13:42:17 UTC righto.com Llama Share 7

The 8087 floating-point coprocessor was introduced in 1980 by Intel, making floating-point operations up to 100 times faster. It was a huge benefit for applications such as AutoCAD and flight simulators.

GLM 5.2 vs. Opus

359 ritzaco 2026-06-22 07:42:17 UTC techstackups.com Llama Share 252

GLM-5.2 is a strong open model that can perform long-horizon tasks at a fraction of the cost of closed models like Claude Opus 4.8, but it's text-only and can't verify its work visually. GLM-5.2 is suitable for tasks that are mostly text and logic, while Opus is better for tasks that require correctness, polish, and visual judgment, and GLM-5.2 should be kept in the arsenal due to its ...

Nintendo Wii U games running from a 1980's Bernoulli disk [video]

14 zdw 2026-06-22 15:42:17 UTC youtube.com Embed Share 1

Granularity comes at a cost – Game Theory

24 sidhantbansal 2026-06-22 13:42:17 UTC sidhantbansal.com Llama Share 4

The user initially thought that more granularity in a system is always a good thing, but two examples showed that finer choices can create new ways to game priority or reduce the incentive to participate honestly. The user learned that there is a sweet spot for granularity, where it allows price competition but not so small that queue priority becomes meaningless and liquidity providers stop ...

Help I accidentally a wigglegram

410 gregsadetsky 2026-06-22 02:42:17 UTC lmao.center Llama Share 95

You created a script to use perceptual hashing to find similar images in your camera roll, automatically identifying potential wigglegrams. The script calculates the hamming distance between image pairs to find a threshold, helping to distinguish between intentional and accidental shots.

Did my old job only exist because of fraud?

749 advisedwang 2026-06-21 21:42:17 UTC david.newgas.net Llama Share 338

The author worked at GenieDB, a UK startup acquired by US VC fund Frost VP, and later discovered the fund was sued for fraud by the SEC. The author questions if their job and life changes were a result of the alleged fraud.

DHL Set to Transport Goods on New Wind-Powered Cargo Ships

59 julienchastang 2026-06-22 15:42:17 UTC wsj.com Full Share 20

Please enable JS and disable any ad blocker

Apertus – Open Foundation Model for Sovereign AI

496 T-A 2026-06-21 21:42:17 UTC apertvs.ai Llama Share 167

Apertus is an open-source AI model developed by Swiss AI Initiative, meeting EU AI Act requirements with opt-out respect and data protection. It's a multilingual model trained on 1000+ languages, competitive with top open models.

Munich 1991: The Roots of the Current AI Boom

172 tosh 2026-06-22 06:42:17 UTC people.idsia.ch GoogleT5 Share 74

The foundations of this trillion-dollar industry were laid down over 30 years ago in munich. these four techniques power today's most advanced llms. lstm is the most cited ai paper of the 20th century, based on deep residual learning through residual nn connections. the year 1991 also marks the emergence of defining features of highway net - a 10 times deeper than previous feedforward ...

Maria Isabel Sánchez Vegara on Her 100th "Little People, Big Dreams" Book

27 zeristor 2026-06-22 12:42:17 UTC amightygirl.com Llama Share 4

Maria Isabel Sánchez Vegara created the Little People, BIG DREAMS series to inspire kids with stories of remarkable women. The series has sold over 7.5 million copies worldwide and features 100 books, including a recent one about Ugandan climate activist Vanessa Nakate.

Why Drawing Tablet Brands Won't Collaborate on Linux Floss Drivers

137 Tomte 2026-06-22 12:42:17 UTC davidrevoy.com Llama Share 60

You've been trying to get drawing tablet brands to collaborate on GNU/Linux drivers, but Gaomon decided not to participate due to branding concerns. You'll continue reviewing tablets and documenting specs, one by one, with the help of Peter and Benjamin.

There is minimal downside to switching to open models

335 amarble 2026-06-21 21:42:17 UTC marble.onl Llama Share 281

The author notes that using Linux has become more viable due to improved compatibility and software ecosystem. However, they express concern about the growing penalty for using open LLM models due to performance and trust issues.

Nvidia Halos

62 ilreb 2026-06-22 14:42:17 UTC nvidia.com Llama Share 33

NVIDIA Halos is a comprehensive safety system for autonomous vehicles, unifying vehicle architecture, AI models, and software to ensure safe development and deployment. It provides a unified safety framework, integrating functional safety, cybersecurity, AI, and regulations to ensure system-level reliability and iterative improvement.

Manticore Search 27.1.5: Auth, sharding, conversational and faster vector search

30 snikolaev 2026-06-22 10:42:17 UTC manticoresearch.com Llama Share 1

Manticore Search 27.1.5 adds built-in auth, sharded tables, conversational search, and performance improvements. It requires staged rollout and careful replication setup due to significant changes.

Rive, Fast and reliable background jobs in Go

19 mountainview 2026-06-22 10:42:17 UTC github.com/riverqueue Llama Share 4

River is a job queue library for Postgres that enqueues jobs transactionally with database changes. It provides a client interface for job insertion and management, with support for workers, subscriptions, and telemetry.

Chevron signs 20-year power agreement with Microsoft for West Texas data center

44 cdrnsf 2026-06-22 14:42:17 UTC chevron.com Llama Share 40

Chevron Corporation's forward-looking statements may differ from actual results due to various risks and uncertainties. These risks include changing oil prices, government policies, and global economic conditions.

Investors get real-time view of UK bond market activity for the first time

82 monkeydust 2026-06-22 07:42:17 UTC fca.org.uk Llama Share 60

The UK bond market now has a single, real-time source of prices and trading activity thanks to the launch of its bond consolidated tape operated by ETS Connect UK. The service provides a comprehensive view of market activity with 98% market coverage and is supervised by the FCA for data quality and reliability.

My 1992 view of the problems of computer programming in 1992

82 speckx 2026-06-22 07:42:17 UTC blog.plover.com Llama Share 37

IBM initially invested in a new FORTRAN compiler, but now compiler quality is not the main issue. The real problem is that programmers lack good methods, languages, and thinking skills to effectively write computer programs.

Sakana Fugu

189 Finbarr 2026-06-22 02:42:17 UTC sakana.ai Prefix Share 106

One Model to Command Them All マルチエージェントを指揮する、一つのモデル Frontier-level performance without single-vendor dependency. Fugu dynamically orchestrates the world's best models to tackle complex, multi-step tasks. Plug collective intelligence directly into your workflows today with a single API. Sakana Fugu ...

Memory Safe Inline Assembly

160 pizlonator 2026-06-22 01:42:17 UTC fil-c.org Llama Share 38

Fil-C supports memory-safe inline assembly for various use cases such as constant-time crypto, SIMD intrinsics, and atomic operations, while rejecting unsafe uses like memory accesses and control flow. The implementation involves parsing and analyzing assembly instructions, constraints, and clobbers, and uses a database of acceptable instructions and their effects to ensure safety.

Good results fine tuning a local LLM like Qwen 3:0.6B to categorize questions

196 dev-experiments 2026-06-21 23:42:17 UTC teachmecoolstuff.com Llama Share 37

The user is developing a chatbot that uses a local LLM to categorize household-related questions, and they tested the performance of a 0.6B parameter model through prompting and fine-tuning, achieving 92% accuracy after tweaking the finetuning approach. The model's performance improved significantly after mapping categories to two-character opaque IDs with no semantic overlap, reducing ...

Everything is logarithms

280 E-Reverance 2026-06-21 21:42:17 UTC alexkritchevsky.com GoogleT5 Share 60

The dimension of a vector space is the cardinality of its basis. [log f = by x] gives something that at least resembles the function's representation as scalar relation, but i suspect these are merely confusions of concepts. when you take general covariance to its extreme you end up asking that all of your mathematics be formulated in an explicit co-variant way, and that

Show HN: I rebuilt the only parts of my IDE I use, in Rust, over a weekend

16 kyle-ssg 2026-06-22 13:42:17 UTC github.com/kyle-ssg Llama Share 6

Kyde is a fast native Git client and code editor for macOS, Linux, and Windows. It offers a familiar interface, side-by-side diff, and syntax highlighting for various languages.

Alan Greenspan Dies at 100; Led Fed During Boom Before 2008 Bust

115 helsinkiandrew 2026-06-22 11:42:17 UTC bloomberg.com Full Share 96

Please make sure your browser supports JavaScript and cookies and that you are not blocking them from loading. For more information you can review our and .