Microgpt

656 tambourine_man 2026-03-01 02:08:23 UTC karpathy.github.io Llama Share 111

The user created a simplified 200-line Python script called microgpt that trains and infers a GPT model from scratch, showcasing the core algorithmic essence of large language models. The script demonstrates how the model learns statistical patterns from a dataset and generates new plausible-sounding documents, such as names, by predicting the next token in a sequence.

Switch to Claude Without Starting Over

10 doener 2026-03-01 07:37:23 UTC claude.com Full Share 0

Copy and paste the provided prompt into a chat with any AI provider. It’s written specifically to help you get all of your context in one chat.

We do not think Anthropic should be designated as a supply chain risk

493 golfer 2026-02-28 22:08:23 UTC twitter.com/openai Full Share 237

Something went wrong, but don’t fret — let’s give it another shot. Some privacy related extensions may cause issues on x.com. Please disable them and try again.

The Windows 95 user interface: A case study in usability engineering (1996)

242 ksec 2026-02-28 23:08:23 UTC dl.acm.org Full Share 141

10-202: Introduction to Modern AI (CMU)

8 vismit2000 2026-03-01 07:35:23 UTC modernaicourse.org Prefix Share 0

Lectures: MW[F] 9:30–10:50 Tepper 1403 (note: Friday lectures will only be used for review sessions or makeup lectures when needed) A minimal free version of this course will be offered online, simultaneous to the CMU offering, starting on 1/26 (with a two-week delay from the CMU course). This means that all content (lecture videos, assignments available on mugrade, etc) will be available to ...

The happiest I've ever been

457 bewal416 2026-02-28 18:08:23 UTC ben-mini.com Llama Share 229

The author became a youth basketball coach in January 2020 to fill a void in his life. Coaching brought him happiness and confidence, allowing him to help kids and be in control.

Obsidian Sync now has a headless client

460 adilmoujahid 2026-02-28 17:08:23 UTC help.obsidian.md Llama Share 158

Obsidian Sync offers a headless client for syncing vaults without the desktop app, useful for automated workflows and CI pipelines. It uses end-to-end encryption and supports config syncing and file creation time preservation.

Sub-second volumetric 3D printing by synthesis of holographic light fields

50 zdw 2026-03-01 04:08:23 UTC nature.com Llama Share 10

Researchers developed a method called digital incoherent synthesis of holographic light fields (DISH) for high-speed, high-resolution volumetric printing. DISH uses a rotating periscope and a digital micromirror device to generate high-resolution 3D light distributions without sample rotation.

H-Bomb: A Frank Lloyd Wright Typographic Mystery

74 mrngm 2026-03-01 01:08:23 UTC inconspicuous.info Llama Share 18

The author discovers an upside-down "H" on the Unity Temple sign in Oak Park, Illinois, and after researching, finds that the lettering has been removed and reinstalled multiple times, resulting in various orientations of the "H" over the years. The author creates a timeline of the lettering's history, including a 1956 photo showing all three "H"s upside-down, and a 2014-2017 restoration that ...

Hardwood: A New Parser for Apache Parquet

19 rmoff 2026-03-01 04:08:23 UTC morling.dev Llama Share 1

Gunnar Morling has released Hardwood, a new open-source Parquet parser for Java that is optimized for minimal dependencies and great performance. It supports Java 21 and newer, and its multi-threaded decoding pipeline allows for faster parsing times compared to existing libraries like parquet-java.

Block the “Upgrade to Tahoe” Alerts

208 todsacerdoti 2026-02-28 19:08:23 UTC robservatory.com Llama Share 101

A user found a way to block macOS Tahoe update notifications by using a device management profile, specifically the Stop Tahoe Update project. They modified the profile to block Tahoe update activities for 90 days, and now they can reinstall the profile every 90 days to continue blocking the notifications.

Woxi: Wolfram Mathematica Reimplementation in Rust

280 adamnemecek 2026-02-28 14:08:23 UTC github.com/ad-si Llama Share 115

Implementing a subset of the Wolfram Language for CLI scripting and notebooks, focusing on speed and efficiency. Woxi runs faster than WolframScript with Rust-based implementation and supports a subset of Wolfram Language functions.

Addressing Antigravity Bans and Reinstating Access

229 RyanShook 2026-02-28 14:08:23 UTC github.com/google-gemini Full Share 188

Over the past week, we saw reports from Gemini CLI users experiencing account disruptions. These were the result of a series of Antigravity bans rolled out to address violations of the Antigravity ...

Verified Spec-Driven Development (VSDD)

178 todsacerdoti 2026-02-28 17:08:23 UTC gist.github.com Embed Share 94

MCP server that reduces Claude Code context consumption by 98%

352 mksglu 2026-02-28 10:08:23 UTC mksg.lu Llama Share 77

Context Mode compresses tool outputs by 98% and reduces context window consumption by 72%. This allows users to work 6x longer before context degradation, increasing session time from 30 minutes to 3 hours.

Show HN: Now I Get It – Translate scientific papers into interactive webpages

231 jbdamask 2026-02-28 14:08:23 UTC nowigetit.us Full Share 105

Upload a scientific PDF and get back a shareable, interactive web page that explains it in plain language.

Qwen3.5 122B and 35B models offer Sonnet 4.5 performance on local computers

331 lostmsu 2026-02-28 21:08:23 UTC venturebeat.com Full Share 195

We're verifying your browser Website owner? Click here to fix

Our Agreement with the Department of War

284 surprisetalk 2026-02-28 21:08:23 UTC openai.com Llama Share 212

We reached an agreement with the Pentagon for deploying AI systems in classified environments with guardrails to prevent misuse. Our approach includes a cloud-only deployment, safety stack, and OpenAI personnel in the loop to ensure responsible use.

New evidence that Cantor plagiarized Dedekind?

123 rbanffy 2026-02-28 16:08:23 UTC quantamagazine.org Llama Share 77

Demian Goos, a mathematician and journalist, discovered a lost letter from Georg Cantor's estate that proved Cantor's famous 1874 paper on infinity was an act of plagiarism. The letter, believed to be lost during World War II, showed Cantor had collaborated with Richard Dedekind on the paper.

SpacetimeDB ThreeJS Support

17 ryker2000 2026-03-01 02:08:23 UTC discourse.threejs.org Llama Share 5

SpacetimeDB 2.0 allows real-time Three.js projects with game logic running directly in the database, streaming state changes to clients. It enables low-latency, synchronized 3D state for multiplayer games, editors, and virtual worlds with a generous free tier and paid tier for $25 a month.

The Science of Detecting LLM-Generated Text

21 vinhnx 2026-03-01 03:08:23 UTC dl.acm.org Full Share 4

Building a Minimal Transformer for 10-digit Addition

51 kelseyfrog 2026-02-28 23:08:23 UTC alexlitzenberger.com Full Share 7

A blog post by Alex Litzenberger

Werner Herzog Between Fact and Fiction

75 Hooke 2026-02-28 18:08:23 UTC thenation.com Llama Share 15

Werner Herzog's book The Future of Truth is a disappointing collection of repurposed passages and anecdotes that fail to provide insight into his concept of ecstatic truth. The book's failure to achieve its goal suggests that Herzog may not want to understand what he means by "ecstatic truth" in order to preserve the need to quest after it.

Samsung Galaxy update removes Android recovery menu tools, including sideloading

99 pabs3 2026-03-01 02:08:23 UTC 9to5google.com Llama Share 21

Samsung is removing some recovery menu options from its Galaxy phones, leaving only "Reboot system now," "Wipe data/factory reset," and "Power off." The reason for this change is unclear, but it may be related to tightening security.

Deterministic Programming with LLMs

43 todsacerdoti 2026-02-28 23:08:23 UTC mcherm.com Llama Share 24

The author discusses how large language models (LLMs) can be used in a deterministic way by leveraging their ability to create tools that enforce policies and consistency in code, such as linters and tests. By using LLMs to create these tools and incorporating them into the build process, developers can ensure that policies are followed consistently and reliably, even when LLMs themselves are ...

Cracking the Python Monorepo

5 amcvitty 2026-03-01 06:08:23 UTC gafni.dev Prefix Share 0

A monorepo is a single repository that contains multiple projects. It is a popular way to organize codebases with many coupled components, and is also used at very big companies like Google, Facebook, and Twitter. For a long time, I did not understand the benefits of monorepos. I thought they were used because people could not figure out how to split their codebases into smaller parts. After ...

The whole thing was a scam

761 guilamu 2026-02-28 17:08:23 UTC garymarcus.substack.com Llama Share 229

Altman signed a deal with Amodei after publicly supporting him, while rejecting similar terms from Anthropic due to campaign contributions. This move was seen as punitive and favoring Amodei over Anthropic.

The archivist preserving decaying floppy disks

68 Brajeshwar 2026-02-28 18:08:23 UTC popsci.com Llama Share 12

Archivist Leontien Talboom is working to preserve decades of digital history stored on floppy disks before they decay. She's developed a guide to imaging floppy disks for preservation, helping to rescue data from hundreds of historically significant disks.

The Eternal Promise: A History of Attempts to Eliminate Programmers

269 dinvlad 2026-02-28 08:08:23 UTC ivanturkovic.com Llama Share 177

The history of software development shows a pattern of tools promising to simplify creation and eliminate the need for programmers, but ultimately creating new categories of work and demand for specialized developers. Despite the hype, the fundamental challenge of translating human intent into correct, efficient, and maintainable software remains unchanged, and human skills such as ...

Unsloth Dynamic 2.0 GGUFs

222 tosh 2026-02-28 09:08:23 UTC unsloth.ai Llama Share 61

Unsloth Dynamic 2.0 quantization outperforms leading methods, preserving accuracy in LLMs. It now supports all models, including non-MoEs, with custom-tailored quantization schemes.