AICodeKing

daniel.haxx.se

David Bombal

finout

Hacker News

InfoQ

<h2>Must know</h2>
<ul>
<li><strong>AI solves a previously unsolved math problem</strong>: GPT-5.2 Pro has reportedly solved Erdos problem #281, a mathematical challenge that human mathematicians had not cracked. Terence Tao, a highly respected mathematician, called this “perhaps the most unambiguous instance” of AI solving an open problem, marking a significant milestone for AI in mathematical research. This achievement, reported on January 18, 2026, highlights the growing capabilities of AI in complex problem-solving. <a href="https://app.daily.dev/posts/pRkd3CPEF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>China plans a massive satellite constellation</strong>: China has filed requests with the ITU for two satellite constellations, CTC-1 and CTC-2, totaling an astounding 193,428 satellites. This dwarfs SpaceX’s Starlink plans by four times. The US FCC views this expansion as a national security concern, accelerating approvals for domestic satellite networks. While China’s current deployment is limited, these filings, made in January 2026, could be a strategic move to secure orbital and spectrum rights, as reported by various sources. <a href="https://app.daily.dev/posts/9Dqo8zul1" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Snap Store publisher accounts hijacked for malware distribution</strong>: Scammers are exploiting the Canonical Snap Store by registering expired domains previously used by legitimate publishers, then using these hijacked accounts to distribute malicious updates to trusted snap applications. This new tactic, specifically targeting cryptocurrency wallet users, undermines trust in the Snap Store’s reputation signals. Users are advised to be extremely cautious with crypto apps from Snap and publishers should secure their domains and enable 2FA, according to cybersecurity reports. <a href="https://app.daily.dev/posts/8RJLmL5a0" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Physical AI and AV industry shifts at CES 2026</strong>: CES 2026 highlighted Physical AI, with Mobileye acquiring Mentee Robotics for $900 million, and Chinese automotive companies dominating the AV tech presence while US automakers were absent. <a href="https://app.daily.dev/posts/MtQzzuuIN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New CLI tool for Figma and AI integration</strong>: A new command-line tool, dannote/figma-use, offers full read/write access to Figma, enabling AI agents and developers to programmatically create and modify designs using JSX syntax, optimized for LLMs. <a href="https://app.daily.dev/posts/DfAoRLrdU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Iran disrupts Starlink internet with GPS jamming</strong>: Iran used military-grade GPS jammers to disrupt Starlink internet, cutting performance by up to 80% during a nationwide internet shutdown, raising concerns about Starlink’s reliability in conflict zones. <a href="https://app.daily.dev/posts/VVK9EFJ8w" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Open source models: Cathedral, megachurch, and bazaar</strong>: Open source projects are categorized into cathedral (exclusive groups), megachurch (large foundations like Linux Foundation), and bazaar (individual maintainers), with bazaar projects forming critical infrastructure but facing significant funding challenges. <a href="https://app.daily.dev/posts/rr8Vo8FaG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google’s Antigravity expands beyond coding</strong>: Google’s Antigravity is now competing with Anthropic’s Co-work by adding productivity features like creating presentations and organizing files directly from the IDE, leveraging Gemini 3 Flash for non-coding tasks. <a href="https://app.daily.dev/posts/HVlI4ECqg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>RE2: A simple AI breakthrough for LLMs</strong>: RE2 (prompt repetition) significantly improves LLM performance by duplicating the input prompt, addressing causal masking in transformers and showing consistent gains across models and benchmarks, especially in retrieval tasks. <a href="https://app.daily.dev/posts/qNNQkTkyP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pure C inference for Flux 2 image generation model</strong>: A pure C implementation of the FLUX.2-klein-4B image generation model now runs without Python, PyTorch, or CUDA, supporting text-to-image and image-to-image generation, with the entire codebase reportedly generated by Claude AI. <a href="https://app.daily.dev/posts/NsHswJDic" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI-generated code: Still getting memory allocation wrong</strong>: Apparently, AI code generation tools consistently produce a specific memory management bug: returning allocated buffer sizes instead of actual decompressed data sizes. This creates a security vulnerability by exposing uninitialized memory. It seems LLMs are just too good at replicating common coding patterns, even the buggy ones, proving that statistical learning still has its limits when it comes to actual correctness. <a href="https://app.daily.dev/posts/x9XQt7sEo" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated navigation: The invisible coupling strikes again</strong>: AI-generated navigation using boolean flags seemed like a clever shortcut, but it created invisible coupling that led to unpredictable back button behavior and maintenance nightmares months later. Who knew that relying on ‘elegant’ AI solutions without proper type systems would result in fragile, untestable code? It’s almost as if architectural foresight is still a human-required skill. <a href="https://app.daily.dev/posts/5NyDqJTPW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated SwiftUI: Making testability a distant memory</strong>: AI-generated SwiftUI code is reportedly embedding business logic directly into views, making it utterly untestable. The ‘solution’ is to refactor and separate business logic from UI, which sounds suspiciously like… good software engineering practices. It seems AI is great at generating code, but not so great at generating <em>maintainable</em> code. Maybe we should ask it to generate a better understanding of separation of concerns instead. <a href="https://app.daily.dev/posts/jCgZBlLZU" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
