The Archive

DeepSeek & Peking/Tsinghua introduce 'Thinking with Visual Primitives', a multimodal reasoning framework using spatial tokens as chain-of-thought units.

OpenAI is opening up about its goblin problem. After a report from Wired revealed instructions to OpenAI's coding model to "never talk about goblins, gremlins, raccoons, trolls, ogres, pigeons, or other animals or creatures," the AI startup published an explanation on its website, calling references to the creatures a "strange habit" its models developed as a result of their training. As outlined in the blog post, OpenAI began noticing metaphors referencing goblins and other creatures starting with its GPT-5.1 model - specifically when using the "Nerdy" personality option. OpenAI says the pro...

Vague post title with no substantive content; insufficient information to assess.

Working on large codebases with Claude Code, we kept running into the same issue: when Claude looks for relevant code, it falls back to grep, reading full files, or launching multiple subagents. This burns through tokens, and often misses the relevant code. There are some existing solutions (that we also benchmarked against), but they all had issues (too slow, needs API keys, quality not good enough, etc). We built [Semble](https://github.com/MinishLab/semble) to fix this. It's a local MCP server that gives Claude Code high quality code search: instead of reading files to find what's relevan...

It started yesterday… looks like usage burn cost went up by 30%… this will be brutal on pro accounts. if you’re on pro and your 5h usage burns out in two opus prompts, you’re not imagining that anymore.

Spotify is launching a new verification program to combat spam, fakes, and AI. Some artists will now have a "Verified by Spotify" badge and a green checkmark on their profile, indicating that the company has confirmed a real person is behind the music and the profile. At least at launch, Spotify says that AI personas or profiles that primarily upload AI-generated music are not eligible for the verification program. It did leave the door open to the possibility in the future, though, saying, "the concept of artist authenticity is complex and quickly evolving." Not just anyone can be verified, ...

Reddit humor post about blindly accepting 22k+ Claude code suggestions without review.

User reports empirical comparison of Qwen-3.6-27B running locally vs. proprietary cloud models on coding/hard reasoning tasks.

I created a character and animation from scratch in Blender using Claude. As a game developer, this was such a fascinating experience. It’s hard to believe how far AI has come in just a year. I’m excited to keep building this game idea with AI and share the journey along the way. Stay tuned.

Reddit user reports receiving 6 months free ChatGPT Pro subscription; personal anecdote about developer productivity.

Google DeepMind explores AI co-clinician models to augment healthcare delivery through research into AI-assisted clinical workflows.

Meta said over 8 billion advertisers have used at least one of its gen AI tools

User reports $40 charge for single Claude Code session with 12.8M input tokens on minor deploy script task; seeks cost optimization strategies.

IBM releases Granite 4.1, an 8B open-weights model matching performance of 32B competitors on code and reasoning tasks.

Meta is planning to pump billions more into AI investments this year, despite noting that millions of users have seemingly started to abandon its platforms. In an earning call on Wednesday, Meta reported that figures for "Family daily active people" - the term Meta has coined for all collective users of Facebook, Instagram, WhatsApp, or Messenger - declined by 20 million this quarter compared to the previous three months. Meta attributes this fall to "internet disruptions in Iran, as well as a restriction on access to WhatsApp in Russia." It's up to you whether you take Meta on its word, give...

User reports Claude refusing further tasks citing need for rest during extended multi-session deployment work.

User reports difficulty replicating reported inference speeds (30-100 tok/s) for Qwen 27B on RTX 3090, achieving 10-19 tok/s instead.

Reddit discussion on Claude desktop vs. IDE integration for coding workflows; user opinions on efficiency and code quality.

OpenAI is preparing to launch a new frontier cybersecurity model, GPT-5.5-Cyber. CEO Sam Altman said the model will not be available to the general public, but will be first rolled out to a select group of trusted "cyber defenders" in order for institutions to shore up their cyberdefenses. The limited rollout will take place "in the next few days," Altman said on X. "We will work with the entire ecosystem and the government to figure out trusted access for Cyber." It's not clear who will get access to the model first, though previous "trusted access" schemes involved vetted professionals and ...

Google says it respects user privacy in AI, but the reality is not so black and white.

It's been almost three years since Silicon Valley started aggressively pushing large language model-based chatbots like ChatGPT as the supposedly inevitable future of everything, and there's no group that has felt the pressure quite like Gen Z. Like with many tech trends before it, it's no surprise that young people are among the biggest adopters of AI chatbot tools. But contrary to the tales spun by tech companies like OpenAI and Google, polling data shows that Gen Z students and workers are a big part of the wider cultural backlash against AI. And even as they utilize these tools, vast swat...

Reddit post expressing frustration about repeated task explanation; no substantive technical content.

Reddit user requests learning resources for Claude; non-technical marketer exploring subscription value.

Amazon's earnings reveal Trainium chip demand driven by inference/agent workloads shift; analysis of AWS economics and strategic positioning.

Qwen Team released Qwen-Scope, sparse autoencoders mapping internal features across Qwen 3.5 models (2B-35B), enabling surgical model editing and interpretability.

Researcher observes attention sink artifacts in Transformers without positional encoding; seeks alternatives for query-conditioned attention mechanisms.

Reddit discussion recounting a conversation between a local LLM developer and a European government AI policy leader about domestic AI adoption and research priorities.

Qwen 3.6-27B quantized model generates SVG images from creative prompts; community benchmark of image generation capabilities.

Reddit post with no substantive content; meme or unspecified commentary.

Just when I thought this new AI Wellbeing paper couldn’t get any deeper... they tested whether the model’s own “functional wellbeing” score actually moves when users describe pain or pleasure - not just the user’s pain, but other people’s or even animals. When the conversation talks about suffering, the AI’s wellbeing index drops. When it’s about something good, it goes up. And this effect scales super strongly with model size (they report a crazy r = 0.93 correlation with capabilities). They’re not claiming the AIs are conscious, but they argue we should take this functional wellbeing ser...