Silv's AI weekly

X is the best way I’ve found to keep up with AI. I like tweets throughout the week, filtering for things I think are actually worth knowing. I use Claude Code to pull those likes automatically and help me turn them into this post (here’s how the pipeline works). This week: 148 tweets liked, filtered down to what’s below. Check out the previous roundup (May 2) if you missed it. AI for Everyone GPT-5.5 Instant Becomes the ChatGPT Default for Everyone (5 mentions) OpenAI started rolling GPT-5.5 Instant on May 5 as the default model for every ChatGPT user (also available in the API as gpt-5.5-chat-latest). The pitch is meaningful improvements in factuality, especially in medicine, law, and finance, plus a noticeable shift toward shorter, less padded answers. The memory update is the part worth digging into: ChatGPT can now use your saved memories, past chats, files, and a connected Gmail to personalize responses, and a new “memory sources” panel shows exactly what it pulled from so you can edit or delete entries. A full-duplex voice mode is coming, where the model can listen and speak at the same time. The companion OpenRouter cost analysis is the catch: short-prompt workloads are about 92% more expensive on GPT-5.5 than GPT-5.4. (source: @OpenAI, @OpenAI, @OpenAI) ...

X is the best way I’ve found to keep up with AI. I like tweets throughout the week, filtering for things I think are actually worth knowing. I use Claude Code to pull those likes automatically and help me turn them into this post (here’s how the pipeline works). This week: 148 tweets liked, filtered down to what’s below. Check out the previous roundup (Apr 24) if you missed it. AI for Everyone ElevenMusic Launches With Creator Payouts (7 mentions) ElevenLabs shipped ElevenMusic, a platform where you discover independent artists, remix tracks, and create songs from a prompt. The economic model has evidence behind it: ElevenLabs has paid out $11M to voice creators through its voice library and is bringing the same revenue share to music. Over 4,000 artists are on the platform at launch, and the Eleven Album Vol. 2 dropped alongside it. The same week added a Voice Changer Skill for ElevenLabs Agents that transforms voice in real time while preserving emotion and timing. (source: @ElevenLabs, @ElevenLabsDevs) ...

X is the best way I’ve found to keep up with AI. I like tweets throughout the week, filtering for things I think are actually worth knowing. I use Claude Code to pull those likes automatically and help me turn them into this post (here’s how the pipeline works). This week: 150 tweets liked, filtered down to what’s below. Check out the previous roundup (Apr 16) if you missed it. AI for Everyone GPT-5.5 Ships in ChatGPT (8+ mentions) OpenAI launched GPT-5.5 on April 23. Same per-token latency as GPT-5.4, meaningfully smarter on knowledge work, research, and writing. OpenAI is calling it their smartest and most intuitive model yet, and the early reactions back the claim up. One team at Chai tested it on huge spreadsheets (100K to 1M+ cells) and called it the Pareto frontier for that work: best accuracy, fastest, most efficient. OpenAI also slipped a line into their announcement that I think will age well: “the last few years have been surprisingly slow.” They’re telling you the release pace is about to get a lot faster. (source: @OpenAI, @kimmonismus, @nicochristie) ...

X is the best way I’ve found to keep up with AI. I like tweets throughout the week, filtering for things I think are actually worth knowing. I use Claude Code to pull those likes automatically and help me turn them into this post (here’s how the pipeline works). This week: 150 tweets liked, filtered down to what’s below. Check out the previous roundup (Apr 5) if you missed it. AI for Everyone Claude Opus 4.7 Ships (12+ mentions) Anthropic launched Claude Opus 4.7 on April 16. Pricing held steady at $5 per million input tokens and $25 per million output. What changed is how the model spends those tokens: it thinks longer before answering and double-checks itself more aggressively. Vision is the part I didn’t see coming. Opus 4.7 now accepts images up to 2,576 pixels on the long edge, triple what prior Claude models took, which actually matters for diagrams, slides, and UI screenshots. If you’ve been running 4.6 for long tasks, swap the model ID and give it something harder. (source: @claudeai, @claudeai, @bcherny) ...

X is the best way I’ve found to keep up with AI. I like tweets throughout the week, filtering for things I think are actually worth knowing. I use Claude Code to pull those likes automatically and help me turn them into this post (here’s how the pipeline works). This week: 149 tweets liked, filtered down to what’s below. Check out the previous roundup (Mar 24) if you missed it. AI for Everyone Gemma 4 Runs Locally on a Mac Mini (10+ mentions) Google released Gemma 4 under Apache 2.0. The 26B version uses mixture-of-experts, only activating 4 billion parameters at a time, so it runs on a 24GB Mac Mini. It scores 79.2% on GPQA Diamond, which puts it in the same neighborhood as Claude Sonnet 4.6’s 74.1% standard score (though Sonnet hits 89.9% with max-effort reasoning). Four sizes from edge mobile to 31B dense, 256K context, multimodal input, 140+ languages. ollama run gemma4:26b and you’re running a competitive model locally for free. I’ve been waiting for an open model this good. (source: @GoogleDeepMind, @PawelHuryn, @ollama) ...

X is the best way I’ve found to keep up with AI. I like tweets throughout the week, filtering for things I think are actually worth knowing. I use Claude Code to pull those likes automatically and help me turn them into this post (here’s how the pipeline works). This week: 104 tweets liked, filtered down to what’s below. Check out the previous roundup (Mar 16) if you missed it. AI for Everyone Claude Computer Use on Mac (5 mentions) Claude can now control your Mac: mouse, keyboard, screen, all of it. Kick off a task from your phone via Dispatch, walk away, come back to finished work. It’s slower than a human on purpose. Best for repetitive stuff that doesn’t have an API: copying data between apps, filling out web forms, clicking through legacy software. It tries connected apps (Slack, Calendar) first and only grabs screen control when needed. macOS now, Windows in a few weeks. (source: @claudeai, @felixrieseberg) ...

X is the best way I’ve found to keep up with AI. I like tweets throughout the week, filtering for things I think are actually worth knowing. I use Claude Code to pull those likes automatically and help me turn them into this post (here’s how the pipeline works). This week: ~110 tweets liked, filtered down to what’s below. Check out the previous roundup (Mar 8) if you missed it. AI for Everyone Claude Opus 4.6 Gets 1M Context Window at No Extra Cost Anthropic dropped the long-context surcharge. 1M tokens is now standard for Opus 4.6 and Sonnet 4.6, same price as any other request. Feed in an entire codebase, a stack of PDFs, hundreds of images. For Claude Code users on Max, Team, or Enterprise, it’s already the default. Benchmark that matters: 78.3% on MRCR v2 at 1M tokens, best-in-class. (source: @claudeai, @bcherny, @ClaudeCodeLog) ...

X is the best way I’ve found to keep up with AI. I like tweets throughout the week, filtering for things I think are actually worth knowing. I use Claude Code to pull those likes automatically and help me turn them into this post. This week: 298 tweets liked, filtered down to what’s below. Check out the previous roundup (Feb 26) if you missed it. AI for Everyone GPT-5.4 Is Here, and It Can Control Your Computer OpenAI launched GPT-5.4 this week. Codex-level coding merged with GPT-5.2-level reasoning, native computer use, and a 1M-token context window, all in one model. You can steer it mid-response. It’s live in ChatGPT (as GPT-5.4 Thinking and GPT-5.4 Pro), in the API, and rolling into Microsoft Copilot Studio. If you want to try it for coding, download the Codex app. You used to need a dedicated Codex model, but 5.4 handles it all. You still need to select a reasoning level. ...

Here’s what caught my attention from Feb 16-26, 2026: Check out the previous roundup (Feb 16) if you missed it. AI for Everyone Google Launches Nano Banana 2: Pro-Quality Image Generation at Flash Speed Google just dropped Nano Banana 2, their new image generation model built on Gemini Flash. The headline numbers: pro-level quality running at Flash speed, which in practice means you get high-fidelity images without waiting around. It handles text rendering (historically terrible in AI image gen), 4K upscaling, aspect ratio control, and subject consistency across multiple generations. Demis Hassabis says it taps into Gemini’s world understanding and real-time search to produce more accurate results, which tracks with what we’re seeing in the outputs. The rollout is wide: Gemini App, AI Studio, the Gemini API (listed as “Gemini 3.1 Flash Image”), Google Search, Flow, Google Ads, and Vertex AI. OpenRouter already has it. Logan Kilpatrick also announced new lower-cost resolutions and an Image Search tool. If you’ve been using DALL-E or Midjourney by default, this is worth trying today. (source: @GoogleDeepMind, @demishassabis, @OfficialLoganK) ...

Personal News I got accepted into the Claude Code hackathon last week. 13,000 people applied, 500 got in. The hackathon just wrapped up today and I built Robo.app. Check it out, and watch the 3 min pitch video! Here’s what caught my attention from Feb 10-16, 2026: 🚀 OpenAI Hires OpenClaw Founder for “Personal Agents” Peter Steinberger joining OpenAI to build next-gen personal agents - Sam Altman personally welcomed him. OpenClaw hit 100K GitHub stars in 2 days (fastest any repo has ever done that), now at 197K+. Both Meta and OpenAI made offers. The fact they’re building a dedicated personal agents team, not folding him into ChatGPT, signals a new product line. OpenClaw will transition to an independent open-source foundation that OpenAI will continue to support. ( @sama) 🏋️ Three Major Model Launches in One Week Qwen 3.5-397B released - 397B total params, 17B active via hybrid linear attention + MoE. 87.8 MMLU-Pro, 83.6 LiveCodeBench v6, 76.4 SWE-bench Verified. Decoding 8.6x-19x faster than Qwen3-Max. Apache 2.0, 201 languages. Already on OpenRouter. ( @Ali_TongyiLab) MiniMax M2.5: Sonnet-level coding at 1/20th Opus cost - 230B MoE, 10B active. 80.2% SWE-Bench Verified, $0.30/M input tokens. OpenHands confirmed it’s the first open model to beat Sonnet on their benchmark. Stock jumped 15.7%. If you’re running coding agents, test this as your workhorse model. ( VentureBeat) ByteDance Seed 2.0 / Doubao 2.0 - Claims GPT-5.2 and Gemini 3 Pro level at ~1/10th cost. Doubao app has 155M weekly active users. China’s frontier labs aren’t 6 months behind anymore. ( @QuanquanGu) 🌐 WebMCP: Every Website Becomes an API for Agents Google and Microsoft co-authored a W3C spec called WebMCP - Websites can now declare capabilities as structured tools via navigator.modelContext. Two modes: declarative (HTML forms self-describe) and imperative (JS exposes custom tools). 89% token efficiency improvement over screenshot-based agents. Available in Chrome 146 Canary behind a flag. ( @aakashgupta) Pietro Schirano built a DoorDash agent demo - Adds items to cart and checks out without ever parsing UI. This is what “agents that talk to apps” looks like. ( @skirano) Links: Chrome blog | W3C spec draft ...