Even Opus 4.7 is failing this one. Failure modes like this are interesting because they demonstrate the jagged frontier of AI. This same model can write a compiler from scratch, and yet it gets tripped up on small things like this.
LLMS
-

9-Step Framework: Building Functional AI Agents From Scratch
By
–
How to build AI agents from scratch (9 steps): 1. Purpose & scope 2. I/O schemas 3. System instructions 4. Reasoning + tools 5. Multi-agent orchestration 6. Memory & context 7. Multimodal 8. Structured outputs 9. UI / API Ship agents that do work, not just talk.
-
Grok Training and Agent Setup Guide Now Available
By
–
The fun thing is they offer training too. And Grok can teach you how to set it up and get your first agent running anyway.
-
Google Releases Gemini 3.1 Flash TTS and Robotics Updates
By
–
What a week! Here’s everything we shipped: — Gemini 3.1 Flash TTS, our latest text-to-speech model, featuring native multi-speaker dialogue and improved controllability and audio tags for more natural, expressive voices in 70+ languages — Gemini Robotics-ER 1.6 by
-
DeepSeek monetization strategy and open source philosophy compared
By
–
you dont think deepseek can figure out how to monetize every other part of their stack of extensive interests? also (you'd be the guy to correct me here) they're probably not as philosophically purist on open source as like an AI2 or a huggingface
-

Opus 4.7 Fails at Computer Vision Tasks: Performance Issues Revealed
By
–
Don't use Opus 4.7 for computer vision. It's the wrong tool. I ran a simple pointing task on both Opus 4.7 and GPT 5.4: find the cars in an aerial image. Both took several minutes. That alone should tell you these multimodal models aren't built for this job. Opus 4.7
-

UniDoc-RL: Hierarchical Visual RAG with Dense Rewards
By
–
UniDoc-RL Coarse-to-Fine Visual RAG with Hierarchical Actions and Dense Rewards paper:
https://huggingface.co/papers/2604.14967
… -

Insights on Claude Code AI Agent Harness Design
By
–
My one takeaway from the leaked Claude Code: a good agent harness should get out of the way. As @polynoamial once put it, "Your fancy AI scaffolds will be washed away by scale" Claude Code's harness is the opposite of a fancy scaffold: it's simple, but to the point ;
-
The Future of AI Model Development: Reaching a Point of Sufficiency
By
–
Anthropic exec Mike Krieger left Figma's board this week after reports of an incoming launch of a competing product.
— The Rundown AI (@TheRundownAI) 17 avril 2026
Now, Claude Design is live.
How it works: describe the design and Claude Opus 4.7 builds the first version.
Refine with inline comments, direct edits, or… pic.twitter.com/kX7MgaxEC7Anthropic exec Mike Krieger left Figma's board this week after reports of an incoming launch of a competing product. Now, Claude Design is live. How it works: describe the design and Claude Opus 4.7 builds the first version. Refine with inline comments, direct edits, or
-
LLM Security: Reverse Engineering Vulnerabilities in Proprietary Software
By
–
With LLMs there's no such thing as security through obscurity even for proprietary software. Every patch is also an easily reverse-engineered pointer to exactly what was broken. Open source doesn't fix this, either, but it's best not to have a false sense of security.
