Anyone know if there's a solid answer from OpenAI or Google DeepMind yet on whether the models they used in their ICPC programming competition wins had access to tools like the ability to run generated code through a compiler?
@simonw
-
Command+Shift+P Developer Chat Debug View Shortcut
By
–
I had NOT seen that! That's awesome – Command+Shift+P and then search for "Developer: Show Chat Debug view"
-
GitHub Copilot CLI Notes and Implementation Guide
By
–
Blogged some notes on GitHub Copilot CLI here
-
Learning Effectively with Top Tier Large Language Models
By
–
If you genuinely want to learn new things the amount you can learn from working thoughtfully with a top tier LLM is enormous
-
Model Limitations: Why AI Cannot Generate Ungrounded Content
By
–
That would be a model that couldn't answer the prompt "write me a story where a walrus teaches me the basic ideas of particle physics" Because a Walrus never said those things!
-
GPT-5 Output Quality Comparison and Visual Results
By
–
It's pretty good – the one I got out of regular GPT-5 is slightly cleaner but a bit more funny-looking
-
Qwen3-VL 235B Vision-LLM Monster Released Today
By
–
Plus notes on the 5 (!) new things Qwen released today, the most exciting of which is the first in their Qwen3-VL vision-LLM series, a 235B 471 GB Apache 2 licensed monster!
-
OpenAI GPT-5-Codex API Now Available with Tool Support
By
–
My notes on OpenAI's gpt-5-codex model, now available via their API – I upgraded my llm-openai-plugin to handle it and had GPT-5-Codex itself implement tool support for that plugin
-
Calendar Invite Prompt Injection Data Exfiltration Attack Vector
By
–
Could this be abused as a prompt injection data exfiltration mechanism? Someone might trick the assistant into sending them a calendar invite where the description of the event includes private data pulled from other assistant-available sources