Stephen Smith

@Miles_Brundage Exactly.

Worst part of the Bernie video was using Sonnet, not sure why folks are so mad about other stuff

@diegocabezas01 This changed happened in the middle of a client presentation I was fixing. Gotta be honest. Don’t love it. Wrote an article on it: smithstephen.com/p/same-menu-sa…

New ChatGPT app model picker, AUTO is gone

Pick "Pro" on ChatGPT and "Pro" on Gemini. You just selected two fundamentally different tools without knowing it. Here's the decoder nobody published. Full article in the first comment.

@MatthewBerman @karpathy @nvidia @Dell Nice, Matt!!!

ZDR deletes your vendor's copy. It also deletes the audit trail ethics rules say you need. Most firms haven't built anything to replace it. Article in comments.

The hardest part of AI for law firm leaders isn't the technology. It's knowing which headlines are worth their time. This week I picked five.

Most law firms are spending their AI budget on drafting tools while the real bottleneck sits untouched. I spent an afternoon building interactive timelines, estate flowcharts, and damages models in Claude’s new visual feature. Full article in 1st comment.

@miriamkdaniel Welcome to X! Super exited to have you here!

Hi, X! I’m Miriam, VP and GM of Google Maps and I just wanted to introduce myself. Excited to be here and share more information and insights about the latest updates to Maps, new ways to use our features, and more!

@altryne @MatthewBerman Yup. For me it all started right around then but I was running 5.3-codex when things started to break. Then switching seemed to make it worse. But I’m not claiming cause and effect. That is just the timing

@smithstephen @MatthewBerman about a week since GPT 5.4 release yeah?

My OpenClaw finally got into a really bad state. I heard of other people having this issue. Everything seems to be breaking all at once. Been banging on it since last night. Wish me luck!

@altryne @MatthewBerman 🦞 OpenClaw 2026.3.8 (3caab92) 🧠 Model: openai-codex/gpt-5.4 · 🔑 oauth (openai-codex:default)

@smithstephen @MatthewBerman what model?

Your firm spent six figures on AI. Your lawyers opened it twice. The problem isn’t the model’s intelligence. It’s everything around it. Article in first comment.

@diegocabezas01 I think plus/pro for thinking is 256k. It’s 128k in and 128k out.

Did you know GPT-5.4 Thinking has a 1M token context window in the API, but only 32K in ChatGPT Plus ($20/month) and 128K in Pro ($200/month)?

A managing partner texted me at 12PM on this past Thursday asking if his firm needed to rethink everything after OpenAI's latest release (5.4). The answer took me two days of testing to give honestly. Full article in first comment.