
Wow. Same assignment to Fable 5.2 and Astra 6.2. Complicated and highly detailed training deck for a new client. Astra: 7/10 (after two prompts). Fable: 9/10 (after one prompt). Honestly not even close. Wow.

The AI training partner for firms that aren't Am Law 100. 3,500+ attorneys trained, 100+ firms. CEO Intelligence by Intent. CLE keynotes.

Wow. Same assignment to Fable 5.2 and Astra 6.2. Complicated and highly detailed training deck for a new client. Astra: 7/10 (after two prompts). Fable: 9/10 (after one prompt). Honestly not even close. Wow.

I keep thinking about this. Kind of feels like the ChatGPT 5 moment. High expectations and a miss. 6.2 Astra and 6.2 SOL are good. But not good enough.

I'm excited to see that @GeminiApp is going to finally allow consumers to turn off training AND keep their history.


The post is evidence. The prompt behind it might be worse. A Miami judge just showed how courts will decide which of your clients' AI chats the other side gets to read.


So - Magnus managed to give me two fractured ribs this weekend. I decided to turn that into a video test (@claudeai Fable 5.2, @ChatGPT GPT-6 Astra, @GeminiApp Gemini 4 Carbon). I think you'll enjoy this one (once you stop laughing at what he did to me).

Favorite new computer use case. Doing a huge build in @antigravity CLI. It's *constantly* asking for permission. Way too much. That really needs to get fixed. So I have @ChatGPT computer use running with 6.1 Terra to monitor it and actively push the damn button for me.

The new @GeminiApp Carbon model is just unbelievable at understanding video and audio. Better than any other model I've used (including Fable 5.2). I was using it today with a 900mb training video to find every hiccup, glitch, pause that shouldn't be there. It build incredibly detailed timestamps with perfect details and the exact things to fix. I've never seen anything like it.


I was recently asked "Aren't you afraid of the dangers of AI?" My instant response was "no." But I've been thinking about that question ever since.


My wife and I were talking about this at dinner again last night: what do kids to today in an era where AI is starting to take away entry level jobs. I talk about this a ton when I do keynotes and client presentations. I decided to also see what the latest models think about this.


Legal AI rollouts don't die on the model. They die on a Tuesday, when someone with a noon filing can't remember the prompt. Claude's new legal gallery is built for that morning. My take:
