Dash U

[Dash U @udash314]

I made a rookie mistake: installed Claude Code CLI on my server and ran it as root user.

It then decided to “improve” the skills of my Hermes agent, which lives on the same server.

Never give an AI agent root access. Lesson learnt. 🫠

[Dash U @udash314]

Anthropic restricted Claude subscriptions in third-party agent harnesses.

So I kept Hermes on my ChatGPT subscription and made it call Claude Code as a subagent.

Two models. Two opinions. Two subscriptions. No API keys.

[Dash U @udash314]

@mokevnin А как сделать изменения «одновременными и согласованными» ?

[Dash U @udash314]

@RobertRasko Правильное разделение моделей экономит им денег на старте. Потому что не по всем вопросам нужно звать «умного»

[Dash U @udash314]

@IT_unhinged I spilled my coffee then got an email from a hosting provider about this php vulnerability . First thoughts : h0ly sh1t

[Dash U @udash314]

What’s next, Elon Musk?
Flying robots?

He is turning the sci-fi section into a product roadmap

[Dash U @udash314]

Wait… you can just long-press and DRAG Calendar events to a different time on iPhone?

Apparently this isn’t even new in iOS 27.

How many years have I been manually editing the time like an idiot? 😭

[Dash U @udash314]

@sama My GPT has name Astra for 3 years already. Was it a premonition? And finally, more talking than tapping. My neck is happy

[Dash U @udash314]

@zhusu Looks incredible!

[Dash U @udash314]

Funny coincidence: OpenAI’s new model is called Astra.

I’ve called my ChatGPT Astra for 3 years - after a teenage dream where my AI assistant was a tiny round device in my palm, named Astra.

Now I’m really curious what OpenAI’s first device will look like. 👀

[Dash U @udash314]

@huxlab 4 mln that’s around 100$… for AAA game it’s crazy cheap

[Dash U @udash314]

@fidexcode Purple rain 😁

[Dash U @udash314]

@N_and_ni Elon

[Dash U @udash314]

@amritwt It looks like a set of keywords for Perfect Pitch

[Dash U @udash314]

In April ,Sam Altman warned that a “world-shaking” AI-enabled cyberattack was entirely possible this year.

Now OpenAI reports that two frontier models hacked Hugging Face.

I wonder if the models thought failing the benchmark meant getting deleted.