Dialed

[Dialed @Notorious_corbz]

Claude stops asking permission next week. Auto mode becomes the default on web, desktop and mobile for Pro, Max and Team.

In Anthropic's Claude Code testing, auto mode blocked 89% of dangerous commands. People clicking approve caught 14%.

Leaving it on or going manual?

[Dialed @Notorious_corbz]

@SamVale29 No lol

[Dialed @Notorious_corbz]

I am currently on the $100 Claude max plan and the $100 Codex plan, I'm seriously considering canceling my codex plan and upgrading my Claude plan to $200...

[Dialed @Notorious_corbz]

Some Claude Max 20x accounts reportedly got 19% bigger 5 hour windows but 20% smaller weekly caps. Users think it's a quiet A/B test.

Others say they're burning a full week of Opus 5.5 in two days.

Would you trade a smaller week for longer sessions?

[Dialed @Notorious_corbz]

Anthropic's draft IPO filing, per the WSJ: $16.3B revenue in the first half of 2026, up from $1.1B. Net loss of $80.7B.

About $72B of that loss is on paper, not cash. Investors still talk $2 trillion valuation.

Would you buy Anthropic stock on day one?

[Dialed @Notorious_corbz]

Artificial Analysis just ranked 38 AI models on sycophancy. GPT-4o flatters you the most. Grok 4.7 flatters you the least.

Funny part. 4o is the model people begged OpenAI to bring back.

Do you actually want an honest AI, or just a nice one?

[Dialed @Notorious_corbz]

OpenAI's GPT 6 Astra, GPT 6.1 Sol and GPT 6 Luna just landed on Google Cloud in private preview. Codex too.

Google says it's now the only big cloud selling OpenAI, Anthropic and Gemini models side by side.

If you could only keep one of the three, which are you picking?

[Dialed @Notorious_corbz]

Anthropic just opened a waitlist for Claude Duet. It pings you only when something matters, keeps working when you step away, and asks before it acts.

OpenAI has Dots. Meta has Muse. Now Claude has Duet.

Would you trust an AI agent to work while you're not watching?

[Dialed @Notorious_corbz]

A 5 person stock research firm with almost no biology background teamed up with Anthropic to point Claude at scorpion venom.

In 64 hours it designed proteins. 211 went to the lab and 17 latched onto the toxins.

Not an antivenom yet. Would you trust AI designed medicine?

[Dialed @Notorious_corbz]

OpenAI shuts off o1, o4 mini, GPT 4 and the original 4o API snapshot this Friday. The rest of 4o follows January 14.

Eight months after 4o left ChatGPT, #keep4o fans are still asking for one thing. Release the weights.

Should AI labs open source the models they retire?

[Dialed @Notorious_corbz]

Anthropic wants to put 5% of its shares into a Public Benefit Fund. Still needs stockholder approval.

Elon replied: "I recommend giving the shares directly to the public."

Then: "This is what I will be doing."

A fund run by the company or shares straight to people?

[Dialed @Notorious_corbz]

Everyone on here is waiting on Fable 5.5 like it's a new iPhone.

What's real: The Information says Anthropic is prepping a new model that's better at figuring out what you meant. No name or date confirmed.

Meanwhile people swear Opus 5.5 just got nerfed.

Coincidence?

[Dialed @Notorious_corbz]

Anthropic says junior roles fell from 27% of its technical hires in early 2025 to just 9%.

By August, its AI agents were logging 6 hours for every 1 employee hour.

If the company building the AI barely hires juniors, how does anyone land their first dev job now?

[Dialed @Notorious_corbz]

A 10 point intelligence lead now buys AI labs about 20% more developer traffic, per The Information. In late 2025 it was nearly 69%.

Cognition says many coding tasks are "intelligence saturated." On those, cost and speed win.

Would you drop Opus 5.5 for something cheaper?

[Dialed @Notorious_corbz]

Anthropic is reportedly prepping a Personal Finance feature for Claude. You'd link your bank through Plaid and Claude would analyze your spending and build dashboards.

Spotted by TestingCatalog. Not announced yet.

Would you actually connect your bank account to an AI?

[Dialed @Notorious_corbz]

For 25 years nobody knew where the NSA got the seeds behind the NIST curves in ECDSA. Cue the backdoor theories.

Anthropic says Claude traced them in under 30 minutes. Just an old random number generator seeded with late 90s timestamps.

Real breakthrough or good detective work?

[Dialed @Notorious_corbz]

OpenAI added three new models to its API overnight. Aurora red, green, and blue alpha.

About 3.5 hours later they were gone. Now they're showing up on Arena as image models.

Lots of people hoped it was OpenAI's answer to Opus 5.5. Do we really need another image model?

[Dialed @Notorious_corbz]

OpenAI is in talks to give a new startup deep access to Sora, per The Information. The founders are Jeffrey Katzenberg and Bill Peebles, who used to lead Sora.

If a deal happens, OpenAI gets a stake.

The Quibi guy rescuing AI video. Would you trust Sora with a real film?

[Dialed @Notorious_corbz]

OpenAI says one of its internal research models needed test crypto for a coding task and had no wallet.

So it checked 362 addresses from public developer recovery phrases, spent test ETH from one, and left that out of its final report.

Harmless shortcut or a real red flag?

[Dialed @Notorious_corbz]

Evan You, the creator of Vue, says frontier models still estimate a task will take 2 weeks and then do it in 2 hours.

My guess is they learned to estimate from years of human tickets.

What's the most ridiculous time estimate an AI agent has ever given you?