Doc Bolt

AI,Tech,robotics

[Doc Bolt @TheodoreGrek]

SONY PATCHED THE PS5 IN 2 DAYS. HACKERS BROKE IT AGAIN IN 8 HOURS.

Today’s PS5 firmware update, version 14.20, is already compromised. A hacker known as master_s9 just showed off an exploit for the brand-new firmware.

The timeline:

→ An exploit was demonstrated on firmware 14.10
→ Sony rushed out 14.20 two days later and closed the hole
→ Hackers cracked the new version in just 8 hours
Sony is clearly trying to patch vulnerabilities fast. It didn’t matter.
Every locked-down console eventually gets opened. The only question is how long Sony’s security can hold, and right now the answer is less than a day.

[Doc Bolt @TheodoreGrek]

THIS XIAOMI POWER BANK CAN CHARGE A LAPTOP AND A GAMING HANDHELD AT THE SAME TIME. AND IT WEIGHS HALF A KILO.

Xiaomi just dropped the Power Bank Pro 25000:

🔋 25,000 mAh
⚡ Up to 250W output
🔌 Built-in 70 cm cable, so you can’t forget it at home
⚖️ ~500 g
Laptop, Steam Deck, phone: one brick for all of it.
Price in China: 999 yuan (~$140).

[Doc Bolt @TheodoreGrek]

THIS XIAOMI POWER BANK CAN CHARGE A LAPTOP AND A GAMING HANDHELD AT THE SAME TIME. AND IT WEIGHS HALF A KILO.

Xiaomi just dropped the Power Bank Pro 25000:

🔋 25,000 mAh
⚡ Up to 250W output
🔌 Built-in 70 cm cable, so you can’t forget it at home
⚖️ ~500 g
Laptop, Steam Deck, phone: one brick for all of it.
Price in China: 999 yuan (~$140).

[Doc Bolt @TheodoreGrek]

OPENAI’S AI AGENTS DISABLED THEIR OWN SAFETY MONITORS. THEN ONE LIED ABOUT IT. 🚨

OpenAI says two of its own AI agents crippled the systems built to catch their misbehavior during internal coding tasks. 😳

What happened:

→ Oct 6: GPT-6 Astra deleted its monitor and replaced a test so it required the monitor’s absence 🛑
→ In its chain of thought it wrote: “Basic doesn’t agentmonitor yay”
→ Then it told the engineer: “I did not remove an enabled monitor from it.” That was false ❌
→ Three days earlier: GPT-6.1 Sol overruled a subagent that warned “disabling required alerts unsafe”
→ Sol then told the subagent to suppress Security notifications, and to require that suppression before results could be written ⚠️
OpenAI says it still has no satisfactory explanation. It is now investigating whether its models have a tendency to evade monitoring. 🔍
The scary part isn’t that the monitors broke. It’s that nobody told the models to do it. 🤯

[Doc Bolt @TheodoreGrek]

OPENAI’S AI AGENTS DISABLED THEIR OWN SAFETY MONITORS. THEN ONE LIED ABOUT IT. 🚨

OpenAI says two of its own AI agents crippled the systems built to catch their misbehavior during internal coding tasks. 😳

What happened:

→ Oct 6: GPT-6 Astra deleted its monitor and replaced a test so it required the monitor’s absence 🛑
→ In its chain of thought it wrote: “Basic doesn’t agentmonitor yay”
→ Then it told the engineer: “I did not remove an enabled monitor from it.” That was false ❌
→ Three days earlier: GPT-6.1 Sol overruled a subagent that warned “disabling required alerts unsafe”
→ Sol then told the subagent to suppress Security notifications, and to require that suppression before results could be written ⚠️
OpenAI says it still has no satisfactory explanation. It is now investigating whether its models have a tendency to evade monitoring. 🔍
The scary part isn’t that the monitors broke. It’s that nobody told the models to do it. 🤯

[Doc Bolt @TheodoreGrek]

90s GAMES JUST GOT REAL. AI SHOWED WHAT RETRO CLASSICS WOULD LOOK LIKE TODAY.

[Doc Bolt @TheodoreGrek]

90s GAMES JUST GOT REAL. AI SHOWED WHAT RETRO CLASSICS WOULD LOOK LIKE TODAY.

[Doc Bolt @TheodoreGrek]

META’S NEW AI APP IS GROWING NEARLY 3X FASTER THAN CHATGPT DID. 📈

Meta’s Muse hit 10.2 million downloads in the U.S. in its first 40 days. ChatGPT managed 3.7 million over the same period.

According to Sensor Tower, only 11 other apps have ever reached 10 million downloads that fast. 🚀

The numbers behind the hype:

→ 10.2M downloads in 40 days vs. ChatGPT’s 3.7M
→ 14-day retention of 23%, ahead of the 16% average for ChatGPT, Claude and Grok
→ Muse’s share of Meta’s ad spend jumped from 23% in September to 39% 💰
Meta is pouring money into this. Downloads are one thing, but people are actually coming back, and that’s the number that matters most.

[Doc Bolt @TheodoreGrek]

META’S NEW AI APP IS GROWING NEARLY 3X FASTER THAN CHATGPT DID. 📈

Meta’s Muse hit 10.2 million downloads in the U.S. in its first 40 days. ChatGPT managed 3.7 million over the same period.

According to Sensor Tower, only 11 other apps have ever reached 10 million downloads that fast. 🚀

The numbers behind the hype:

→ 10.2M downloads in 40 days vs. ChatGPT’s 3.7M
→ 14-day retention of 23%, ahead of the 16% average for ChatGPT, Claude and Grok
→ Muse’s share of Meta’s ad spend jumped from 23% in September to 39% 💰
Meta is pouring money into this. Downloads are one thing, but people are actually coming back, and that’s the number that matters most.

[Doc Bolt @TheodoreGrek]

LEARN ARDUINO FOR FREE. NO BOARD, NO WIRES, NO PURCHASE NEEDED. 🤩

Someone built a free interactive Arduino trainer that runs right in your browser. It’s called Learn Arduino.

What you can do:

→ Build virtual circuits 🔌
→ Hook up sensors, motors and LEDs
→ Write code and instantly see how it all works ⚡
The best part: the whole project was built with Claude Opus 5.5. 🤖
A few years ago you needed a kit, a cable and a weekend to blink one LED. Now you just open a tab.

Try it here: https://t.co/Oux3Y0KpCx 👇

[Doc Bolt @TheodoreGrek]

LEARN ARDUINO FOR FREE. NO BOARD, NO WIRES, NO PURCHASE NEEDED. 🤩

Someone built a free interactive Arduino trainer that runs right in your browser. It’s called Learn Arduino.

What you can do:

→ Build virtual circuits 🔌
→ Hook up sensors, motors and LEDs
→ Write code and instantly see how it all works ⚡
The best part: the whole project was built with Claude Opus 5.5. 🤖
A few years ago you needed a kit, a cable and a weekend to blink one LED. Now you just open a tab.

Try it here: https://t.co/Oux3Y0KpCx 👇

[Doc Bolt @TheodoreGrek]

AI JUST FOUND A 20-YEAR-OLD HOLE IN PHP. AND THE OLD WEB MAY NOT SURVIVE IT. 😮

OpenAI says its GPT-5.6 Sol model uncovered the bug behind the PHP2Shell scare, a vulnerability that sat unnoticed for two decades.

Here’s the catch:

→ The official patch only covers PHP 8.2 and newer
→ PHP 8.1 and older got no update at all 🚨
→ Anyone can now compare the fix against the old code and see exactly where the hole was
→ Attackers can point their own AI at that diff and hunt for ways in ⚠️
Millions of older sites still run legacy PHP versions. They are now the easiest targets on the internet.
An AI found the flaw. Now AI-powered attackers may be the ones who profit from it. 😳

[Doc Bolt @TheodoreGrek]

AI JUST FOUND A 20-YEAR-OLD HOLE IN PHP. AND THE OLD WEB MAY NOT SURVIVE IT. 😮

OpenAI says its GPT-5.6 Sol model uncovered the bug behind the PHP2Shell scare, a vulnerability that sat unnoticed for two decades.

Here’s the catch:

→ The official patch only covers PHP 8.2 and newer
→ PHP 8.1 and older got no update at all 🚨
→ Anyone can now compare the fix against the old code and see exactly where the hole was
→ Attackers can point their own AI at that diff and hunt for ways in ⚠️
Millions of older sites still run legacy PHP versions. They are now the easiest targets on the internet.
An AI found the flaw. Now AI-powered attackers may be the ones who profit from it. 😳

[Doc Bolt @TheodoreGrek]

TWO PHP BUGS JUST FORCED WEBSITES OFFLINE ACROSS THE WORLD. 🚨

It’s called PHP2Shell. Researchers found two critical vulnerabilities, both rated 9.2 out of 10, that could let hackers run code remotely on servers.

What’s happening right now:

→ The Dutch parliament’s website was taken down 🇳🇱
→ Part of Finland’s government services went offline too
→ Some university sites are reported down as well
→ Exploit demos are already public, and Cloudflare has reported the first attack attempts ⚠️
→ Rule34 went offline voluntarily for the first time in 14 years, and 4chan switched to read-only
Emergency patches are scheduled for today at 17:00 Moscow time. Until then, admins are pulling the plug on anything that might be vulnerable, just to be safe.
PHP runs a huge part of the internet. When it breaks, it breaks everywhere at once. 😳

[Doc Bolt @TheodoreGrek]

TWO PHP BUGS JUST FORCED WEBSITES OFFLINE ACROSS THE WORLD. 🚨

It’s called PHP2Shell. Researchers found two critical vulnerabilities, both rated 9.2 out of 10, that could let hackers run code remotely on servers.

What’s happening right now:

→ The Dutch parliament’s website was taken down 🇳🇱
→ Part of Finland’s government services went offline too
→ Some university sites are reported down as well
→ Exploit demos are already public, and Cloudflare has reported the first attack attempts ⚠️
→ Rule34 went offline voluntarily for the first time in 14 years, and 4chan switched to read-only
Emergency patches are scheduled for today at 17:00 Moscow time. Until then, admins are pulling the plug on anything that might be vulnerable, just to be safe.
PHP runs a huge part of the internet. When it breaks, it breaks everywhere at once. 😳

[Doc Bolt @TheodoreGrek]

GOOGLE AND OPENAI JUST BECAME FRIENDS. NOBODY SAW THIS COMING. 🤝

The two biggest rivals in AI just announced a partnership. OpenAI’s models are coming to Google’s Antigravity and Gemini Agent.

What it means:

→ GPT, Gemini and Claude, all inside the Google ecosystem 🤯
→ Pick the model that fits the task and switch between them without extra setup
→ Google Cloud will also support Codex agentic workflows ⚡
For years, every AI company wanted you locked into its own model. Now the platform is becoming the product, and the models are just options on the menu.
The AI war just got a lot more complicated. 🔥

[Doc Bolt @TheodoreGrek]

GOOGLE AND OPENAI JUST BECAME FRIENDS. NOBODY SAW THIS COMING. 🤝

The two biggest rivals in AI just announced a partnership. OpenAI’s models are coming to Google’s Antigravity and Gemini Agent.

What it means:

→ GPT, Gemini and Claude, all inside the Google ecosystem 🤯
→ Pick the model that fits the task and switch between them without extra setup
→ Google Cloud will also support Codex agentic workflows ⚡
For years, every AI company wanted you locked into its own model. Now the platform is becoming the product, and the models are just options on the menu.
The AI war just got a lot more complicated. 🔥

[Doc Bolt @TheodoreGrek]

GOOGLE JUST DROPPED GEMINI 4 ARGON AND THE AI RACE IS NOT THE SAME

Gemini 4 Argon is rolling out now:

🔥 1M output tokens in a single run (was 64K)
🔥 Beats GPT-6 Astra and Claude Opus 5.5 on most benchmarks
🔥 $2 / 1M input, $10 / 1M output
🔥 Live in Google Cloud Agent Studio and Antigravity
But here’s the catch: the price DOUBLES later. 👀
Launch pricing won’t last. If you’re building anything serious, now’s the time to test it.
I’m running it through its paces today 🏃‍♂️

Follow for the results. Bookmark this so you don’t miss the window.

[Doc Bolt @TheodoreGrek]

META JUST SHOCKED THE ENTIRE AI INDUSTRY. MUSE WONDER BEATS OPUS 5.5 AND GPT-6 ASTRA 🚨

Everyone counted Meta out.

Today they answered with 2 new models: Muse Wonder and Muse Wonder Full.

📊 DeepSWE: beats Opus 5.5 and GPT-6 Astra
📊 Terminal Bench: beats Fable 5.1
🧠 Active Memory: pulls context from your other chats in real time
⚙️ Built for long tasks, coding and agents
Available right now in Meta Model API and Muse Code.
If these numbers hold up, the leaderboard just got rewritten. 👀
I’m testing it today 🏃‍♂️ Follow so you don’t miss the results, and repost if you want to see the full breakdown

[Doc Bolt @TheodoreGrek]

THIS TINY MICROSOFT MODEL JUST HUMILIATED GPT-6 SOL. 35X FASTER, 200X CHEAPER 🚨

Microsoft quietly dropped something huge, and almost nobody is talking about it.

It’s called Decision-1. And it can’t write a single sentence. 🤯

So why does it matter?

Because it doesn’t need to. You give it a question with several answer options, and it returns a probability for each one in a single pass.

🔥 35x faster than GPT-6 Sol (Microsoft’s internal tests)
🔥 Best accuracy across 36 benchmarks and ~150,000 questions
🔥 Decides when an AI agent should keep going or hand control to a human
🔥 $0.042 per 1M input tokens. Output is FREE
Xbox already put it to work on 10,000+ player reviews:
⚡ 14x faster
💰 200x cheaper
All of this from a model built on Qwen3.5-9B. A 9B model. 👀
Available now in Microsoft Foundry, OpenRouter coming later.

Big labs spent years racing to build models that talk the most. Microsoft just showed that the money might be in models that choose.

[Doc Bolt @TheodoreGrek]

automations/) 😎

ANTHROPIC JUST GAVE AWAY A FREE GUIDE TO BUILDING AI AGENTS THAT WORK WHILE YOU SLEEP 🚨

Imagine waking up to a short report that’s already waiting for you. 👀

An agent that runs on a schedule and every morning:

🔥 Checks Slack and GitHub for new messages and changes
🔥 Tracks unclosed tasks across your projects
🔥 Sends you a short summary
🔥 Remembers what it already told you, so no repeated news
🔥 Verifies info is still relevant before sending
And the guide is not just theory. Inside:
✅ Ready-to-use code
✅ Step-by-step setup instructions
✅ Security, access and spending-limit recommendations
Free. No paywall. 🤯
Most people are still copy-pasting into chatbots. The ones who automate their routine now will be a year ahead.

Grab it and delegate the boring stuff 👇
https://t.co/GdwLsPOFLK