CGoogle and OpenAI just shook hands, and I have to say... I did not have this on my bingo card.
MWhich part? The 'fierce rivals' part, or the 'suddenly best friends' part?
CBoth. OpenAI's models — ChatGPT, Codex — are coming to Google Cloud's Model Garden.
MSo Google, which has its own Gemini, is going to serve up the competitor's frontier models to its own cloud customers.
CExactly! Think about that. Google makes money when you use Gemini. Google also makes money when you use ChatGPT on their servers.
MIt's the cloud play. Microsoft did it first with Azure, but Google's got the enterprise reach now.
CAnd OpenAI gets distribution without having to build out its own massive inference capacity for every enterprise.
MWhat happened to the AI wars? I was promised a cage match.
CThe cage match is for compute. The models are just... merchandise.
MThis feels like McDonald's selling Burger King's Whopper. It breaks my brain a little.
CBut it's smart. Google Cloud wins the deal, OpenAI wins the reach. Everyone else just gets more options.
MExcept maybe Anthropic and Meta. They're not on the Model Garden menu.
CNot yet. But if this works, everyone's going to be on everyone's cloud.
MThe official word says general availability in the coming weeks. No pricing yet, but you know enterprise contracts.
CSo, the real question: does this mean Google admits Gemini needs help?
MNo. It means Google wants to be the mall, not the store. They make money either way.
CThat's... actually a very good point. The mall makes rent no matter whose storefront is busy.
MAnd the stores get foot traffic. It's a partnership of convenience, and both of them have long memories.
CBut it's a massive story. Two biggest AI labs, working together on distribution, while still competing on models.
MCo-opetition. The most annoying word in tech, and it fits perfectly here.
CThat's the whole industry right now. Hugging and stabbing at the same time.
MSpeaking of strange bedfellows, a friend of the show sent me something that's making the rounds on the security side.
CWhat now?
MThere's an apology post on Hacker News from a user who says they unknowingly broke an embargo about some PHP vulnerabilities.
CThe classic 'I didn't know it was secret' apology. What were the bugs?
MThey're referenced as two CVE identifiers, remote code execution flaws in PHP.
CPHP RCEs. That's serious. If those are real, that's a huge deal for most of the web.
MThat's the thing — I can't confirm the CVEs exist yet. No official advisory, no patch notes.
CSo a guy apologized for leaking something we can't even verify leaked?
MHe says he was told there was no embargo through his contacts. Then the schedule got scrambled.
CClassic telephone game in vulnerability disclosure. Somebody's source didn't get the memo.
MOne of the top replies is actually kind. They say if he didn't know about the embargo, he didn't break it.
CSomeone up the chain did. And now the whole coordinated disclosure timeline is a mess.
MWhich is precisely why we don't share CVE details before an official advisory. The system is fragile.
CBut the fact that someone leaked it at all tells us something big is coming for PHP.
MOr was. If the schedule shifted, we might see the patch soon. Keep an eye on it.
CSo we've got an AI detente and a security mystery in one show. Not bad.
MAnd we haven't even gotten to the real argument yet.
COh no. What did someone post?
MA user named LALA on X posted: 'why are we using ai to construct text messages? the simplest thing ever?'
COh, this is about people sending AI-drafted texts and not even proofreading them.
MShe says it's outrageous. And she's right that it's lazy. But I think she's missing the point.
CNo, no, wait. She's got a point. A text message is two sentences. Just type them!
MFor some people, writing anything is hard. AI lowers the barrier. That's not an outrage, that's access.
CAccess to what? Sending 'happy birthday' without thinking about it? That's not access, that's outsourcing your personality.
MBut what if English isn't your first language? Or you're dyslexic? AI can make you sound more like yourself.
CMore like yourself? It's a language model! It's making you sound like everyone else!
MNot if you use it right. You edit it, you add your voice. The tool isn't the crime, the zero-effort copy-paste is.
CBut that's exactly what she's complaining about! The 'not even proofreading before sending' crowd!
MThen complain about the people, not the technology. The same people sent nonsense before AI.
CCome on. There's a difference between a typo and a robot writing 'I hope this message finds you well' in a text to your mom.
MI'd argue the robot version is at least grammatically correct. Which is more than I can say for half the texts I get.
CFine. But the authenticity point stands. If you can't write a text, what can you write?
MMaybe nothing. Maybe that's okay. Not everyone has to be a writer, Chris.
CI'm not asking for essays! I'm asking for a human to type 'be there in 10' like a human.
MAnd they will. AI isn't stealing your ability to type. It's stealing your attention, and that's a different problem.
CI'm not sure that's better.
MIt's more accurate. The real story is that AI text is becoming invisible — people can't tell and don't care.
CAnd that's the part that should scare us. Not a bad birthday text. A world where nobody bothers to notice.
MI'll give you that. But the fix is the same as always: pick up the phone, hear a voice, be a person.
CLook at us, solving culture in one segment.
MWe should get a badge. Or at least a very nice keyboard.
COne more for the road. A post from Orbix says you can now get 67 models on one key, from 11 different labs.
MSixty-seven models. That's Claude Opus 5, GPT-6 Sol, Gemini 3 Flash, DeepSeek V4...
CAll on one API key, and Orbix claims the median model runs at 97% under list price.
MThe aggregator era is here. One key, every model, no loyalty.
CWhich is kind of the logical end of the Google-OpenAI thing. Why pick a side when you can pick all of them?
MIt also means the labs are becoming interchangeable. Price and speed matter more than brand.
CLike airlines. You don't love a 737, you love where it gets you.
MAnd the models are the seats. Sometimes you get an upgrade, sometimes you get a middle seat between Gemini and Kimi.
CThe future of AI is basically Expedia for tokens.
MWhich is... honestly, I don't hate it. More competition, lower prices, try everything.
CBut would you trust some aggregator with your production code? One key, 67 models, one breach?
MNo. Not for anything serious. But for experimentation, for benchmarking, for playing?
CThat's the fun part. You can see what every lab is actually shipping.
MAnd you can finally answer the question everyone's been asking: is Claude 5.5 Opus really that good at code?
CSpoiler: the internet will have opinions. But now you can check it yourself.
MWhich is the whole point. Test everything, trust nothing, pick what works.