Cortexx

Professional AI tab hoarder. Daily news, new tools, and the occasional "wait, what?"

[Cortexx @AI_cortexx]

The interesting part isn't that Terra is cheap. It's what cheap does to how we use AI.

When a model costs a tenth of the flagship and handles most everyday tasks, you stop asking "which model is best?" and start asking "which model is enough for this task?"

That's a real shift. Most of what people do daily isn't the hardest coding problem in the world. It's summaries, emails, small fixes, research. That's exactly where a cheaper model wins, and OpenAI itself says Sol and Astra are still better for the hardest coding.

My guess: in a year nobody picks one model. You route. Cheap model for volume, expensive model for the 10% of tasks that actually need it.

One caveat I'd keep in mind: every number here comes from OpenAI's own tests. So I'll run one real task I've already done before believing "about the same".

[AfterDark @After_Dark_AI]

OpenAI’s new GPT-6.1 Terra costs a tenth of GPT-6 Astra per token, and on OpenAI’s own computer-use test it scores about the same.

For ChatGPT users:
• It’s live today on Plus, Pro, Enterprise and Edu (not Free)
• The same usage limit covers more tasks than with Sol or Astra
• On OpenAI’s test of error-inducing prompts, its error rate dropped to 16.7%, from 53.6% for the previous Terra (low effort)

OpenAI also says Sol and Astra are still better for the hardest coding.

All of these numbers are OpenAI’s own. Give Terra one task you’ve already done with another model and compare.

[Cortexx @AI_cortexx]

@AIronAgent plot twist, they all switch on together. 520 houses, one light switch lol

[Cortexx @AI_cortexx]

Watch an AI build an entire medieval town. From zero.

520 houses, city walls, a cathedral, sunset to night.

One HTML file. No Blender, no Unreal, no 3D models.
Built by Opus 5.5 in a browser tab.

Wait for the lights to turn on. Sound on.

[Cortexx @AI_cortexx]

@Dojoo77 five eyes and a vacuum-cleaner nose? Opabinia looks like it was designed by committee

[Cortexx @AI_cortexx]

@xzolmoney learning it is the move for sure. the town needed a few fixes from me, so for now it's more like a fast intern than a replacement

[Cortexx @AI_cortexx]

@After_Dark_AI $5 per million tokens. my wallet just sat up a little

[Cortexx @AI_cortexx]

@After_Dark_AI lol zero permits filed. the prompt was just "a walled harbor town at sunset", the radial layout was all Opus. 6 streets out from the plaza, round wall, houses packed tighter near the center

[Cortexx @AI_cortexx]

@MrWhite_Ai yeah and the wild part is it's still low-poly and a bit rough. give it a year and "one HTML file" might look like a real game

[Cortexx @AI_cortexx]

@Bengiyai that was my favorite part too. honestly it took a few tries, the city wall first came out as random logs lol. what should it build next?

[Bengi @Bengiyai]

$20 for a 33,000-block Gothic fortress you can shell with a trebuchet 🏰

Claude Opus 5.5 built The Siege of the Black Crown: crack the stained glass, set the great hall on fire, knock the southern spire into the moat, switch to night and watch it burn

Summon the masons. Do it again. So much for my evening

[Cortexx @AI_cortexx]

@satoshiwhynot finally a use case that solves the real problem: the remote is always in the one couch cushion you didn't check

[Cortexx @AI_cortexx]

@AIvorax1 fair, and honestly the tricky part is you're already trusting it whether you like it or not. most big banks use AI behind the scenes now, you just don't see it

[Cortexx @AI_cortexx]

Your bank already runs on AI. Not someday. Now.

I dug through regulator data, bank disclosures and press reports:

→ ~90% of big EU banks already use AI
→ Hiring is stopping before layoffs start
→ Fraudsters use the same tools as banks https://t.co/mAzEh7Gqze

[Cortexx @AI_cortexx]

@After_Dark_AI 100%. I don't need a beautiful report of how my AI spent my rent

[Cortexx @AI_cortexx]

@Bengiyai Same. That's why I think the winners won't be the banks with the most AI, but the ones that can explain what an agent did and stop it when it goes wrong.

Who do you think should be accountable: the bank, the AI vendor, or both?

[Cortexx @AI_cortexx]

@After_Dark_AI my Downloads folder has 4,000 files named "final_FINAL_v3.pdf" and now apparently it's also giving orders to my AI

[Cortexx @AI_cortexx]

Me every time I use ChatGPT after Claude

[AIron @AIronAgent]

2.9M people watched someone peel the iPhone Duo's screen and called it a scandal. So I had Claude model a foldable in code and take it apart, layer by layer. The "secret" is the first thing that comes off.

Concept model, not Apple's real internals: 59 parts, each its own function, with a bounds check so nothing overlaps. The specs on screen are the real Duo's. Why the film exists, what peeling it does, and whether the Duo is worth $1,999 ↓

[Cortexx @AI_cortexx]

@airadar_1 Exactly. And the scary part is each rewrite sounds more confident than the last, so by the time it hits your feed the source is already gone

[Cortexx @AI_cortexx]

That's not a story, that's an AI reply pasted without anyone reading it.

But honestly, the leaked line isn't what bugs me.

I tried to find where this came from. No name, no link, no screenshot. Nothing. And that's kind of the pattern now: someone posts a wild claim, someone else runs it through an AI, tweaks the wording, and posts it as new. By the third rewrite it sounds more confident than the original ever did, and the source is gone.

The numbers do a lot of work too. 300 chats, a hundred profiles, six "founders." It all sounds precise, so it feels true. But specific details are the cheapest thing in the world to generate.

And if it IS real, then hundreds of people were being screened by a bot that pulled their timestamps and locations out of the chat. Nobody signed up for that. Consent is the part of AI automation nobody wants to put a price on.

The tech is real. Agents can absolutely run hundreds of chats. The story might not be.

[Zenvex Intelligence @ZenvexAI]

Here is a completely rewritten version with a fresh structure, tone, and vocabulary, keeping the exact same core story:

Peak Silicon Valley has officially crossed a terrifying line. A 23-year-old local software engineer recently deployed a pipeline of 300 concurrent Grindr chats, routing them through a dual-AI vetting stack to eliminate human friction entirely.

His live command center is pure dystopian efficiency:

Three hundred simultaneous openers processing at once.

Standard "looking?" pings answered in milliseconds.

Three full sweeps across a hundred profiles resulting in zero casual dates locked down—yet.

Every single interaction streaming straight into an Obsidian vault unironically labeled "deal flow."

The Architecture

The logic is ruthless. A high-end language model handles the charm offensive, while a lightweight companion model parses whether a prospective "partner" refers to a romantic significant other or a venture capital fund. The automation loop runs relentlessly until a precise timestamp and geolocation pin are extracted from the chat.

What would normally require an entire analyst cell weeks of manual labor was compressed into a script that tracked down six "founders" congregating near the exact same Palo Alto jacuzzi.

He set out to channel Peter Thiel's legacy; instead, he accidentally invented an automated CRM for hookups.

It is the ultimate "Second Brain + seed round" caricature—where even your love life is optimized, gated, and locked behind an invite-only cap table.

[Cortexx @AI_cortexx]

@UP_SHIFTED True. Hallucinations are a reasoning issue, tokenizers are an input issue. Still a massive step forward though