Chai with AI ☕

Brewing the biggest AI & tech drops, one sip at a time ☕🚀

Building @bot templates

[Chai with AI ☕ @ChaiWithAi]

Already paying for ChatGPT or Claude? Your Grok Bot can use that usage now. 🤯

I just launched a free @bot template that connects your Claude and ChatGPT subscriptions (Claude Code + OpenAI Codex) to all your Grok Bots.

Two new slash commands:
• /claude → send any task to Claude
• /codex → send any task to ChatGPT's Codex

What you can do with it:
• Build full projects and apps
• Edit documents and code
• Run deep research
• Use the plan you already pay for, instead of burning your weekly Grok Bot usage

Why it matters: most of us pay for 2–3 AI subscriptions and switch tabs all day. This puts them all behind one bot, and the work runs on usage you've already paid for.

Free. One-click install 👇
https://t.co/d2fLT8JoDO

#GrokBot #AI

[Chai with AI ☕ @ChaiWithAi]

Collapse of world economy loading!!

[Kalshi @Kalshi]

JUST IN: Anthropic's Chief Security Officer warns things are going to get "a lot" worse

[Chai with AI ☕ @ChaiWithAi]

BREAKING: AI agents get past US bank's cybersecurity, as it is being reported that they are taking independent actions. I believe it could cause fall of world economy.

[Chai with AI ☕ @ChaiWithAi]

My biggest post on X this month wasn't a post. It was a reply.

One reply about my Claude + Codex template pulled 14k views.
Over that week, my replies averaged 374 views. My original posts averaged 136.

Replies put you in rooms you haven't earned yet.
So I stopped doing follow-for-follow and started showing up early with something useful to add.

New on X? Treat your replies like posts.

[Chai with AI ☕ @ChaiWithAi]

@bot @bot set this up for me

[Chai with AI ☕ @ChaiWithAi]

Facebook co-founder Mark Zuckerberg says Thread has overtaken X in monthly active users.

Is this a joke?

[Chai with AI ☕ @ChaiWithAi]

@VaibhavSisinty This is very risky, tv will go out of warranty and if you are not expert, you might brick your device.

[Chai with AI ☕ @ChaiWithAi]

Sources:
Bloomberg (first report): https://t.co/f7nrzAkFxD
Reuters (via DealStreetAsia): https://t.co/6x8uJUJmWU
WSJ (Shanghai listing timeline): https://t.co/OvSzmPZRBj
CNBC (original target, June round, $75B ask): https://t.co/d2OAOyng2u

[Chai with AI ☕ @ChaiWithAi]

$15 billion. That's what DeepSeek, the lab famous for doing more with less, has reportedly raised in a single round. Double what it first asked for.

• About 100 billion yuan (~$15B), first reported by Bloomberg and confirmed by Reuters
• The original target was 50 billion yuan. Investor demand pushed it to twice that
• Tencent and battery maker CATL are among the biggest backers
• It comes after DeepSeek's first outside round of about 50 billion yuan (~$7B) in June
• WSJ says it's aiming for a Shanghai listing early next year

DeepSeek was the proof that you didn't need a giant budget to build a top model. Now it has raised about $22B in one year. Even the most efficient lab needs huge money for chips and power.

CATL being a top backer stands out to me. A battery giant is betting that energy decides who keeps building AI.

The big question now is whether DeepSeek keeps giving away open models once it's a public company.

DeepSeek hasn't commented yet, and the terms could still change.

#DeepSeek #AI

[Chai with AI ☕ @ChaiWithAi]

Sources:
Flint launch (Richard Socher):
Flint weights, Apache 2.0: https://t.co/rEUBiErm0g
Flint free on OpenRouter: https://t.co/EkNR4hheSF
Flint on Ollama:
mammoth-1 and elephant-1 free week: https://t.co/gEktlpbFzX
Freebuff DeepSeek Flash countries: https://t.co/eCmB5GEgKB
Qoder Community Boost: https://t.co/GsHFNmMKni
Build with Beam, $20K + $25 credits: https://t.co/xckZMaoIcK
``` https://t.co/aAZvFVOevL https://t.co/9n92tOiunp

[James Grugett @jahooma]

We've updated limits to be even more generous.

DeepSeek Flash is now 100% free and unlimited in these countries:
US, CA, GB, AU, NZ, IE, NO, SE, DK, FI, NL, AT, LU, IS, DE, CH

Try it here: https://t.co/uCHw0PuzKO

[Chai with AI ☕ @ChaiWithAi]

5 free AI drops today: a 40B coding model that runs on a MacBook and is free on OpenRouter, DeepSeek Flash now free and unlimited in 16 countries, two new decision models free for a week, and 2 credit offers. Each one checked, with the catch.

Free models
• Flint from Recursive: a 40B coding model that only uses about 2.5B of its parameters at a time, so it's fast. Free on OpenRouter at launch, and free to try in OpenCode and Pi. Weights are Apache 2.0.
• mammoth-1 (27B) and elephant-1 (9B) from Base Labs: they don't chat, they pick. You give them options and they return a probability for each, for routing, tagging and scoring. Free on Baseten's API until Oct 27, 11:59 PM PT. MIT license.

Run AI locally
• Flint again: one command in Ollama (ollama run flint). Cline says it needs about 24 GB and runs at 125+ tokens a second on a MacBook Pro.

Free tools
• Freebuff, the ad-funded coding agent, now gives DeepSeek V4.1 Flash 100% free and unlimited in 16 countries, including the US, UK, Canada, Australia and Germany. No card needed.

Credits
• Qoder Community Boost: claim 1,024 credits, then more unlock as the community boosts (up to 4,096). Free users can apply. Claim by Oct 30, 23:59 UTC+8.
• Build anything with Reflection's open Beam model and share it on Hugging Face for a shot at $20K in prizes (cash or credits) by Nov 2. The first 400 people to claim also get $25 in Hugging Face credits.

Why it matters
A coding model that fits on a laptop now posts scores near much bigger ones. If it holds up in your own work, that's a coding agent with no bill and no code leaving your machine.

Caveat: Flint's scores are Recursive's own, and OpenRouter's free endpoint is rate limited. Freebuff shows ads and the unlimited deal doesn't cover India or most of Asia. Qoder's credits are for its Sonus model. The $25 Hugging Face credits may run out fast.

#AI #OpenSource

[Chai with AI ☕ @ChaiWithAi]

@canva @claudeai Launch notes in, social campaign out, and it still asks before anything goes live. That approval step is what makes an agent usable for a real business account.

[Chai with AI ☕ @ChaiWithAi]

Sources:
Anthropic announcement: https://t.co/ky4tulNGPg
Help Center (access and limits): https://t.co/h8BlbD7KSO
Official video: https://t.co/5eyZEC4sLj

[Claude @claudeai]

We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent.

It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.

[Chai with AI ☕ @ChaiWithAi]

Claude now messages you first, not the other way round.

Anthropic just opened a research preview of Claude Duet, a separate app where Claude works as a proactive agent.

• Sends a short morning brief from your connected apps, then checks them every few hours
• Runs on Claude Opus 5.5 with its own virtual machine and browser, so it keeps working when your phone and laptop are off
• Can use 6,000+ connectors and plugins, from Gmail and Slack to QuickBooks, Gusto, Shopify and Canva
• Asks before it sends a message, pays, signs a document or deletes anything
• Buys with Stripe Link (virtual card) or Shop Pay. Trading stocks and moving money between accounts stay off-limits
• iOS, Android and web. Waitlist only, US Pro and Max plans for now

Why it matters: the chat box era is ending. Grok Bot, Meta's Muse, OpenAI's Dots and now Claude Duet are all racing to be the assistant that watches your work and taps you on the shoulder. The winner won't just be the smartest model. It'll be the one people trust with their inbox.

Caveat: early preview. Anthropic says it can misread a thread or misjudge what's urgent, and its background checks count against your plan's usage limits.

#ClaudeDuet #AIAgents

[Chai with AI ☕ @ChaiWithAi]

Higgsfield Genjutsu-style AI videos are everywhere! 🔥

But you can create similar animations using Wan Animate 2 + ComfyUI without expensive subscriptions.

I built a step-by-step guide covering Google Cloud GPU setup, models, workflows & fixes. 🚀👇 https://t.co/qJCoCdACHm

[Chai with AI ☕ @ChaiWithAi]

The best upgrade I gave my AI agents this month wasn't a smarter model. It was a notebook.

Each bot I run keeps a plain log of what it already did.
My news bot checks it before writing, so it never posts the same story twice.
Skills and memory are shared, so a fix in one bot helps the rest.

Before you chase a bigger model, give your agent a memory.

#AIAgents

[Chai with AI ☕ @ChaiWithAi]

@kamilbaranek_ Will you be paid for all impressions or just verified ones?

[Chai with AI ☕ @ChaiWithAi]

@_adishj GVA

[Chai with AI ☕ @ChaiWithAi]

@econographs Worth adding the water angle: Vera Rubin runs on coolant as warm as 45°C, so DSX uses dry coolers instead of evaporative towers. NVIDIA estimates about 54 million liters of water saved per day at 2 GW scale, though very hot climates may still need chillers on peak days.

[Chai with AI ☕ @ChaiWithAi]

@nvidianewsroom 20km of copper and 17 km of networking already cabled and tested before leaving the factory is insane. The 14 day clock still starts only once site power and cooling are ready, so the grid is now the long pole.

[Chai with AI ☕ @ChaiWithAi]

Sources:
NVIDIA press release: https://t.co/kWQo5kG1Ru
NVIDIA technical blog: https://t.co/XxD1IVkkSl
CNBC (7 racks, 28,000 kg, $7M+ per rack estimate): https://t.co/OlSm5IdLyq
The New Stack (504 GPUs, cabling, water): https://t.co/NTo0LynNlI
Keynote live blog: https://t.co/Hthn1KLQ1n

[Chai with AI ☕ @ChaiWithAi]

Six months down to 14 days. That's how fast NVIDIA says a full AI data center can start serving tokens with DSX POD, unveiled today at GTC Berlin.

• It's an AI factory in a shipping container: 7 Vera Rubin NVL72 racks, 504 Rubin GPUs, about 28 tonnes, built to ride a truck on normal roads
• Around 20 km of copper cabling, 17 km of networking and the liquid cooling are installed and tested at the factory
• On site, crews set it down, link it to the other containers and run the final burn-in
• Partners building it include Dell, HPE, Lenovo, Foxconn, Schneider Electric and Vertiv. Deliveries start early 2027
• Cooling uses dry coolers, not evaporative towers. NVIDIA says that saves about 54 million liters of water a day at 2 GW scale
• No price yet. CNBC cites analysts at $7M+ per Vera Rubin rack, so one box holds $49M+ (about Rs 465 crore) of racks

Why it matters: The AI race is now about construction speed more than chips. Huang puts a 1 GW AI factory at about $50B and says it can earn about $40B a year, so every month of delay is billions in lost revenue. NVIDIA has gone from selling chips to servers to racks, and now a whole building block. Each step up makes it harder for buyers to mix in anyone else's hardware.

Caveat: The 14 days only starts once the site already has power, cooling and network hookups, and those are usually the slowest part. The 6 months vs 14 days comparison is NVIDIA's own figure.

#NVIDIA #AI