AI Tech Updates

AI • Technology • Open Source🔥🔥🔥

[AI Tech Updates @AIInfoShare]

LinkedIn vs the lab.
Alexandr Wang, Meta’s Chief AI Officer. Tibo, Codex & ChatGPT at OpenAI.
A few weeks in an AI lab, apparently. 😅
#AI #OpenAI #Meta #Codex

[AI Tech Updates @AIInfoShare]

This is the part that actually matters: not that the cipher was impossible, but that Astra ran the whole loop — transcribe 1,300 hand-drawn signs, recover a key, crack the rest, and fix the date — in about 6 hours from one image and one goal. Unsolved backlogs don’t need more specialists. They need curious outsiders with a model that can do the boring cross-disciplinary work. 🔐📜
#AI #GPT6 #Napoleon #Cryptography #History

[Carter Church @CarterWChurch]

I used GPT-6 Astra to break an unsolved cipher to one of Napoleon's generals that had gone unread for 217 years.

What makes this impressive isn't actually the codebreaking, but that Astra completed the entire multi-modal workflow in ~6 hours from a single image and goal.

1/6🧵

[AI Tech Updates @AIInfoShare]

Tencent just inked its biggest overseas lease ever with Oracle: a five-year deal across multiple data centers in Southeast Asia, worth about $7 billion, with roughly 30% paid upfront. 🤝
The pact gives Tencent access to around 100,000 advanced AI chips it can’t get inside China, fueling its models and agentic tools. ☁️💻 That upfront payment also helped push Tencent’s Q2 free cash flow negative for the first time in over a decade — about Rmb 13.8bn ($2bn). 📉
Reported by the Financial Times, citing people familiar with the matter. Neither company immediately commented. 🗞️
#Tencent #Oracle #TencentCloud #AI #AIChips #CloudComputing #Semiconductors #TechNews

[AI Tech Updates @AIInfoShare]

🤣🤣🤣

[Xanderwow @Xanderwow_A]

你最喜欢哪家的Si大模型？

视频模型：Seedance2.5
舞蹈/音乐：極楽浄土 - GARNiDELiA (ガルニデリア)

[AI Tech Updates @AIInfoShare]

Hahahaha, this is too funny 😂 Does releasing a new AI model really just mean changing a few numbers? 🤣

[Spectro @Spectromachina]

Introducing Steve v1.0 🧠

Our new frontier AI model, built to push agentic intelligence to absurd levels.

• 99.8% Vals Index
• 98.7% AutomationBench
• 98.4% DeepSWE

Steve v1.0 is currently available to a limited group of trusted testers.

[AI Tech Updates @AIInfoShare]

🚀 Google just unveiled Gemini 4 Argon — its next-era frontier model!
💻 Leads on long-horizon software engineering (DeepSWE v1.1: 77.9%) 📊 Tops knowledge-work benchmarks including Vals Index (68.9%) and finance/legal agents 🛡️ Strong on cyber defense (ties GPT-6 Astra on CWE-bench) 📝 Industry-first 1M token output limit (up from 64K) for deep, multi-step workflows
Rolling out first to trusted cyber defenders via the Fairwind Program. Intro pricing: $2 / $10 per million input/output tokens.
#Gemini4 #GoogleAI #Argon #AI #Cybersecurity

[Google @Google]

Today we’re introducing Gemini 4 Argon.

It delivers frontier performance in complex workflows across real-world software engineering, knowledge work, and cybersecurity defense with an industry-leading 1M token output limit.

[AI Tech Updates @AIInfoShare]

Just came across Ant Group’s latest Ling-3.1-Flash, and this model is pretty interesting. 👀

It uses a Mixture-of-Experts (MoE) architecture with 560B total parameters, while activating only around 25B parameters per token.

What’s even more interesting is that it’s not just built for chatting. It focuses heavily on coding, complex reasoning, multi-step tasks, and AI agents. 🤖

It reminds me of the three key metrics I’ve been looking at on Artificial Analysis:

🧠 Intelligence
⚡ Speed
💰 Cost per Task

The AI model race may be shifting from “who has more parameters and higher benchmark scores?” to:

Who can complete complex tasks faster and at a lower cost?

If Ling-3.1-Flash can combine the capacity of a 560B-parameter model with the efficiency of activating just 25B parameters, it could be a good example of where AI models are heading:

More model capacity + better inference efficiency + lower cost per task.

The next AI race may not just be about who is smarter, but who delivers more Intelligence per Dollar. 🚀

#AI #AIAgents #LLM #ArtificialIntelligence #Ling31

[AI Tech Updates @AIInfoShare]

I’ve been digging through the latest charts on Artificial Analysis, and I think the AI race is becoming much more interesting. 👀

The Intelligence Index is no longer just about traditional benchmarks. It now combines 10 evaluations covering real-world work, agentic workflows, coding, science, reasoning, knowledge reliability and long-context tasks.

But the score is only one part of the story.

Artificial Analysis also tracks ⚡ Speed and 💰 Cost per Task.

That creates a much more interesting question:

How much intelligence are we getting for every dollar and every second?

A model can score extremely high, but if it takes much longer or costs significantly more to complete the same task, its real-world advantage may be smaller than the headline score suggests.

To me, the next phase of AI competition looks less like a simple leaderboard and more like an economic optimization problem:

🧠 Intelligence × ⚡ Speed × 💰 Cost

The next AI leaders may not simply be the models with the highest scores, but the ones that can turn intelligence into useful work at the right speed and cost.

The AI race is becoming an efficiency race. 🚀

#AI #ArtificialIntelligence #AIAgents #LLM #MachineLearning

[AI Tech Updates @AIInfoShare]

On Polymarket, people are betting on which AI company will have the best model by the end of October.

Anthropic is currently #1, followed by Google and OpenAI. Meta is in 4th with less than 1% of the votes.

People are also predicting when Gemini 4.0 will launch, with October 31 and November 30 currently having the highest probabilities.

So… do you think Gemini 4.0 is coming in October? 👀🤔

#AI #Gemini #Google #Anthropic #OpenAI #Polymarket

[AI Tech Updates @AIInfoShare]

🤖 https://t.co/JJ2rssmyC0’s OpenVuln is an AI-powered vulnerability discovery tool driven by GLM. It can scan open-source GitHub projects, analyze code, and automatically identify potential security vulnerabilities.

This shows AI is moving beyond “helping you write code” toward auditing code and discovering vulnerabilities on its own. 🛡️

AI cybersecurity is becoming a new battleground for the next generation of AI agents. ⚔️

#AI #CyberSecurity #GLM5 #OpenVuln

https://t.co/LfoWAi3Cvn

[AI Tech Updates @AIInfoShare]

🚨 Anthropic’s latest research shows that GLM-5.3’s cyber capabilities are already approaching Claude Mythos Preview, with the model able to autonomously discover vulnerabilities and build complete exploit chains.

What’s even more concerning is that GLM-5.3 is open-weight, and its safety mechanisms can be bypassed.

AI-powered cyber warfare is entering a new phase: AI isn’t just writing code anymore—it’s starting to find and exploit vulnerabilities on its own. 🤖⚔️

#AI #CyberSecurity #GLM #Claude

https://t.co/NyxVDrQ8xa

[AI Tech Updates @AIInfoShare]

🚨 Anthropic’s latest research shows that GLM-5.3’s cyber capabilities are already approaching Claude Mythos Preview, with the model able to autonomously discover vulnerabilities and build complete exploit chains.

What’s even more concerning is that GLM-5.3 is open-weight, and its safety mechanisms can be bypassed.

AI-powered cyber warfare is entering a new phase: AI isn’t just writing code anymore—it’s starting to find and exploit vulnerabilities on its own. 🤖⚔️

#AI #CyberSecurity #GLM #Claude

https://t.co/NyxVDrQ8xa

[AI Tech Updates @AIInfoShare]

🏠 Bloomberg: Apple enters the smart home on Oct 13 — CEO Ternus’s first big fight.
Main course: J490 home hub. ~6-inch square display, wall-mount or tabletop. Recognizes who is in the room and personalizes content.
Sides: first HomePod mini update in 6 years (new pink/green) + new Apple TV. Same look, faster chips for Siri AI.
All three were slated for 2024, delayed by AI. Still coming: a privacy camera that doesn’t record video, and a 9-inch screen on a robotic arm. HomeKit expands categories and pulls in third parties.
Apple’s actually serious this time. #Apple #SmartHome #HomePad #SiriAI #AppleEvent

https://t.co/q4nJtpy1OH

[AI Tech Updates @AIInfoShare]

I asked my Muse agent to make three cats dance for me 😂🐱🐱🐱 What do you think of their moves? 💃🕺 Music: Raving Energy by Kevin MacLeod (https://t.co/f2vGSVNpgV), CC BY 4.0

[AI Tech Updates @AIInfoShare]

This video is seriously so damn cool! 🔥🤩

[edaᚐ҉ᚐ @edosnury]

CILDIRIYORUM BUNU HANGINIZ YAPTI AHAHSHDHDHD

[AI Tech Updates @AIInfoShare]

☕️☀️

[AI Tech Updates @AIInfoShare]

AI is starting to improve the efficiency of government operations. 🤖🏛️👍

[kanav @kanavtwt]

I can't believe this is a government website

what the fuck.

[AI Tech Updates @AIInfoShare]

🤖 AI is evolving at an incredible pace. It feels like a new model is being released every 11 days now. 🚀

Could AI eventually become capable of evolving on its own? 🧠

If so, we could see AI models constantly changing and improving in real time. ⚡️🔥

[josh :) @jfonsecarivera]

you're not crazy

between anthropic and openai, a new model used to come out every ~10 weeks

now it's every ~11 days

[AI Tech Updates @AIInfoShare]

Didn’t OpenAI think to secure the https://t.co/btsJCcRhQH domain before launching Dots? 😂

Why does https://t.co/btsJCcRhQH now redirect to Grok Bot? 🤣

According to WHOIS data, the domain information was updated on September 28, 2026.

[AI Tech Updates @AIInfoShare]

The battle between AI assistants is heating up. The ones that offer the most powerful features at the lowest cost will ultimately come out on top.

[Yuchen Jin @Yuchenj_UW]

I tried Grok Bot 3 weeks ago.
Installed Instint 2 weeks ago.
Installed Muse last week.
Now apparently I need to try Dots.

Let the personal AI assistant wars begin.

[AI Tech Updates @AIInfoShare]

How to use OpenAI Dots
Dots are always-on agents inside ChatGPT. Unlike a normal chat, a Dot keeps working after you leave, uses its own cloud computer and browser, and only pings you when it needs a decision.
Who can use it
ChatGPT Pro (100 / 200 / 500) or Business Premium. Pro rollout excludes the EEA, UK, and Switzerland. Access is rolling out gradually. Create a Dot on desktop web or the desktop app — not on mobile web. After that, you can talk to it in the mobile app. Your first Dot is included; usage does not count against plan limits for the first month.
Set it up
1. Open https://t.co/X3VaBpA3TP or ChatGPT on desktop.
2. Create your Dot, give it a name and look. The handle starts as @yourname-dot.
3. Let it introduce itself and suggest what it can take on.
Give it work
Tell it a goal, not just a one-off question. Examples:
• “Check my calendar each morning and flag conflicts.”
• “Research this topic, draft a doc, and update me when it’s ready for review.”
• “Track this project and ask before you send anything out.”
Use + to attach files or photos. In the Dot’s profile, check In progress, Scheduled, and Completed. Open its cloud computer to watch the browser. If a site needs your login, choose Take over, finish the step, then Return control.
Connect tools (these are separate)
• Plugins / apps — Profile → Customize → Plugins. Connect Gmail, calendar, Drive, GitHub, etc.
• Slack / Teams — add the channel on desktop. Being in a channel does not mean it is watching it; tell it what to monitor.
• Your computer — off by default. Enable it in the desktop app. The machine must stay online with the app open. Only one personal computer at a time.
Texting is a limited US Pro beta. Dots cannot call you yet; you can start a voice call from the conversation.
Keep control
In Customize → Custom rules, set whether it can act on its own, act only if you already asked, ask first, or hand off to you. Pause from the profile ••• menu. Reset deletes the Dot, its chats, memories, and scheduled tasks.
Best use
Give it ongoing work: research, drafts, calendar, follow-ups. Check important results yourself. Official guide: https://t.co/EpCpxMOean

[Tibo @thsottiaux]

Announcing dots. Dots work 24/7 for you, learn from your feedback, have their own computer, browser and can be connected to over 4k apps in our ecosystem.

They’re powered by Astra our best model yet. Included in your Pro plan, without drawing down on any of your usage. You can even call a dot while it’s working.

We’re getting you started with your primary dot today and soon you’ll be able to create entire teams of them.