🤖 AI • Research • Tech AI papers, new models, interesting projects & useful tools. Exploring what’s next in AI — one post at a time. 🔬 IISER Bhopal
Joined August 2026
- Tweets141
- Following358
- Followers61
- Likes125
The AI war isn’t just about technology anymore—it’s about who gets to decide what AI can say and create. ⚖️🤖
Interesting twist: the AI regulation battle is now becoming a fight over free speech, federal vs. state power, and America’s AI competitiveness. ⚖️🤖
The hard part is protecting people without regulating innovation into irrelevance.
NEWS: The U.S. Department of Justice is backing Elon Musk’s xAI in its legal challenge to Minnesota’s new AI image-generation law.
The DOJ says the law goes far beyond federal rules and could punish AI platforms even when adults consent or create images of themselves.
The DOJ also warned that overly broad state laws could restrict protected speech, slow AI development and hurt America’s global leadership in AI.
The next space race won’t just be about reaching orbit—it’ll be about who can build and operate there. 🚀🤖
On-orbit robotics could completely change how we think about satellites, servicing, and even defense.
Where robotics meets defense missions.
S3, our robotic satellite for on-orbit servicing, is fully integrated and flight-ready.
It reflects the vertically integrated capability we’ve built at GITAI, satellite buses, robotics, avionics, software, manufacturing, integration, and test.
We’re applying that same capability to LEO constellations for missile defense, Earth observation, and on-orbit servicing.
We’re looking for people who are hungry to build, take on real responsibility, and own a big piece of the system. If that sounds like you, come build with us.
grnh.se/k4q7j83y8us
Humans spent decades chasing this record. Robots showed up and said, “Move over.” 😂🤖🏃♂️
🚨HUMAN WORLD RECORD BROKEN: Tiangong just clocked 2:21.6 in the 1500m at the World Humanoid Robot Games.
The human world record had stood at 3:26.00 since 1998.
Tiangong beat it by over a full minute.
For context, last year’s champion, Unitree H1, ran 6:34.40.
#WHRG2026
The recovery is almost more impressive than the jump. 🤖😂
Robots are learning one of the most human skills: falling, then getting back up.
Swift recovery! At the 2nd #WHRG, a humanoid robot tumbled on long jump landing, yet quickly regained its footing. #WorldHumanoidRobotGames #HumanoidRobot #LongJump
The irony is hard to miss: chip restrictions were meant to slow China’s AI progress, but scarcity pushed Chinese labs to innovate harder on efficiency.
Sometimes constraints don’t stop innovation—they accelerate it. 🤖
China’s AI startups are closing the gap on OpenAI and Anthropic.
Leap was powered by a single Tsinghua University computer science lab.
> Professor Tang Jie and student Yang Zhilin founded AI unicorns Z
> AI and Moonshot. Chinese labs used open-source code and model distillation
> DeepSeek invented MLA architecture to reduce memory use on scarce chips
> Chinese startups built models on less than one-fifth the capital of US rivals. Z
> AI made $1 billion in annual recurring revenue after going public.
so US chip bans have improved hardware efficiency, and Chinese models have become more efficient.
🇨🇳 China has solved its low birth rate problem…
Robot kids are here. 🤣🤖
🇨🇳 China's solved its low birth rate problem...
Robot kids 🤣
Writer: Ian
xitter.cf/pianhangx/status/20911…
The day robots start competing against humans, “robot Olympics” stops sounding like science fiction. 🤖🏅
🤯 Humans are becoming the minority users of AI.
According to a16z, AI agents now consume nearly 5× more tokens than humans — up 14× since February.
The shift is clear:
AI is moving from answering prompts to doing the work. Source: a16z
Humans are the minority user of AI
Agents burn nearly 5x the tokens people do, up 14x since February
Charts of the Week: a16z.news/p/charts-of-the-we…
🚨 Grok Imagine is climbing the AI image-generation leaderboard.
It’s now ranked #3 on Arena for text-to-image generation. 🎨
The AI image race is getting seriously competitive. 🔥
Grok is closing the gap fast.
BREAKING: Grok Imagine is now ranked #3 on the Arena Leaderboard for text-to-image generation. 🎨🤖
Grok Imagine continues to climb the AI image-generation rankings, placing among the top models in the benchmark.
The competition in AI image generation is heating up. 🔥
Source: Arena Leaderboard
Norway is spending $2.3B on 21 U.S. Black Hawks 🇺🇸🇳🇴
The package includes missile-warning systems, countermeasures, 46 engines, weapons, communications and Arctic winterization.
With Russia on its doorstep, Norway is clearly strengthening its NATO northern flank.
🇺🇸🇳🇴 Norway's about to drop $2.3 billion on American Black Hawks.
The State Department just cleared the sale, 21 UH-60M choppers plus a whole pile of extras.
We're talking 46 engines, missile-warning gear, countermeasures, comms and nav kit, machine guns, rescue equipment, and even winterization gear for the cold up north.
Here's the interesting part. Back in June, Norway walked away from the pricier, more specialized HH-60W "Jolly Green II" rescue version.
This time they went for the UH-60M, which does a bit of everything instead of one job really well.
Norway sits right on NATO's northern flank next to Russia, so this is Oslo quietly beefing up while the U.S. keeps its gear flowing into Europe.
Source: U.S. State Department / Writer: Daniyal
🚨 Grok 4.5 tops VulcanBench Eval Suite 3.
90% Pass@1, ahead of Claude 5.1 (89%) and DeepSeek V4-Flash (88%).
The benchmark uses 23 real-world software engineering tasks from merged open-source PRs.
Grok is setting the pace in AI coding. 🤖
AI agents just got a serious RL upgrade.
ClawGym II improved Qwen3-30A3B’s Pass@1 by 9.98 points on OpenClaw and 14.81 on Claude Code.
One model, multiple agent harnesses. 🤖
Agent training is moving beyond the model itself.
Grok 4.5 is still holding the #1 spot on VulcanBench v3.
The benchmark tests 23 real-world software engineering tasks based on merged open-source PRs.
21/23 solved 91.3%.
That’s not chatbot trivia. That’s actual engineering work.
Grok is becoming a serious coding-agent.
BREAKING: Grok 4.5 is still sitting at #1 on VulcanBench Eval Suite 3.
This benchmark is built around genuinely hard, real-world software engineering — 23 frontier-level tasks taken straight from actual merged open-source pull requests.
And once again, Grok 4.5 is right at the top.
Its engineering performance is just on another level @elonmusk
