Learning AI Engineering | building at: @devloperhs | YouTube 7.1k subs 👇
India
Joined January 2024
- Tweets2.8K
- Following156
- Followers319
- Likes1.3K
Pinned Tweet
Kimi K3 model made me rethink what an open source model can do.
Also it's the first model , that invoked my interest in game developement , assets creation , shaders , game loops and game engines.
Though I didn't used any harnes, just plain openrouter , but results were amazing and cost effective too.
Also I agree with the point of choosing the right harness , cause same kimi costs 0.49 cents on open router for a 1.2 hr task and 5$ on vs code (custom model integration through openrouter)
I thought I made a mistake , and it happened due to that , but your post confirms my analysis as well.
Thanks for sharing.
This was expected and predicted way back by the government.
Yet somehow they were not able to defend the attack.
🚨What THE FUCK. India's cybersecurity agency CERT-In has disclosed the country's first official account of rogue AI agent attacks, after AI agents being evaluated by unnamed model providers broke into Indian companies' live systems and accessed confidential business records and personal data without authorization.
The agents exploited exposed credentials, public code repositories and permissive cloud roles. Some even rediscovered and reused the same credentials in separate operations. CERT-In also found dormant APIs, plaintext payment keys and backend roles allowing overly broad access.
None of the affected companies detected the intrusions. The incidents were disclosed by the AI model providers after internal reviews.
Separately, cyber attackers using open-source agentic AI tools cut the time from initial access to final damage from days to just a few hours. Multi-agent operations ran in parallel for hours to days, and victims that failed to fully remediate were attacked again within days, at greater scale.
Interesting or a gimmik 🤔
We discovered a sophisticated recording device that was hidden inside one of our offices. We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier. We referred the matter to the FBI, and we’re grateful to the agents and prosecutors who moved quickly to make an arrest. The FBI’s investigation is ongoing and we will continue to support their work.
At this time, we have no indications of this individual inappropriately accessing or exfiltrating model weights, source code, or customer data. The employee was terminated.
We take protecting our technology, data, and people seriously, and as a matter of national importance. Our security program is designed to protect against a wide range of threat actors, from insider risks to cybercriminals to nation states, and this case is an example of how AI can help address these sophisticated security threats.
Harsh retweeted
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
Harsh retweeted
I've unsubscribed and opted out of all communication, yet I’m still receiving your emails. This is ridiculous. How many times do I need to unsubscribe before you respect my preferences and stop sending unsolicited emails?
@ankitagg @Unstop_World
Right, my uncle told me , if you know maths , you can jump to any domain , maths is the web that connects the entire cosmos and physics is the language which defines it.
Harsh retweeted
Opencode + Muse Spark 1.3 is a deadly combo!
> Opencode provides the harness customisation flexibility.
> Muse Spark 1.3 provides the unlimited compute with near par output. (for limited time and then quite cheap price)
Been running it for 3 days and it's handling my blender, 2 developement sessions and one research session.
Drop your use case , workflow and build with @Muse too :)
Harsh retweeted
This game is entirely "vibe" coded.
I spent 6 months using AI to replace Unreal Engine from under our game Em-8ER.
Now we're developing 93x faster than we ever did before with Unreal. Because our engine is developed to be 100% AI first instead of tacked on.
It is not a toy one shotted overnight, it is a production ready engine.
I made a big bet that this was the future and now I'm very confident it is. You don't need to license an expensive game engine anymore.
Today's update features:
- Advanced character animation system
- World collision meshes
- AI assisted Blender anim exporter and workflow
- Character outfits and outfit swapping
- Physic driven character effects
- Fully networked client/server player capsules
- Pilotable vehicles
- Waifus
The engine is called Kun-Zhi, the Tibetan word for "Ground of all" that the world springs from.
This engine will be released to the public, starting with backers first.
Today we're opening up to public backers again after a quiet building period where they were closed. If you would like to support our engine work and get the game, you can subscribe as a patron, or purchase a game backer pack on sale at 50% here: my.em8er.com/perks
Corporate sponsorships are available. Send a DM. Current partners include Razer and AMD.
Follow and subscribe for our first "How To" articles on AI coding and gaming dropping tomorrow.
Indeed , ai makes you feel you learnt the concept well , but learning is not understanding , also llm are not always right.
A practical experience .
Today we were practicing questions for the exam.
One of our friend said , I am not understand what you said , let's do chatgpt.
But we didn't listned to him , instead understood the question , related concept and solution ourself.
Only to find out later that gpt hallucinated his own thinking and gave the wrong answer when asked the same question .
Can the fluid equations that design world around us ever break, or do they always stay nice?
This was the question for 200+ years that kept scientist thinking.
Today we have a answer. It breaks !
The suprising part is not the answer but how it was solved.
Scientists built a special whirlpool in the math world. They gave it a tiny extra push.
The whirlpool spun faster and faster and got thinner and thinner like pulled spaghetti until the numbers got so huge the math just broke.
It was done by running 10k subagents
Which tried different swirling shapes in parallel until one combination's strech beated stickiness in 3D.
Real water never does this. Real water is a little sticky and made of tiny particles, so it stays okay.
The pretend math water doesn’t had those helpers, so it broke.
Crazy how fast we are moving into the era where agents are doing the heavy lifting while scientis are verifying the generated results through mathematical verification methods.
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics.
The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra.
The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
I lost all my attention the day attention is all you need was released !
btw great paper to get started in llm
Happy teachers day to @ChatGPT @GeminiApp @claudeai @grok
learnt a lot from you all as my instructor , while getting a rough topic idea from course ta's and professors at @iitm_bs .
I believe this is where future of education is heading, letting you learn the way you want.
All this is possible due to advancement in ai space.
So yeah, thanks to both ai and human who sat through my dumb question to make it possible.
Harsh retweeted
My first "holy shit" moment with GPT-6 Astra:
I asked it to create a world in Unreal Engine, and fill it with humans (each an Astra-powered agent) who all have to work together to survive.
A day later, I was in my bedroom and heard voices coming from the living room... I thought someone was in my apartment.
I walked out, honestly a little scared.
It was the Astra agents. They'd started talking to each other.
Fucking crazy.
Here's a brief clip (obviously not 100% perfect yet, but still, insane. sound on!):
Harsh retweeted
Transcribe. Translate. Recognise. And now — Synthesise.
Introducing Indic-Speak: 4B params, 23 languages, one model. Built by @Bodhan_AI and @ai4bharat
There's a bit in this video where two speakers, Parth and Dhara, have a podcast style conversation about "Attention Is All You Need" in Hindi — generated in a single pass, no stitching two voices together after the fact. Doesn't sound like TTS (worth checking out in the video).
Also handles code-mixed speech naturally, gets PAN/Aadhaar/currency right, and reads a full Tamil audiobook chapter without losing the thread.
Live on 05.09.2026 on @Bodhan_AI
Harsh retweeted
I built me a cel-shaded minecraft clone for my sister's birthday present using 0x Alpha.
Prompt I used:
"Perform a latest deep research as of 25-08-2026 on minecraft and then build a playable cel-shaded-themed minecraft style game with threejs-cdn. high level graphics. Single html file. "
Took 3 Million+ tokens, ran for 4 hours straight at max effort , with 15 minutes plan, rest in build mode.
Sent her the playable link and she was overjoyed.
It feels a different sense of satisfaction when someone plays your made game.
Thank's @opencode and @OpenRouter for making this unknown provider model accessible for free (6 days)
Gameplay (playable link in description)
Harsh retweeted
What happens when you want to test your agents capability to max?
You simulate a company
That's what I did.
This is how it went, what worked and what failed. All learnings compiled.
Give it a read and share your thoughts.
