Rabimba

Something I helped built with @developers.google.com is live now :)
developers.googleblog.com/beyond-the-c...
But I am incredibly proud of what we all built with @developers.google.com

Will be at @neuripsconf.bsky.social this weekend.
Come say Hi if you are there.
Would love to meet and chat.

The Context Trap: Why Scaling Laws Can’t Break the Ceiling of Uncertainty A blog about my journey through Mozilla,Open Source Contribution, Virtual,Augmented,Mixed Reality, Security,Hackathon and in general technology.

Jotted down some of my unconventional takes on "AI". Or what we call AI nowadays in applied science, I guess.....


rkrants.blogspot.com/2025/11/the-...

YouTube video by Rabimba Karanjai AI Powered Smart Contract Repair - AIWare

Smartify Talk is up. Not really a polished one, but this is what was presented in @aiware-conf.bsky.social

www.youtube.com/watch?v=s65v...

Wish I could be there to receive it personally, but honored nevertheless @ACM@mastodon.acm.org @aiware-conf.bsky.social

Lei was able to send it back to us :D

@ACM@mastodon.acm.org @aiware-conf.bsky.social

Excited to have 3 papers presented today & tomorrow at BRAINS (Zurich) and AIWare (Seoul)! 🌍

While I can't attend,my team is demoing our work. Humbled that our "Securing Smart Contract Languages" paper won the ACM SIGSOFT Distinguished Paper Award at AIWare! 🏆

🔗 2025.aiwareconf.org/track/aiware...

Building serious AI with zero cloud overhead? It’s here.

This post covers running LLM fine-tuning, streaming chatbots, and even quantum hybrid models—using free Colab GPUs, orchestrated from VS Code.

For #ColabSprint.
rkrants.blogspot.com/2025/11/cola...

The Google Colab VS Code Extension: Beyond Notebooks — Building Production AI Systems with Free… How I’m using Google’s latest extension to prototype LLM applications, fine-tune models, and deploy AI agents — all from VS Code

Notebooks are great for demos—but what about real products?

This post breaks down how to use the new Google Colab VS Code extension to ship production AI systems while keeping the ease of Colab and the power of VS Code.

👇
medium.com/@rabimba/the...

@developers.google.com

Showcasing our work at BRAINS 2025 this weekend!
“Unlocking On-Chain Intelligence (OGAIS)” lets smart contracts trigger deterministic GenAI on-chain with sub-second latency. DeFi prototypes show 15–40% gains (sim). Presented by Yang Lu.
More: brains.dnac.org/2025/
#OnChainAI #BRAINS2025

Big news: I defended my thesis this summer. The final version is public now: uh-ir.tdl.org/items/0625c9...

Grateful to everyone who helped along the way. #PhDone

Apple's new Siri will secretly use Google Gemini models behind the scenes - 9to5Mac Via Mark Gurman, Apple has landed on its strategy for the new Siri update coming as soon as iOS 26.4...

So much for #Apple Intelligence :D

9to5mac.com/2025/11/03/a...

Perplexity is using stealth, undeclared crawlers to evade website no-crawl directives Perplexity is repeatedly modifying their user agent and changing IPs and ASNs to hide their crawling activity, in direct conflict with explicit no-crawl preferences expressed by websites.

Not preplexity

blog.cloudflare.com/perplexity-i...

Perplexity AI Is Lying about Their User Agent • Robb Knight Perplexity AI claims it sends a user agent and respects robots.txt but it absolutely does not

Apart from perplexity, others seem to honor it (or used to).
Perplexity on the other side


rknight.me/blog/perplex...

www.malwarebytes.com/blog/news/20...

When this picture was taken at Google stage, I was trying to make a point about Open Weight AI models, while....wearing a @mozilla@apoil.org T-shirt with a @Chrome@mastodon.world hoodie....

www.uh.edu/nsm/news-eve...

And yes. I did talk about it again youtu.be/cHrWxuHOcyg

YouTube video by Rabimba Karanjai Google I/O 2025: GDE Summit, From Whiteboard to Users

Something I talked about in Google I/O, GDE Summit :)

youtu.be/cHrWxuHOcyg

Interested to know more about #AI Red Teaming and how you can protect your AI Infra and Models?

I'll be speaking at @kcdc.bsky.social tomorrow on exactly that with some "questionable" demos *wink*

Drop by if that's of interest to you

#kcdc2025

The only way you can stop @Apple and Siri to not send your data, in plainly , not being part of private compute. And by data, your precise location Everytime you use your iPhone, any app you use, you do inside them including your message, documents, communication and text.

Disable fictation

Wow.
Apple sends all your WhatsApp messages or any other message with contact numbers to their server when you use Siri!

So much for respecting E2E even in Whatsapp

And also @apple scans all your apps including whats inside VM and your saved documents (inside notion or whatever) and sends it to Apple Server......

So much for privacy

So apple sends you precise location to its server no matter what you ask Siri.
Even if it's "what is 1+1"

Will be presenting "you know what" with "you know who" at "you know where" tomorrow

Happy to chat if you are around.
#strawhatpirates

The white hat among #BlackHat2025

Our paper, "Collaboration is all you need: LLM Assisted Safe Code Translation," is now published in at #FSE25).

We propose a method for using #LLMs to perform safe #CodeTranslation, aiming to prevent security and stability issues in migrated code.

dl.acm.org/doi/10.1145/...

Excited to have spoken at #Google I/O 2025 GDE Summit on making my research—AI, code generation & smart contracts with Move—accessible to real users! Grateful to @suifoundation.bsky.social & #GDE program for backing this open research. See the recording: youtu.be/cHrWxuHOcyg

🎤 SpeakWise: Build an AI Public Speaking Coach with Gemma 3n A blog about my journey through Mozilla,Open Source Contribution, Virtual,Augmented,Mixed Reality, Security,Hackathon and in general technology.

🚀 Launched SpeakWise, an AI-powered public speaking coach using Google’s Gemma 3n.

🎙️ Transcribes your talk
👁️ Analyzes body language & slides
✅ Gives actionable, privacy-first feedback

Multimodal, on-device, in real time.

Try: rabimba.me/link/speakwise

#AISprint

YouTube video by Rabimba Karanjai AI, LLMs, and the Future of the Web: A Conversation with Rabimba and Hemanth (PayPal) at Google I/O

Thrilled to share my recent conversation with Hemanth from PayPal, straight from the Google Studio at I/O 2025. We explored the impact of LLMs on human intuition, the security challenge of "model poisoning," and the evolving landscape of web development with integrated AI.

youtu.be/tErgHp_qtOk

Collaboration is all you need: LLM Assisted Safe Code Translation (FSE 2025 - Ideas, Visions and Reflections) - FSE 2025 The ACM International Conference on the Foundations of Software Engineering (FSE) is an internationally renowned forum for researchers, practitioners, and educators to present and discuss the most rec...

🚀 Our paper "Collaboration is all you need: LLM Assisted Safe Code Translation" is being presented tomorrow at #FSE2025!

We introduce UniTranslator, a multi-agent system of compact LLMs working collaboratively for safer, more accurate code translation.

rabimba.me/link/fse25

#AI #LLMs

From Models to Agents: Shipping Enterprise AI Faster with Google’s MCP Toolbox & Agent Development Kit Build production AI agents fast with Google MCP Toolbox & ADK—YAML-config tools, Gemini integration, embedded slides, and live code demo from GDE Rabi

Skip prompt spaghetti: MCP Toolbox exposes safe SQL/vector endpoints; ADK lets any LLM call them with one line. My deep-dive post + deck is live → rabimba.me/link/adkmcp #AISprint

"Apple’s celebrated “Illusion of Thinking” paper on AI is facing serious pushback.

A new response, co-written with Claude Opus, flags three fatal issues: one is counting tasks as fails when the model just runs out of tokens.

The hype was loud; the correction deserves equal attention."

It was 22nd 4:40 PM.
It should have been recorded. I'll send you the link when I have it

But also can share the slides :D