Prateek Singh

Founder @ganakailabs | Making your favourite Cloud ☁️ Visual & Conversational with AI 🤖 | Ex-Staff SWE @LinkedIn 💻 | 5x Author 📚

[Prateek Singh @mentalhotfix]

Your first task:
1. Join Discord server, where can can discuss more.
2. Navigate to https://t.co/6K2QATsyiu .
3. Try one real Azure Cloud or Infrastructure as code workflow.
4. Share what felt useful, confusing, or missing.
5. Tell me what would make this worth paying for.

[Prateek Singh @mentalhotfix]

Cloudeval AI (https://t.co/nBXyM1xNeE) Beta is now LIVE!

CloudEval AI turns your cloud infrastructure into a decision system.

• Visualize cloud architecture instantly.
• Run cost, security, and Well-Architected evaluations.
• Query your infrastructure conversationally.

No more digging through dashboards, diagrams, exports, and cloud portals trying to understand what’s actually happening.

[Prateek Singh @mentalhotfix]

If you manage cloud infrastructure and want to help shape the future of cloud intelligence:

Join the beta here:
https://t.co/fLjE0LNzR6

Docs:
https://t.co/cvzPtzYJsX

Discord:
https://t.co/0Uv28H85c2

Cloud infrastructure keeps getting more complex.
The tools for understanding it haven’t evolved fast enough.

That changes now!

[Prateek Singh @mentalhotfix]

@Google built CodeWiki - https://t.co/nGsIcsCbMN

A system that turns code into structured, queryable knowledge.

Not docs. Not comments.

Actual understanding:
– What depends on what
– How systems evolve
– What breaks if you change something

This is where things shift.

You won’t read codebases anymore.
You’ll query them.

That’s context engineering at scale.

And this is exactly how AI agents will work with code.

Send this to someone still debugging by scrolling files.

#ai #softwareengineering #llm #systemdesign #aiagents

[Prateek Singh @mentalhotfix]

Here are the 4 AI YouTube channels I try to watch every day. They shape how I think, build, and stay ahead in AI.

My personal shortlist 👇

1. AI Engineer - https://t.co/8QrMcEhqti

2. The AI Automators - https://t.co/f853iJvwTq

3. IBM Technology - https://t.co/fn9QdVYk8Y

4. Yannic Kilcher - https://t.co/zHxheD2Jwj

[Prateek Singh @mentalhotfix]

Can AI feel emotions?

Anthropic Research: https://t.co/CpJE3UMv1P

Anthropic thread summary:
https://t.co/UZz9m0c08q

[Anthropic @AnthropicAI]

New Anthropic research: Emotion concepts and their function in a large language model.

All LLMs sometimes act like they have emotions. But why? We found internal representations of emotion concepts that can drive Claude’s behavior, sometimes in surprising ways.

[Prateek Singh @mentalhotfix]

Google DeepMind’s “AI Agent Traps” highlights a new kind of risk.

Not the model. But the environment.

AI agents can be manipulated through:
1. Hidden instructions
2. Prompt injections
3. Memory poisoning

If you're building AI, this changes how you think about security.

Follow @mentalhotfix for more AI and tech content.

#ai #aiagents #genai #machinelearning #cybersecurity tech startups
https://t.co/i3kH74Lph5

[Prateek Singh @mentalhotfix]

@chamath The'll end up using @deepseek_ai

[Prateek Singh @mentalhotfix]

Started writing my 6th book 👇
"The UX Foundations of Agentic AI Systems"

The more I work with AI agents, the more I realize this: AI products don’t succeed or fail on intelligence alone... they succeed or fail on UX.

The real challenge is engineering the interface between probabilistic systems and deterministic human expectations.

https://t.co/8aORI0d8gr

[Prateek Singh @mentalhotfix]

A good UX doesnt just need to be corrwct, but also consistent with reality 😅

[Prateek Singh @mentalhotfix]

Its all me 😅

[Prateek Singh @mentalhotfix]

Original thinking? but AI can do that in 20$ 😅

[Prateek Singh @mentalhotfix]

Why treating prompts as an Environment changes LLM Scaling (@MIT Paper)

MIT proposes Recursive Language Models (RLMs) that flips the usual paradigm:
Instead of stuffing massive prompts into a transformer, they treat the prompt as part of the environment, with with LLMs can interact.

And its outperforming GPT-5 on some tasks already!

An RLM:
→ Loads the prompt as a variable in a Python REPL
→ Uses code to peek, slice, and decompose it
→ Recursively calls itself on only the relevant pieces
→ The model doesn’t remember everything, but it interacts with the prompt symbolically.

Research Paper: https://t.co/Hv9ydteuQL

[Prateek Singh @mentalhotfix]

If you’re on @ChatGPTapp Plus and you’re not using Codex in @code or the CLI, you’re leaving value on the table.

I’ve been testing it from 2 days and the high-reasoning limits are actually pretty decent, and it’s included in your regular ChatGPT subscription.. access in both IDE and CLI integrations is worth trying.

One thing that still feels missing: the ability to attach any existing ChatGPT chat or a Project as context. That would be a game-changer... you could brainstorm in ChatGPT, then hand off the full context to the agentic coding model for much stronger, more consistent work.

[Prateek Singh @mentalhotfix]

Brainstorming with AI on improving diagrams in https://t.co/6K2QATsyiu , and it suggested an open-source tool 'AzViz' I built 6 years ago. Life really does come full circle. 🙂 #buildinpublic

[Prateek Singh @mentalhotfix]

How AI Code Editors Work? #livestream on 7th Oct 11 AM IST 👇

Attend here Online: https://t.co/Tk0CVRRzv0

https://t.co/DSeqr3JNIY

[Prateek Singh @mentalhotfix]

How AI Code Editors Work? #livestream on 7th Oct 11 AM IST 👇

Attend here Online: https://t.co/Tk0CVRRzv0

https://t.co/DSeqr3JNIY

[Prateek Singh @mentalhotfix]

Weekend AI Rundown #LIVESTREAM - @Microsoft AI Models, @Google's EmbeddingGemma & New research from @OpenAI Why LLMs Hallucinate?

Follow @ganakailabs for more, every week!

https://t.co/Krv0uTi8l7

[Prateek Singh @mentalhotfix]

Google’s Gemini 2.5 Image Model (aka 'Nano-Banana') is the new state-of-the-art model 🔥 👇

https://t.co/ftiVMG4mN6