Aayush Singh Chauhan | Data & AI

BI, Data Science & AI Platform Engineering | IIT Roorkee '20 | Guitar & Singing on the side

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

https://t.co/Fd5pJn7R9O

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

Indian stocks fell for an eighth week in a row. That is the longest stretch in 25 years. On 1 Oct, foreign investors sold ₹9,484 crore of shares. Indian institutions bought ₹10,042 crore. Local buying matched the foreign selling. Matching it has not turned the fall around.

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

@devops_nk You’re allowed to sweat the MVP deadline, but that’s no excuse to stop optimizing for an edge case even v10 will never face. Deadlines are temporary. An abstract base Class is permanent.

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

@MySoctr Three losing sessions was the morning note. The close was a fourth down day, inside an eighth down week. FIIs sold. DIIs bought more. Absorption, not a bottom.

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

@Financewitpalak The list is the backdrop. The 1 Oct tape is simpler. FIIs sold ₹9,484 crore. DIIs bought ₹10,042 crore. Eighth down week. The buyers are local. The direction has not flipped.

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

@i_m_garg Crude and the US 10-year explain the pressure. They do not explain the flow. FIIs sold ₹9,484 crore on 1 Oct. DIIs bought ₹10,042 crore. Local money is the bid. The trend is still down.

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

@AVallimohd Don't call Monday a crash off this tape. India just had an eighth down week, and on 1 Oct DIIs bought more than FIIs sold. That is absorption. It is not a turn.

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

@TheTradingwolf0 FII selling is real. On 1 Oct they sold ₹9,484 crore. DIIs bought ₹10,042 crore. The week is still the eighth down week in a row. Domestic buying is absorbing the sell. It has not turned the trend.

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

https://t.co/Fd5pJn7R9O

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

5/n
OpenAI says it closed the replay path. It added checks on streamed output that might leak reasoning. It banned accounts. It shared the pattern with industry partners.

Expect more of this. As frontier models get better, copying the thinking becomes more valuable than copying the polished reply.

Primary source:
https://t.co/EgIHnzEinv

#AI #OpenAI #MoonshotAI #AdversarialDistillation

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

OpenAI shut down a July campaign that tried to pull "protected reasoning" out of its models.

That is the hidden scratchpad. The steps the model takes before it shows you a clean answer.

They say a core cluster of the activity came from people linked to Moonshot (Kimi). On the peak days, about 16,000 extraction-pattern requests came from more than 4,000 accounts.

This was not a break of encryption. Not a database hack. Not reading stored chats. They replayed encrypted reasoning across conversations until the model wrote those steps in plain text.

I have spent years around data systems and AI tooling. The final answer is only half the story. If you steal the working notes, you can copy capability without copying the safety stack.

That is the real fight now. Not just who has the biggest model. Who can stop someone else from copying the thinking.

https://t.co/EgIHnzEinv

#AI #OpenAI #MoonshotAI

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

4/n
Why this matters, from years of building around data and AI systems:

The final answer is curated. The scratchpad is closer to how the model actually works. Harvest that, and you can push capability into another model without carrying the same safety filters on the user-facing side.

That is not a cute research trick. It is a shortcut with safety and security consequences.

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

@zomatocare Can you promise a resolution time for such a small issue or not?

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

@zomato messed up my order and the app bot just tells me to email them. I did, got Ticket 220728606, and have been ghosted for 5 days. Stop making customers jump through hoops to get a refund for YOUR mistakes. Bring back in-app support. Order 8633757469 @zomatocare

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

@zomato @zomatocare @deepigoyal This feels like a deliberate tactic. Zomato knows people will give up if they have to draft emails and chase them for small refunds. If you can take my money in two seconds on the app, you can fix a wrong order on the app. Wake up. (Order ID: 8633757469)

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

Has @zomato completely given up on customer support? I got the wrong order and their app bot literally just tells you to write an email. Why do I have to do extra work because your restaurant messed up? @zomatocare @deepigoyal #Zomato 🧵👇

[Aayush Singh Chauhan | Data & AI @256BitEncrypted]

@zomato @zomatocare @deepigoyal I actually went ahead and emailed them. Got an automated reply saying wait 24 to 48 hours (Ticket ID: 220728606). It has been 5 days and total silence. You make reporting issues a nightmare and then ghost your customers anyway.