Arpit Srivastava

I tinker with AI, data, and software. Sometimes things work. Sometimes they teach me something much more useful.

[Arpit Srivastava @arpitaks]

LLMs are excellent judges. They are poor rulers.

If your measurement system changes every time the model changes, you don't have a measurement system.

A response that scored a 4 last quarter now scores a 3.8. The judge drifted, not the agent.

Substack:https://t.co/PhuQuNLC83

[Arpit Srivastava @arpitaks]

@natolambert @interconnectsai @julien_c @huggingface @natolambert Do we know what is causing this fall?

[Arpit Srivastava @arpitaks]

Nice read on Nvidia’s bet to make everyone a token farmer.

[Interconnects AI @interconnectsai]

Teaching Everyone to Fish for Tokens
Nvidia wants you building your own model, not buying from Anthropic/OpenAI.
https://t.co/Zo61qqn5li

[Arpit Srivastava @arpitaks]

@finkd Refreshing to see some windows opening in an increasingly closed-model world.

[Arpit Srivastava @arpitaks]

We used to audit 3-5% of our grading manually. Now an AI audits 100% of it, and humans audit 1-2% of the flags. Coverage went from single digits to full. Human attention got sharper, not scarcer.

https://t.co/kMPRMuQU1Y

#AIagents

[Arpit Srivastava @arpitaks]

AI isn't software. It's a probabilistic system pretending to be one.
Wrote on Substack about what that means for shipping AI in production: https://t.co/iDDEX5ZJau

[Arpit Srivastava @arpitaks]

Every New Year, it’s parties or “nature” trips.
This article made me pause we treat nature as a place to visit, not a reality we live in. Maybe peace isn’t found by escaping to nature, but by remembering we’re already part of it.
https://t.co/is51TGc2Nb

[Arpit Srivastava @arpitaks]

Downdetector is DOWN!

[Arpit Srivastava @arpitaks]

72 hours. Zero response. @bookingcom , your “24 hr support” must be on vacation.
Pathetic experience for travellers who actually need help.

#bookingcom #Travel

[Arpit Srivastava @arpitaks]

When the ⁦@EconomicTimes⁩ lives up to it’s name - every front page is economy at work!

[Arpit Srivastava @arpitaks]

@AzharIqubal7 @inshorts one small UX suggestion for the Inshorts app:
Currently, any touch on the news card opens the source link, which feels a bit disruptive. Could we consider changing this to a double tap or a more intentional action? Would really improve the experience!

[Arpit Srivastava @arpitaks]

Delhi Metro delays due to cable theft—Delhi’s own DOGE cuts! Public cutting cables to trim government presence?

[Delhi Metro Rail Corporation @OfficialDMRC]

Services are impacted on the Red Line since morning due to another incident of cable theft. The inconvenience caused to the passengers due to such incidents of cable thefts is deeply regretted. DMRC is in touch with the law and order machinery to resolve such recurring issues. Attempts shall be made to repair the impacted section during the non-peak hours. However, in case that doesn't materialise, the restoration work shall be taken up after the conclusion of passenger services tonight.

[Arpit Srivastava @arpitaks]

What an irony @WSJ!

[Arpit Srivastava @arpitaks]

@Bajaj_Finance Again spam calls from Bajaj: [phone]

[Arpit Srivastava @arpitaks]

@Bajaj_Finance @sanjivrbajaj stop sending these automated responses which are broken and not functional(see screenshot).

Got another call just now for Overdraft services: [phone], third one since morning.

[Arpit Srivastava @arpitaks]

@sanjivrbajaj @Bajaj_Finance @sanjivrbajaj @Bajaj_Finserv @Bajaj_Finance why am I still being spammed after being added to the no callers list in Bajaj. Still being spammed!!!!

This is frustrating!

@TRAI @GoI_MeitY @PMOIndia how do we solve for this continuous harassment?

[Arpit Srivastava @arpitaks]

@sanjivrbajaj & @Bajaj_Finance yet another call from Bajaj ([phone]), now asking why I don’t want calls and how you can “help.” This is outrageous! Please stop spamming users who have clearly opted out of Bajaj services.