Abhivardhan

79.8K posts

Abhivardhan banner
Abhivardhan

Abhivardhan

@IndusThink

AI Governance @IndicPacific. President @IndianSocietyAI. Posts personal, unrelated to organisations I am affiliated with. My #AI book: https://t.co/IdknPNfVVL

Indo-Pacific ⇄ Europe ⇄ World Katılım Haziran 2015
1.5K Takip Edilen3K Takipçiler
Sabitlenmiş Tweet
Abhivardhan
Abhivardhan@IndusThink·
Glad that @NDTVProfit invited me and asked questions on what AI governance and public-private partnership trends one can fathom from the Pentagon vs Anthropic situation. Here’s my complete take on this. While the Trump admin cannot be relied upon to create policy precedents - the incident created a policy precedent: LLMs remain unreliable.
Abhivardhan@IndusThink

Well, thanks @NDTVProfit. :)

English
0
1
5
3K
Abhivardhan retweetledi
NEET INTEL
NEET INTEL@neetintel·
A post "decoding" X's new algorithm has gone viral. It tells you what's dead, what wins, and to screenshot it. X open-sourced the entire algorithm on GitHub, so I downloaded it and checked the claims against the real code. Most of it doesn't hold up. What the post got WRONG: → "Small accounts get a 3x boost from out-of-network reach." It's the opposite. One part of the code (a file called oon_scorer) exists purely to turn DOWN posts from people you don't follow. Its own comment says "prioritize in-network." The thread printed the algorithm backwards. → "Media gets 2x the weight." There's no 2x. The code just records whether a post has an image. It's a plain yes/no without any multiplier attached. → "Posting 4+ times a day triggers a penalty." There's a real rule that stops one person flooding your feed. But here's the deal: it only spaces out how often you show up in a single scroll. There's no daily count, and no number 4. That was invented. → "Closers like 'what do you think?' get you flagged." There is no engagement-bait detector anywhere in the code. → "Long 4,000-character posts get boosted." I searched the whole codebase for "4000." Nothing. What it got RIGHT (one thing): → Replies really are judged by WHO replies, not just how many. The code has a setting for whether a large account joined your thread. Credit where due. The irony? The repo ships a file that scores post quality. One thing it measures is literally called a "slop score" — X built a tool to detect low-effort filler. A recycled "what's dead / what wins" thread is exactly that. The takeaway? X's algorithm is public. Anyone can open it, but almost nobody does. Instead, they reshare a thread that summarized a blog that paraphrased a tweet. When a post hits you with confident numbers, ask the one question that matters: did they actually open the file?
NEET INTEL tweet media
Ronin@DeRonin_

BREAKING: X dropped the new ranking algo today. What's dead and what wins for bangers: [ what just died ]: - spam posting (4+ posts/day = author dilution penalty) - reply farming (replies weighted by WHO replies, not how many) - text-only posts (media gets 2x signal weight now) - recycled viral templates (the new content classifier flags them) - generic AI tool roundups (no original POV = low signal) - engagement bait closers ("what do you think?" gets flagged) - motivational fluff without specifics - engagement pods (mutual-follow scores reweighted) [ what just won ]: - original takes from small accounts (out-of-network discovery 3x'd) - threads with clear narrative arcs (model reads full thread context now) - text + media combos (out-performs either alone) - contrarian takes with personal proof (numbers, screenshots, names) - long-form posts (the 4000-char format gets heavier signal weight) - responding to your own replies in the first 30 min (ranking gold) - consistent posting cadence (regularity > volume) [ formats that work right now ]: - tactical playbooks: hook + 5-8 numbered steps + closer - personal proof posts: "$X → $Y in Z weeks" + breakdown - contrarian takes backed by screenshot proof - threads with arc structure: setup → friction → resolution - image carousels (3-7 slides), one bold claim per slide - short videos (under 90s) showing real work, not promo - long-form (4000-char) breakdowns of trending topics [ tone of voice that wins ]: - first-person specific ("I built X / I shipped Y") over abstract observation - concrete numbers and names over vague claims - builder energy over motivational fluff - one strong opinion per post, not three hedged ones - "here's what I shipped" over "here's what's possible" - direct address to the reader ("you") over generic third person [ the play starting today ]: 1. cut your posting to 2 a day max 2. always pair text with image, video, or thread 3. reply to every comment in the first 30 min 4. write in first-person specific, not third-person abstract 5. one bold opinion per post, backed by proof 6. let out-of-network discovery do the heavy lifting screenshot this. algorithm cycles repeat every 6 months

English
6
28
133
9.1K
Abhivardhan retweetledi
कुशल मेहरा
A news channel has editors and they still did what this channel did. Imagine what YouTubers are doing. Imagine the wild west that YouTube is! When I keep telling people to develop intuitive tools to decipher news there is a reason. You are being fed information by illiterate non experts.
English
8
109
606
11.9K
Abhivardhan
Abhivardhan@IndusThink·
If Sanjeev Sanyal becomes WB Fin min, UP should also eventually have a policy czar as an industry minister. 🙂
English
0
0
3
743
Abhivardhan retweetledi
Nirmalya Dutta
Nirmalya Dutta@NonsensicalNemo·
You can ignore any sentence that has the phrase 'sparks debate' in it.
English
0
1
4
170
Parmita Mishra
Parmita Mishra@parmita·
I love this dude so much lmao
English
6
2
70
4K
Ravi Sharma
Ravi Sharma@ravishar313·
Rest assured grass will be touched
Ravi Sharma tweet media
English
1
0
12
275
Abhivardhan retweetledi
Sougat Chakraborty
Sougat Chakraborty@sougat18·
A thread 🧵on the latest paper published by @hjethva05 in the Remote Sensing Journal on Stubble Burning in Northwestern India. cc: @VishnuNDTV
Sougat Chakraborty tweet media
Hiren Jethva@hjethva05

New Publcation Alert: mdpi.com/2072-4292/18/1… Crop fires are kept hidden from NASA-NOAA polar orbiting satellites (MODIS, VIIRS) in northwestern India, but effectively captured in Gostationary time-resolved observations. Fire dataset distributed through zenodo.org/records/200847…

English
1
2
6
542
Abhivardhan retweetledi
Tushar Gupta
Tushar Gupta@Tushar15·
The countries that built lasting fiscal resilience didn't wait for a crisis to justify bold action. They treated sovereign assets like transactions, not schemes. FDI like a deal pipeline, not a policy posture. moneycontrol.com/news/opinion/i… That shift in orientation is available to any economy willing to make it.
English
0
17
44
1.5K