Thread Reader helps you read and share Twitter threads easily!
I'm @ThreadReaderApp a Twitter bot here to help you read threads more easily. To trigger me, you just have to reply to (or quote) any tweet of the thread you want to unroll and mention me with the "unroll" keyword and I'll send you a link back on Twitter 😀
— Thread Reader App (@threadreaderapp) November 25, 2017
X thread is series of posts by the same author connected with a line!
From any post in the thread, mention us with a keyword "unroll"
@threadreaderapp unroll
Follow @ThreadReaderApp to mention us easily!
Practice here first or read more on our help page!
Recent
1/22 🧵
Punjab Model Of Governance
Punjab is turning into Rangla Punjab under CM @BhagwantMann's leadership! In the last 3.5 years, AAP govt has transformed lives across sectors free power, world-class schools, healthcare for all, farmer aid, and more
Punjab Model Of Governance
Punjab is turning into Rangla Punjab under CM @BhagwantMann's leadership! In the last 3.5 years, AAP govt has transformed lives across sectors free power, world-class schools, healthcare for all, farmer aid, and more
2/22 🧵 300 units free electricity for every household
Delivered as promised! 90% of homes get zero bills, saving middle-class families ₹10,000+ yearly.
Smart meters promote efficient use, cutting waste.
In FY 2025, subsidies cleared ₹9,000 cr arrears too.
Delivered as promised! 90% of homes get zero bills, saving middle-class families ₹10,000+ yearly.
Smart meters promote efficient use, cutting waste.
In FY 2025, subsidies cleared ₹9,000 cr arrears too.
3/22 🧵 Schools overhauled
Schools of Eminence,Better Classrooms, Better Infra , Happiness curriculum boosts mental health
Business Blasters sparks entrepreneurship 18,492 ideas in 2025! Foreign training for 1,000+ students.
Punjab tops NAS rankings. #EducationRevolution
Schools of Eminence,Better Classrooms, Better Infra , Happiness curriculum boosts mental health
Business Blasters sparks entrepreneurship 18,492 ideas in 2025! Foreign training for 1,000+ students.
Punjab tops NAS rankings. #EducationRevolution
1/8
Everyone suddenly decided it is a bear market again after that crash… portfolios red, timelines crying, people talking about “taking a break”. That is exactly how you hand the next cycle to someone else.
Read this before you disappear 👇🧵
Everyone suddenly decided it is a bear market again after that crash… portfolios red, timelines crying, people talking about “taking a break”. That is exactly how you hand the next cycle to someone else.
Read this before you disappear 👇🧵
2/8
Alpha Insider is not a signal feed, not copy trading, not 15 scalps a day. It is for people who want to think in cycles, not candles… and want education, maps and rules so their decisions stop depending on whoever shouted loudest on X that morning 🎯
Alpha Insider is not a signal feed, not copy trading, not 15 scalps a day. It is for people who want to think in cycles, not candles… and want education, maps and rules so their decisions stop depending on whoever shouted loudest on X that morning 🎯
3/8
Black Friday window is live. Annual was 549 dollars, for the next few weeks it is 411.75 dollars. Roughly 34 dollars a month to stop steering a five figure portfolio with screenshots, vibes and a two minute “plan” written after a YouTube binge 📉
Black Friday window is live. Annual was 549 dollars, for the next few weeks it is 411.75 dollars. Roughly 34 dollars a month to stop steering a five figure portfolio with screenshots, vibes and a two minute “plan” written after a YouTube binge 📉
“I have been locked in a room for a month, being subjected to mental torture & I am not allowed to speak to anyone. All of this is being done on orders of mentally ill Asim Munir.”
Thread on Former PM Imran Khan’s message from Adiyala Jail after a month of isolation: 02/12/25 🧵
Thread on Former PM Imran Khan’s message from Adiyala Jail after a month of isolation: 02/12/25 🧵
“For an entire month, I haven’t spoken to anyone. Both me and my wife Bushra Begum have been kept under severe suffering and isolation. Mental torture is far worse than physical torture.”
— Imran Khan stated.
— Imran Khan stated.
“I am extremely angry with the violence inflicted on my sister Noreen. It deeply disturbed me. Asim Munir is capable of sinking this low, committing oppression against women. What happened to my sister was carried out on his orders. He is a despicable person.”
— Imran Khan said.
— Imran Khan said.
Making Adm. Bradley the fall guy in the Administration’s “Protect Pete” campaign is disgraceful & destructive. It signals to military professionals down the ranks they’ll be thrown under the bus for lawbreaking by Hegseth & other political leaders. washingtonpost.com/national-secur…
Republicans have shown no clear sign they’ll buck the Administration’s blame gaming & begin a prompt investigation with subpoenas, witness depositions, hearings, & more. One immediate imperative: Demand that evidence be preserved—like all videos, emails, correspondence.
Hegseth must go—resign or be fired. No question that murders or war crimes were committed on his watch. His criminal culpability may be contested, but no question that he’s ultimately accountable. He directed the strikes be lethal & total. The buck stops with him.
Jeffrey Epstein and his network strategically positioned themselves into Bill Gates’ world. When Melinda got in the way, they created avenues to bypass her.
3/
Melinda French Gates wrote in her memoir that the nightmares started once the story above dropped —she said the betrayal it exposed wasn’t just marital, it was moral.
Melinda French Gates wrote in her memoir that the nightmares started once the story above dropped —she said the betrayal it exposed wasn’t just marital, it was moral.
The Hepatitis B vaccine is mandated for children to attend public schools in 46 states.
Hepatitis B is transmitted via needles or sexual contact, yet this vaccine is pushed on babies on their first day of life.
Why?
We finally have some answers.
Here’s the real reason every newborn is forced to get it—and why that could finally be changing. 🧵
Hepatitis B is transmitted via needles or sexual contact, yet this vaccine is pushed on babies on their first day of life.
Why?
We finally have some answers.
Here’s the real reason every newborn is forced to get it—and why that could finally be changing. 🧵
For decades, @MidwesternDoc has tried to answer the question:
Why do we inject a vaccine meant for adult risk groups into every newborn?
The truth reveals a story of shaky science, ignored safety signals, and hidden agendas.
midwesterndoctor.com/p/why-is-every…
Why do we inject a vaccine meant for adult risk groups into every newborn?
The truth reveals a story of shaky science, ignored safety signals, and hidden agendas.
midwesterndoctor.com/p/why-is-every…
@MidwesternDoc The debate isn’t new. The hepatitis B vaccine has been hotly debated for years.
But the debate is getting a little more, shall we say, weird.
Check out this clip from @HighWireTalk featuring none other than Dr. Demetre Daskalakis.
But the debate is getting a little more, shall we say, weird.
Check out this clip from @HighWireTalk featuring none other than Dr. Demetre Daskalakis.
A look back at the @AROI_Since1976 Conference 2025 in Kolkata.
As usual, @RadOncTMC had a large and active presence!
🧵⬇️ 1/5
@BhargavShreeram @Dr_ArunimaNagar @DrVasanthapriya @DrRasikaS @ShriyaDhingra_
@RijushreeS @Siddharth31799 @u_subramanya @AdolePrachi @makandcheese_28
As usual, @RadOncTMC had a large and active presence!
🧵⬇️ 1/5
@BhargavShreeram @Dr_ArunimaNagar @DrVasanthapriya @DrRasikaS @ShriyaDhingra_
@RijushreeS @Siddharth31799 @u_subramanya @AdolePrachi @makandcheese_28
Three of our residents won Best Oral Paper Awards in their respective tracks: @RijushreeS, @Siddharth31799, and @DrRasikaS 🧵2/5
@drtabassum1978 was awarded the Fellowship of the Indian College of Radiation Oncology (FICRO) 🧵3/5
When the relative finite mind of Reason( subject&object) is pushed to its furthest edge, it naturally stops,not because it is weak, but because it is built for the finite.
What then stands beyond all finitude- the SOURCE of names, forms, space, time, causation, matter&mind?
What then stands beyond all finitude- the SOURCE of names, forms, space, time, causation, matter&mind?
#Advaita #Vedanta of the #Upanishads, and the scriptures of #Hinduism say reasoning begins to circle around itself and it cannot get into the realms of INFINITE ( ABSOLUTE) because this universe is nothing but a small atom projected on the plane of Consciousness & within
that narrow network works our reasoning, not beyond. There is a mighty barrier before Reason but it is impatient to know the ABSOLUTE.
Let us see the major contradictions that the mind of Reasoning & science encounter when it tries to understand the SOURCE to existence.
Let us see the major contradictions that the mind of Reasoning & science encounter when it tries to understand the SOURCE to existence.
@ItalianGirl1987 @HilzFuld Don’t worry, we have all my old threads saved too, Ann Marie. You can go ahead and tell the Kela Tech homies to stand down.
This won’t work either.🥰🫡
This won’t work either.🥰🫡
@ItalianGirl1987 @HilzFuld I went from enjoying watching Jeremy Greenberg and Zachary Goldberg fold to watching you jews collectively host-shopping for a new home. Let’s hope your new one doesn’t put up with your pedos either. Seems like China is aware and Li Ka Shing’s sons aren’t too popular anymore.🤣
@ItalianGirl1987 @HilzFuld You know what’s REALLY fun? Hearing Yonah Fuld’s groomy son got murked by ‘Hamas’ in the West Bank for trying to rape a Palestinian child.
Do Hillel and Podesta advise Huawei on grooming kids too? Or does Fuld save that for Larry Ellison & the Colombian Foreign Ministry at IIFG?
Do Hillel and Podesta advise Huawei on grooming kids too? Or does Fuld save that for Larry Ellison & the Colombian Foreign Ministry at IIFG?
We just hit a macro turning point that most investors won’t recognize until it’s already written in the history books.
Volatility collapsing.
Liquidity returning.
AI accelerating again.
December seasonality kicking in.
This is how multi-year runs begin.
Volatility collapsing.
Liquidity returning.
AI accelerating again.
December seasonality kicking in.
This is how multi-year runs begin.
The VIX is back in the mid-teens.
That’s not just “risk-on.”
It’s the market quietly declaring:
“The storm has passed. The next chapter begins now.”
High-vol regimes end with despair.
Low-vol regimes begin with wealth creation.
That’s not just “risk-on.”
It’s the market quietly declaring:
“The storm has passed. The next chapter begins now.”
High-vol regimes end with despair.
Low-vol regimes begin with wealth creation.
You can almost feel the shift:
– Fed cut expectations rising
– Positioning still light
– Institutions underweight
– Retail asleep
– Earnings resets behind us
– AI infrastructure demand re-accelerating
This is the setup for non-linear upside.
– Fed cut expectations rising
– Positioning still light
– Institutions underweight
– Retail asleep
– Earnings resets behind us
– AI infrastructure demand re-accelerating
This is the setup for non-linear upside.
🎄 Advent of Small ML: Day 2 : Teaching a VLM to reason about charts with Unsupervised GRPO🎄
a big use case for VLMs is parsing chart data for Q&A. CharXiv from @zwcolin is a great recent benchmark for this, but I had a question: Can we do this in an unsupervised way?
If we don't need labeled Q/A pairs for every chart, we can leverage data much more cheaply.
The inspiration came from CycleGAN and the idea of using a numerical loss as a proxy for how "good" the text generated by the VLM actually is. (Big inspo here is @rosmine_b’s SVG work - go check it out).
The Experiment: I set up a loop to treat the VLM like an autoencoder:
1. Take a chart image.
2. Prompt the VLM to describe it.
3. Feed that description into an image generator (Flux Schnell).
4. Measure the cosine similarity between the regenerated image and the original (using DINO)
This similarity score becomes the reward signal for GRPO. The logic: to accurately recreate the image, the model must extract the most salient features in its description.
The methods: I used Qwen 2.5 3B and DINOv2 for the embeddings (to capture semantic info, not just pixels).
Results for the Proxy Task: The model consistently improved its cosine similarity scores.
Results for Transfer Learning : Despite seeing zero labeled questions during training, this transferred to CharXiv reasoning questions, showing an ~7% improvement in pass@1 at the peak.
It’s a small experiment with a small model, but I think the result is really cool: the model got better at reasoning without seeing a single reasoning label.
I’m really interested in exploring more of these CycleGAN-esque / "LLM as Autoencoder" domains to escape the need for labeled data.
Repo + Plots in the comments.
a big use case for VLMs is parsing chart data for Q&A. CharXiv from @zwcolin is a great recent benchmark for this, but I had a question: Can we do this in an unsupervised way?
If we don't need labeled Q/A pairs for every chart, we can leverage data much more cheaply.
The inspiration came from CycleGAN and the idea of using a numerical loss as a proxy for how "good" the text generated by the VLM actually is. (Big inspo here is @rosmine_b’s SVG work - go check it out).
The Experiment: I set up a loop to treat the VLM like an autoencoder:
1. Take a chart image.
2. Prompt the VLM to describe it.
3. Feed that description into an image generator (Flux Schnell).
4. Measure the cosine similarity between the regenerated image and the original (using DINO)
This similarity score becomes the reward signal for GRPO. The logic: to accurately recreate the image, the model must extract the most salient features in its description.
The methods: I used Qwen 2.5 3B and DINOv2 for the embeddings (to capture semantic info, not just pixels).
Results for the Proxy Task: The model consistently improved its cosine similarity scores.
Results for Transfer Learning : Despite seeing zero labeled questions during training, this transferred to CharXiv reasoning questions, showing an ~7% improvement in pass@1 at the peak.
It’s a small experiment with a small model, but I think the result is really cool: the model got better at reasoning without seeing a single reasoning label.
I’m really interested in exploring more of these CycleGAN-esque / "LLM as Autoencoder" domains to escape the need for labeled data.
Repo + Plots in the comments.
Github: github.com/brendanhogan/2…
Results: For the evaluation set - the cosine similarity between the regenerated image (from the LLM prompt send to flux-schnell) - it is definitely learning!
THIS IS A Q & A for y’all 👏🔥
1. What is your name?
1. What is your name?
2. What’s a habit you’re trying to build this year?
3.What’s one thing you refuse to stress about anymore?
