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
When a DOJ/FBI/IC document starts a sentence with "Moreover," pay attention to what comes next.
"This CIOL, believed to have been missing for several years, was dated September 07, 2016 and contained certain intelligence related to the 2016 U.S. presidential election campaign."
"This CIOL, believed to have been missing for several years, was dated September 07, 2016 and contained certain intelligence related to the 2016 U.S. presidential election campaign."
"The CIOL was found in a storage closet adjacent to the Director’s office and was subsequently transported to the 9582 SCIF. Former Director Comey previously testified before the Senate Judiciary Committee that he was unfamiliar with this CIOL as well as its related intelligence."
I swear, the SEC and its defenders would claim the conference is better than the NFL.
Hypothetical scenario: Alabama is competing with the LA Rams for a playoff spot, SEC sycophants are saying,“Bama beat two SEC ranked teams on the road” & the Rams are saying, “but we beat all these NFL teams” and Paul Finnebaum is arguing the Rams didn’t play in Athens.
(No, I don’t know why the Rams and Alabama woukd be vying for a playoff spot, that’s why it’s a hypothetical!)
Shareholder yield > dividend yield
Most of us know dividend yield.
Few of us know shareholder yield.
Let's learn more.
Most of us know dividend yield.
Few of us know shareholder yield.
Let's learn more.
Shareholder yield is a financial metric measuring the amount of capital returned to shareholders through various means, including dividends, share buybacks, and debt reduction.
It's a comprehensive way of looking at how much cash the company is returning to its shareholders.
It's a comprehensive way of looking at how much cash the company is returning to its shareholders.
The components of shareholder yield typically include and can be found on the cash flow statement, under the financing section:
1. Dividends:
This is the portion of a company's profits that is distributed to its shareholders on a regular basis.
1. Dividends:
This is the portion of a company's profits that is distributed to its shareholders on a regular basis.
Automate your trading strategies in Python
How to build your first trading bot:
(a thread)
How to build your first trading bot:
(a thread)
1. What is a trading bot?
A trading bot is a software program that automates buying and selling financial assets like stocks and cryptocurrencies based on pre-defined strategies and rules.
These automated systems can manage portfolios without human intervention, operating 24/7.
A trading bot is a software program that automates buying and selling financial assets like stocks and cryptocurrencies based on pre-defined strategies and rules.
These automated systems can manage portfolios without human intervention, operating 24/7.
2. Let's make a Bitcoin Trading bot
We'll use investing-algorithm-framework in Python
We'll use investing-algorithm-framework in Python
Because:
(1) it's actually ~57% today (2025)
(2) it was ~81% in 2011
(3) at the current trajectory it is headed to sub-20% over the next decade
China has found a way to grow its economy without incremental emissions, which have plateaued.
This may not be especially titilating for Robin, but I think it is exciting for policymakers in many countries, especially developing ones for whom lack of access to fossil fuels has historically been a growth constraint.
(1) it's actually ~57% today (2025)
(2) it was ~81% in 2011
(3) at the current trajectory it is headed to sub-20% over the next decade
China has found a way to grow its economy without incremental emissions, which have plateaued.
This may not be especially titilating for Robin, but I think it is exciting for policymakers in many countries, especially developing ones for whom lack of access to fossil fuels has historically been a growth constraint.
(1) Robin's chart doesn't have a source, and I don't know how he's calculating 65%.
What I do know is that China's power grid will generate ~10.6 PWh in 2025 and solar/wind/hydro/nuclear will account for ~4.6 PWh. This is up from 10.1 PWh and 3.4 PWh, respectively, in 2024.
What I do know is that China's power grid will generate ~10.6 PWh in 2025 and solar/wind/hydro/nuclear will account for ~4.6 PWh. This is up from 10.1 PWh and 3.4 PWh, respectively, in 2024.
My suspicion here, given that this is a monthly series, is that Robin is drawing from the NBS monthly data series which he may not know excludes small-scale plants.
These figures are adjusted for in the annual numbers.
These figures are adjusted for in the annual numbers.
🎄 Advent of Small ML: Day 7 🎄 Topic: Entropy-Based Rewards (Forcing the model to "keep its options open")
there’s a fascinating recent paper (Layer by Layer: Uncovering Hidden Representations in Language Models - arxiv.org/abs/2502.02013 - shown to me by @aditjain1980) showing that reasoning models tend to have higher entropy in their middle layers
basically, instead of collapsing to an answer early, they keep more possibilities "alive" in their hidden states while thinking.
it made me think - if high entropy correlates with better reasoning, can we force the model to reason better by explicitly rewarding high entropy?
so I added a Matrix-based Entropy reward (Rényi entropy on eigenvalues) to GRPO training on the MATH500 dataset, rewarding the amount of entropy on the middle 10 layers of qwen 2.5 7b
the initial results were mixed.
when I just rewarded entropy, the model definitely increased its entropy... but it didn't get better at math. It just learned to be "confused" and exploratory without actually converging on answers.
It produced some pretty funny outputs, going on weird tangents and "overthinking" simple problems (examples below)
But then I changed the rewarding rule: Only reward high entropy if the final answer is CORRECT.
this worked (sort of) - it gave a 2.5% performance boost over the baseline.
this is a proof of concept that we can use RL to shape the internal dynamics of how a model thinks, not just its final output tokens.
Repo + Plots below
there’s a fascinating recent paper (Layer by Layer: Uncovering Hidden Representations in Language Models - arxiv.org/abs/2502.02013 - shown to me by @aditjain1980) showing that reasoning models tend to have higher entropy in their middle layers
basically, instead of collapsing to an answer early, they keep more possibilities "alive" in their hidden states while thinking.
it made me think - if high entropy correlates with better reasoning, can we force the model to reason better by explicitly rewarding high entropy?
so I added a Matrix-based Entropy reward (Rényi entropy on eigenvalues) to GRPO training on the MATH500 dataset, rewarding the amount of entropy on the middle 10 layers of qwen 2.5 7b
the initial results were mixed.
when I just rewarded entropy, the model definitely increased its entropy... but it didn't get better at math. It just learned to be "confused" and exploratory without actually converging on answers.
It produced some pretty funny outputs, going on weird tangents and "overthinking" simple problems (examples below)
But then I changed the rewarding rule: Only reward high entropy if the final answer is CORRECT.
this worked (sort of) - it gave a 2.5% performance boost over the baseline.
this is a proof of concept that we can use RL to shape the internal dynamics of how a model thinks, not just its final output tokens.
Repo + Plots below
Entropy results - entropy of 10 layers throughout training
1/
People like this fellow keep circulating graphics claiming “Nature and Science prove a massive Steppe migration into India after the Harappans.” - This is false.
The actual papers say something very different — and none of them show a large-scale foreign influx.
People like this fellow keep circulating graphics claiming “Nature and Science prove a massive Steppe migration into India after the Harappans.” - This is false.
The actual papers say something very different — and none of them show a large-scale foreign influx.
2/
First, the favourite talking point: “10–30% Steppe ancestry in modern Indians.”
This number isn’t a measurement.
It is a model fit, and model fits ≠ real historical migrations.
Even the authors warn that these are “statistical constructs, not literal demographic events.”
First, the favourite talking point: “10–30% Steppe ancestry in modern Indians.”
This number isn’t a measurement.
It is a model fit, and model fits ≠ real historical migrations.
Even the authors warn that these are “statistical constructs, not literal demographic events.”
3/
Next: R1a-Z93 = Aryans coming from the Steppe - Nope
R1a-Z93’s highest diversity is in India, not Russia or Kazakhstan.
Higher diversity = deeper origin.
Some of the South Asian R1a clades are older than their Steppe counterparts.
This is the opposite of a one-way migration
Next: R1a-Z93 = Aryans coming from the Steppe - Nope
R1a-Z93’s highest diversity is in India, not Russia or Kazakhstan.
Higher diversity = deeper origin.
Some of the South Asian R1a clades are older than their Steppe counterparts.
This is the opposite of a one-way migration
🧵Last month, FiLiA welcomed our new Director of Research, @DrLauraFavaro, who will bring together FiLiA’s diverse strands of research to establish a dedicated space for the production of high-quality, feminist-driven research: the FiLiA Research Hub.
#16Days #16DaysOfActivism
#16Days #16DaysOfActivism
The Hub’s first project focuses on Women in the vicinity of Sexual Entertainment Venues (SEVs) in #Cardiff, #Edinburgh & #Manchester, gathering attitudes among Women in the area, alongside interviews with Women who live or work locally & feel affected by the presence of SEVs.
Over the last few weeks, Laura has been in the three cities collecting data on the wide range of ways SEVs can impact the lives of local Women.
The ‘pubic triangle’ is a colloquial name for an area in #Edinburgh known for having several strip clubs in close proximity.
#16Days
The ‘pubic triangle’ is a colloquial name for an area in #Edinburgh known for having several strip clubs in close proximity.
#16Days
OpenAI just released 10 free prompt courses.
Beginner to advanced levels (I included all links):
Use these courses to master ChatGPT prompts.
Beginner to advanced levels (I included all links):
Use these courses to master ChatGPT prompts.
1. Introduction to Prompt Engineering
→ Learn the basics of writing clear, effective prompts from the comfort of your home.
→ Course: lnkd.in/gxm69Kqr
→ Learn the basics of writing clear, effective prompts from the comfort of your home.
→ Course: lnkd.in/gxm69Kqr
2. Advanced Prompt Engineering
→ Learn advanced prompt structures for precision & control.
→ Course: lnkd.in/gYGWjSQ4
→ Learn advanced prompt structures for precision & control.
→ Course: lnkd.in/gYGWjSQ4
@realMaalouf A secular sovereign may, in certain international or multicultural contexts, apply a foreign legal system by analogy when assessing culpability, provided that such application is interpretive rather than authoritative.
1/4
1/4
@realMaalouf In this framework, a Muslim offender may be viewed as a subject culturally and normatively shaped by a foreign legal regime (e.g., sharia). The crime is defined solely by the secular criminal code—such as the violation of Christian sacred symbols under national law —
2/4
2/4
@realMaalouf yet the degree of punishment is calibrated by reference to the sanctions that the offender’s own normative tradition prescribes for an equivalent offense (e.g., Islamic blasphemy law).
3/4
3/4
During this interview, Trevor graciously asked JP from $PEY.to a question about decline rates.
This is an important and telling thread for many reasons, including misinformation on X and the truth about resource plays.
Buckle up 🧵 1/x
This is an important and telling thread for many reasons, including misinformation on X and the truth about resource plays.
Buckle up 🧵 1/x
Here's the question:
“Why is Peyto’s decline rates higher than your peers (if that’s the case)?
Trevor followed, “I don’t think it is”
And JP responds, “I don’t think it is either so I don’t know how to answer that”
In short JP said… What the hell is he talking about?
3/x
“Why is Peyto’s decline rates higher than your peers (if that’s the case)?
Trevor followed, “I don’t think it is”
And JP responds, “I don’t think it is either so I don’t know how to answer that”
In short JP said… What the hell is he talking about?
3/x
JP then explains how Peyto has a 10 yr PDP reserve life index and concludes, “so I wouldn’t consider our decline rates higher.”
(For ref. - $TOU 6 yrs & $ARX 4 yrs PDP RLI)
He ends by says Peyto is also transparent, the decline rate is in the corporate presentation. 😂
4/x
(For ref. - $TOU 6 yrs & $ARX 4 yrs PDP RLI)
He ends by says Peyto is also transparent, the decline rate is in the corporate presentation. 😂
4/x
Noah Baumbach's film Jay Kelly is an anatomy of a movie-star. A must-watch for all those who get overexercised when famous Hollywood actors publish their musings about Israel, Palestinians, or any issue whatsoever.
Kelly' daughter calls him an "empty vessel". /1
Kelly' daughter calls him an "empty vessel". /1
Another character asks,
"Is there a person in there? Maybe you don't actually exist,"
emphasizing the lack of any authentic meaningful life outside of his roles. /2
"Is there a person in there? Maybe you don't actually exist,"
emphasizing the lack of any authentic meaningful life outside of his roles. /2
The film explores the idea of actors being so detached from reality that they struggle to "matter" anyway they can.
The film opens with a Sylvia Plath's line:
"It's much easier to be somebody else". /3
The film opens with a Sylvia Plath's line:
"It's much easier to be somebody else". /3
