LightNews
soodoku.bsky.social
soodoku
@soodoku.bsky.social
530 followers 890 following 1.3K posts
gojiberries.io github.com/soodoku I also run a South Asia focused bsky handle: @sudasia.bsky.social
Posts Media Videos Starter Packs
Reposted by soodoku
protzko.bsky.social
Protzko @protzko.bsky.social · 5d
Always remember:

There are no effects without causality
1 1 3
soodoku.bsky.social
soodoku @soodoku.bsky.social · 2h
interviewer bias is super downstream and even basic stats on college major enrollment/graduation rates etc. ought to allow you to develop a mental model of -- if i were hiring at random, i will expect 1 in X to be Y ..
soodoku.bsky.social
soodoku @soodoku.bsky.social · 2h
given your focus, there is a classic (but i think now in some disrepute) paper on blind auditions: www.aeaweb.org/articles?id=... rooney rule paper: www.sciencedirect.com/science/arti... and bertrand/mullainathan, etc.
Orchestrating Impartiality: The Impact of "Blind" Auditions on Female Musicians
(September 2000) - A change in the audition procedures of symphony orchestras--adoption of "blind" auditions with a "screen" to conceal the candidate's identity from the jury--provides a test for sex-...
www.aeaweb.org
1
soodoku.bsky.social
soodoku @soodoku.bsky.social · 11h
Stereotypical Understanding - By Gaurav in Gender — 11 Jul 2016 Stereotypical Understanding The paucity of women in Computer Science, Math, and Engineering in the US is justly widely lamented. Sometimes, the imbalance is attributed to gender stereotypes. But only a small fraction of me…

Read more →
Reposted by soodoku
sungkim.bsky.social
Sung Kim @sungkim.bsky.social · 19h
I guess Intel’s strategy is to take a consumer GPU and just slap on a big a** memory. I kinda like it. 👍

www.tomshardware.com/pc-component...
Intel unveils Crescent Island, an inference-only GPU with Xe3P architecture and 160GB of memory
And a mysterious configuration.
www.tomshardware.com
2 2 22
Reposted by soodoku
johnpfaff.bsky.social
John Pfaff @johnpfaff.bsky.social · 1d
I dread seeing this get the “few bad apples” treatment, in which these … 2,900 pages of texts … get treated as idiosyncratic personal failures for which there is no systemic explanation.
carlquintanilla.bsky.social
Carl Quintanilla @carlquintanilla.bsky.social · 1d
POLITICO: “.. They referred to Black people as monkeys and ‘the watermelon people’ and mused about putting their political opponents in gas chambers. They talked about raping their enemies .. and lauded Republicans who they believed support slavery.

@politico.com
www.politico.com/news/2025/10...
6 20 54
soodoku.bsky.social
soodoku @soodoku.bsky.social · 1d
Hey @gowder.io does this violate the hatch act

www.whitehouse.gov/government-s...

Or Kristi Noem's request to play the message
Government Shutdown Clock
Track the imminent government shutdown as Democrats' demands risk federal services for millions. Americans don’t agree with Democrats’ actions.
www.whitehouse.gov
Reposted by soodoku
thosjleeper.bsky.social
Thomas Leeper @thosjleeper.bsky.social · 1d
Trump wants to turn his assault on cities into visual propaganda for suburbanites.

The inflatable animal suits ruin the imagery.

But so would everyone carrying American flags.

The left should reclaim patriotic imagery so the only videos ICE can make are of beating up banner-waving patriots.
2 2
soodoku.bsky.social
soodoku @soodoku.bsky.social · 1d
Another Coding Issue in the ANES Cumulative File - By Gaurav in Survey Research — 29 Dec 2010 Another Coding Issue in the ANES Cumulative File Technology has made it easy to analyze data. However, we have paid inadequate attention to developing automation in data analysis software that…

Read more →
Reposted by soodoku
gergely.pragmaticengineer.com
Gergely Orosz @gergely.pragmaticengineer.com · 2d
I still think about when New Relic wrote a blog arguing that customers should leave vendors acquired by private equity (PE) b/c prices will go up+quality will go down. It was a stab at Sumo Logic just bought by PE.

6 months later New Relic got bought by PE...and deleted the blog
4 25 220
Reposted by soodoku
soodoku.bsky.social
soodoku @soodoku.bsky.social · 2d
www.gojiberries.io/decoupling-r...
A Penny for Your Thoughts: Budgeting Thought and Action in Agents
Current LLM interfaces hide the choices that determine both quality and cost. A single request triggers internal reasoning, external actions such as retrieval or small program runs, and some amount of...
www.gojiberries.io
1
Reposted by soodoku
soodoku.bsky.social
soodoku @soodoku.bsky.social · 2d
ML (O)Ops: What Data To Collect? (part 3) - By Gaurav in ML/Statistics — 15 Jun 2021 ML (O)Ops: What Data To Collect? (part 3) The first part of the series, “Improving and Deploying On-Device Models With Confidence,” is posted here . The second part, “Keeping Track of Changes,” is post…

Read more →
1 2
Reposted by soodoku
soodoku.bsky.social
soodoku @soodoku.bsky.social · 2d
www.gojiberries.io/decoupling-r...

cc @apoorvalal.com and @johnlist.bsky.social
1
soodoku.bsky.social
soodoku @soodoku.bsky.social · 2d
www.gojiberries.io/decoupling-r...

cc @apoorvalal.com and @johnlist.bsky.social
1
soodoku.bsky.social
soodoku @soodoku.bsky.social · 2d
ML (O)Ops: What Data To Collect? (part 3) - By Gaurav in ML/Statistics — 15 Jun 2021 ML (O)Ops: What Data To Collect? (part 3) The first part of the series, “Improving and Deploying On-Device Models With Confidence,” is posted here . The second part, “Keeping Track of Changes,” is post…

Read more →
1 2
soodoku.bsky.social
soodoku @soodoku.bsky.social · 2d
www.gojiberries.io/decoupling-r...
A Penny for Your Thoughts: Budgeting Thought and Action in Agents
Current LLM interfaces hide the choices that determine both quality and cost. A single request triggers internal reasoning, external actions such as retrieval or small program runs, and some amount of...
www.gojiberries.io
1
Reposted by soodoku
soodoku.bsky.social
soodoku @soodoku.bsky.social · 4d
www.gojiberries.io/testing-when...
Testing LLMs: Engineering Confidence Without Certainty
Software engineers have long leaned on determinism for confidence. Given a function and a specification, we wrote unit tests, fixed the edge cases those tests revealed, and expected tomorrow to look l...
www.gojiberries.io
1 2
Reposted by soodoku
soodoku.bsky.social
soodoku @soodoku.bsky.social · 4d
www.gojiberries.io/decoupling-r...
Decoupling Reasoning from Compute
Current LLM interfaces hide the choices that determine both quality and cost. A single request triggers internal reasoning, external actions such as retrieval or small program runs, and some amount of...
www.gojiberries.io
1
Reposted by soodoku
soodoku.bsky.social
soodoku @soodoku.bsky.social · 3d
www.gojiberries.io/what-problem...
What Problem Is Traditional RAG Solving?
Traditional RAG architecture is a solution to a specific information retrieval problem where quickly retrieving small chunks of information and reasoning over them is the optimal thing to do. The work...
www.gojiberries.io
1
Reposted by soodoku
justinwolfers.bsky.social
Justin Wolfers @justinwolfers.bsky.social · 3d
The Trump-China trade relationship reminds me of my middle school love life: On one week, off the week, high drama, passing notes, heartache, renewal, gossip, and we did it all again the next week.

The reality, of course, is that despite the drama, no-one was making long-term commitments.
20 120 420
soodoku.bsky.social
soodoku @soodoku.bsky.social · 3d
Another ANES Goof-em-up: VCF0731 - By Gaurav in Social Science — 29 Aug 2020 Another ANES Goof-em-up: VCF0731 Photo by Sarah Kilian / Unsplash By Rob Lytle At this point, it’s well established that the ANES CDF’s codebook is not to be trusted ( I’m repeating “not to be trusted to inclu…

Read more →
soodoku.bsky.social
soodoku @soodoku.bsky.social · 3d
www.gojiberries.io/what-problem...
What Problem Is Traditional RAG Solving?
Traditional RAG architecture is a solution to a specific information retrieval problem where quickly retrieving small chunks of information and reasoning over them is the optimal thing to do. The work...
www.gojiberries.io
1
soodoku.bsky.social
soodoku @soodoku.bsky.social · 4d
www.gojiberries.io/decoupling-r...
Decoupling Reasoning from Compute
Current LLM interfaces hide the choices that determine both quality and cost. A single request triggers internal reasoning, external actions such as retrieval or small program runs, and some amount of...
www.gojiberries.io
1
soodoku.bsky.social
soodoku @soodoku.bsky.social · 4d
www.gojiberries.io/testing-when...
Testing LLMs: Engineering Confidence Without Certainty
Software engineers have long leaned on determinism for confidence. Given a function and a specification, we wrote unit tests, fixed the edge cases those tests revealed, and expected tomorrow to look l...
www.gojiberries.io
1 2
Reposted by soodoku
emollick.bsky.social
Ethan Mollick @emollick.bsky.social · 4d
AI is apparently already accelerating science as a co-intelligence.

Measuring academic publications of authors: “we find that productivity among GenAI users rose by 15 percent in 2023 relative to non-users and further increased to 36 percent in 2024” and the quality of publications also went up.
8 17 79