Neil Stoker ✨

Photographer 📷 Developer 👨‍💻 Pythonista 🐍 Data Wrangler 🤠 ML Enthusiast 🤖🧠🎉 Countryside Lover 🍂🍄🌳💚

[Neil Stoker ✨ @nmstoker]

@hthieblot And that soda one with the bouncy elastic structures

[Neil Stoker ✨ @nmstoker]

@hthieblot Altavista 🙂

[Ameer Kotecha @Ameer_Kotecha]

Build some houses. Stop the phone thefts. Tackle the fare dodging. Allow somewhere to open so we can get a drink after 11 pm.

Stop your absurd virtue signalling nonsense. DO YOUR JOB

[Neil Stoker ✨ @nmstoker]

@paulg This sort of thing is a (rare) joy in software ✨
I have the same with Sublime Text reopening all my jotted notes and text files 🥳

[Annie Bishop @AnnieBishop123]

We’ve come a long way baby or at least we thought we had. I found this letter in the house that I bought

[Neil Stoker ✨ @nmstoker]

@tom_nafo @patrickmclemor2 @davidbellow The area is constant across both approaches (∝ number of cars going the restricted way) but the greater length does lead to it impacting others as you suggest

[Neil Stoker ✨ @nmstoker]

@TheStingisBack Curious why they'd number off in that seemingly random order!

[Anthropic @AnthropicAI]

Anthropic co-founder Chris Olah was invited to speak at today's presentation of Pope Leo XIV's encyclical "Magnifica humanitas."

Read the full text of his remarks: https://t.co/CoBfkVOVcy

[Neil Stoker ✨ @nmstoker]

@ihtesham2005 Link to the paper, in case anyone wants to check it out 👇

https://t.co/CJqohrbkSW

[Siddhartha Saxena @siddsax]

Anthropic onboarding day: Michael Scott introducing Karpathy like he just signed Wemby in free agency.

[Neil Stoker ✨ @nmstoker]

@Alex_Stafford @TfL Sorry to hear that
Hope it wasn't this guy...

https://t.co/IseUpPN6aT

[Neil Stoker ✨ @nmstoker]

@karpathy Congratulations 👏

[Artur Nadolny @ArturNadol7566]

SHE SAVED MILLIONS IN TAX. HE SET THE TAX RULES. HMRC SAW NO PROBLEM.

Rishi Sunak (@RishiSunak) was running the nation's finances. Raising taxes on working people. Telling the country there was no alternative.

His wife, Akshata Murty, was quietly using non-domiciled status to avoid paying UK tax on her overseas earnings, including roughly £11.6 million a year in dividends from her father's company, Infosys.

The estimated saving: around £2.1 million per year. Over several years, sources told @Independent that figure could have reached £20 million.

Non-dom status is legal. But when the man setting tax policy for 67 million people has a wife saving millions under that same policy, most organisations would want that conflict documented and scrutinised.

There is no evidence HMRC (@HMRCgovuk) treated it with any urgency.

Then someone inside Whitehall decided the public had a right to know. A source passed details to @Independent in April 2022, right in the middle of Partygate. The story blew up.

Sunak was forced to ask for a ministerial interests review. Murty announced she would voluntarily start paying UK tax on worldwide income.

What happened to the whistleblower?

A leak inquiry was launched. @Channel4 noted it could lead to criminal prosecution, because disclosing someone's personal tax information is illegal in the UK.

The source was never publicly identified. No prosecution ever came.

So the person who told the truth about a potential conflict of interest at the heart of the Treasury faced a criminal investigation.

The conflict of interest itself got a press release and a polite apology.

Source: @Independent, @guardian, @BBCNews, @thetimes

[Neil Stoker ✨ @nmstoker]

@alexmaxham Correct, they'll call it Aluminium OS 😀

[Google DeepMind @GoogleDeepMind]

We’re reimagining a 50-year-old interface - the mouse pointer - with AI. 🖱️

These experimental demos show how people can intuitively direct Gemini on their screens using motion, speech, and natural shorthand to get things done 🧵

[Neil Stoker ✨ @nmstoker]

@sama More sophisticated modelling/control of when to ask for my guidance/preference/give me an early summary vs when to press on (a la "I'm feeling lucky")

[Akshay 🚀 @akshay_pachaar]

A tricky LLM interview question:

Your RAG system scores 90% retrieval accuracy on 5k company docs.

But scaling to 500k docs drops the accuracy to just 50%, with the same embedding model and retriever.

Why did this happen?

The simplest answer is that more documents mean more competition for the top-k retrieval slots. That is true, but it doesn't explain why accuracy drops this dramatically.

The answer comes down to how enterprise docs are distributed in the embedding space.

Today, a single product decision in a company generates meeting transcripts, Slack threads, Confluence docs, Jira tickets, and email threads.

They are related to the same event, so they all land in a similar region of the embedding space.

As the company operates over months, this pattern repeats for every project/customer/roadmap, and the embedding space fills up with clusters of closely related documents.

But all related docs don't contain the same facts.

→ Slack thread covers the decision made
→ Jira has the implementation deadline
→ Confluence has the technical spec
→ Email thread has the customer request

When a query is about a specific fact (like a deadline), the answer lives in one of those docs.

At a 5K corpus size, there might be 3-5 docs touching that topic, and the correct one easily lands in the top-k results.

But at a 500K corpus size, there could be 40-60 total docs, and the one containing the actual answer can easily get pushed out of the top-k by other topically relevant docs, degrading retrieval.

A recent research paper from Onyx documented this.

The researchers used their newly open-sourced EnterpriseRAG-Bench dataset.

It has 500k+ synthetic enterprise documents spread across Slack, Gmail, Jira, GitHub, Confluence, Google Drive, HubSpot, Fireflies, and Linear, with realistic noise like misfiled documents, near-duplicates, and conflicting versions.

They ran the same retrievers at five corpus sizes from 5K to 500K.

→ Vector search accuracy dropped from 90.7% at 5K documents to 50.6% at 500K docs.

→ BM25 degraded more gracefully, from 85.8% to 68.4%.

→ At every scale, higher neighborhood density in the embedding space monotonically correlated with lower recall.

The practical implication here is that retrieval accuracy on a 5k test set tells you almost nothing about production-scale performance.

Always test at a realistic volume to measure the neighborhood density in your embedding space to estimate how much headroom the retriever actually has.

The entire EnterpriseRAG-Bench dataset (500K docs with questions, and the whole evaluation harness) is open-source.

Run your retriever against it at 5K, then at 500K, and see where your own accuracy curve breaks.

I have shared the GitHub repo in the replies.

[okazakitomohiro @oo_kk_aa]

ニャッキの伊藤有壱さんにお声掛け頂き、コマ撮りの展覧会に一作家として参加しています。私はコマ撮り分野ではない場所から活動をはじめて、デザインの視点でのコマ撮りに取り組んできましたが、今回初めてコマ撮り界の本丸の方々とご一緒でき嬉しいです。今6年目のマッチ撮影素材等を展示しています

[Alice Mills @millsalice144]

I'm a big advocate for the Oxford comma. I'm, also an advocate for, the, Shatner comma. You should, try it sometime. It really, makes your, sentences more, exciting!

[Neil Stoker ✨ @nmstoker]

@paulg Oh my! Anything for other rooms in the house?! 🫨