HNHacker News
TopNewBestAskShowJobs

rahimnathwani

16,483 karma · joined March 9, 2012

I've lived in San Francisco since 2019.

2010-2019 I lived in China.

Before that I lived in London, where I was born.

My personal web site: https://encona.com/

Random tools: https://tools.encona.com/

Other random stuff about me:

- English is my native language, but I also speak (to varying degrees of fluency) Mandarin, Gujarati and Spanish.

- My career so far has been a mix of startups/cofounding, big tech (I was a PM at Amazon and at Google), and exec roles.

- I'm a qualified accountant (CIMA), have two undergrad degrees (one in CS) and an MBA (sorry!).

- I got married in 2012 and our son was born in 2016 (I spend a lot of time thinking about how best to raise him to be a capable and kind adult)

Email: rahim AT encona DOT com

LinkedIn: http://www.linkedin.com/in/rahimnathwani

X: https://x.com/rahimNathwani/

submissionscomments
rahimnathwani··on Best of British Design
Here's my chart: https://best-of-british-design.vercel.app/#r=R_D4nrm510s
rahimnathwani··on Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms
$0.042 not $0.42
rahimnathwani··on Jeff – Jev-compatible 0.8B decision models, trained at home, ~30 ms
"Using embeddings + logistic classifier, the architecture matches or beats Jev and Laya in all basic classification"

Have I understood correctly that you trained only the logistic classifier, but didn't need to train the embedding model?

If so, I'm curious whether you compared that approach (A) with:

B) Jev only, with a single output.

C) Jev with multiple outputs fed into a logistic classifier.

Obviously C has cons (can't be self-hosted, needs some up-front work on deciding the shape of the output) but it might be somewhat more interpretable. (And I suppose it might have better performance?)

rahimnathwani··on Cf: The Agentic CLI for the Cloudflare API
You linked to /skills/productivity/to-questionnaire/SKILL.md

Did you mean to link to /skills/engineering/wizard/SKILL.md ?

rahimnathwani··on Reflections on 1,000 Days of Math
If you want to do this, there are at least two ways:

A) use the AoPS books, which present exercises before explanations, and expect students to attempt them first, or

B) Use Math Academy the way my son does: skip the explanation and worked example and immediately attempt the first problem. Only if stuck go back to the explanation and example.

(B) isn't what Math Academy recommended, presumably because you might be able to solve the problems in the current lesson, but still miss out on something else the lesson is supposed to teach you.

rahimnathwani··on Jev and System One Models: Calibration Beats Accuracy
The way he proposed to deal with "How urgent is this customer support message?" is "a score over ordered levels".

I'm not sure that's a good way to use Jev for this use case. If I had this problem, I would ask Jev to answer several different yes/no questions about each message, and then use the probabilities as inputs into a logistic regression model that predicts urgency.

If you ask the model specific yes/no questions which can be answered reasonably objectively from the input, I think the answers are going to be more stable over successive generations of models.

e.g. if you ask 'Is the customer angry?' I'd expect that answer to have high agreement between models and between models and humans. But directly answering the 'is it urgent' question is much harder. (Although I suppose you can try to put the rules in the prompt.)

rahimnathwani··on Contrastive Language Models
From a tweet:

https://x.com/jackyk02/status/2102947585208258799?s=20

  I actually explored this for robotics last year, and CLM grew directly out of that work a few months ago. Feel free to check out my earlier post for more details on how CLM could be applied to robotics:
  https://x.com/jackyk02/status/2026368947210289660
But of course the repo with the TypeSafe compatible API is new: https://github.com/Contrastive-LM/CLM
rahimnathwani··on Show HN: Make math automatic with Mathy
This is great, Gabe.

One small piece of feedback: on my Pixel, the Android keyboard obscures the green button. So when I'm answering a numerical question, I use your number pad to enter the digits, and then use the enter key on my Android keyboard. This feels weird, but I guess with regular use it wouldn't.

rahimnathwani··on Stanford R&DE Uses AI to Race Swap Students for Advertising
Life imitates art: https://theonion.com/black-guy-photoshopped-in-1819565862/
rahimnathwani··on Claude Opus 5.5
"Opus 5.5 requires less compute to serve than Opus 5, and its pricing reflects that."
rahimnathwani··on Jev – A curation of Jev demos on X, tools, skills, and integrations
I'm curious why you didn't use a spaCy for this. It would probably be faster, and you'd avoid a third party API dependency.
rahimnathwani··on Vacate a drone restriction that criminalized recording immigration agents
If you read the notice, it seems to effectively make it illegal to fly drones anywhere near a road. How can a drone operator possibly verify that they're NOT within 3000ft of a vehicle owned by DOD, DOE or DHS? Are all these cars marked in such a way that a drone that's over 3000ft away would be able to recognize it, so the operator knows not to go any closer?

  PURSUANT TO 14 CFR SECTION 99.7, SPECIAL SECURITY INSTRUCTIONS (SSI), ALL UNMANNED ACFT ARE PROHIBITED FROM FLYING WITHIN A STAND-OFF DISTANCE OF 3000FT LATERALLY AND 1000FT ABOVE UNLESS INDICATED BY THE FAA BY NOTAM OR OTHER MEANS. TO: DEPARTMENT OF DEFENSE (DOD), DEPARTMENT OF ENERGY (DOE), AND DEPARTMENT OF HOMELAND SECURITY (DHS) FACILITIES AND MOBILE ASSETS, INCLUDING VESSELS AND GROUND VEHICLE CONVOYS AND THEIR ASSOCIATED ESCORTS, SUCH AS UNITED STATES COAST GUARD (USCG) OPERATED VESSELS.
rahimnathwani··on You can use any LLM just like JEV
I'm skeptical about the described 'multiple questions at once' technique scaling to more that 2^3 questions.
rahimnathwani··on Qwen Image 2.1
How are you converting the bitmaps into vector images?
rahimnathwani··on Learning another language may be one of the best ways to keep your brain healthy
Yes, but in their 30s most people are:

- unwilling to move to a different country

- unmotivated to spend time chatting with people about random things in a foreign language

rahimnathwani··on Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint
M1 Pro, same prompt, same cli options:

  32,706 tokens
  38min 19s
  14.22 t/s
rahimnathwani··on Canto: A speech model built for the real world
Same in Latin.

And in Italian and Portuguese, but I think they might use pronouns, too.

rahimnathwani··on Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint
If you want to download the gguf to your regular huggingface cache directory instead of to /tmp, you can download the model and run the server in one step:

  export HF_TOKEN=xxx # optional, speeds up the download
  
  ./llama-prism-b10685-7dffb15/llama serve \
    -hf prism-ml/Ternary-Bonsai-2-27B-gguf:PTQ1_0 \
    --port 8331 -ngl 99 -fa on -c 32768
rahimnathwani··on Jev Ultrafast: A browser agent with a dynamic, indexed action space
I tried it just now. After you click 'Start Demo' you should drag the newly-created tab into its own window (so you can watch it). Then click the 'Run automatically' button.

When I did this it worked as described.

rahimnathwani··on Show HN: Nari Qwen3-TTS and Qwen3-ASR – High accuracy, low latency and cost
For some reason it switched voices half way through a 33 second clip.

For OP the clip name is nari-nina-01a0a12f-980a-765e-8029-fa56bd23210d.wav

rahimnathwani··on US Reading Scores Plummet on Global Exam
Thanks, Roshan. I didn't know this interface existed.

I had thought the only way to access the data was via downloadable data files (no longer published) or the PDF reports.

rahimnathwani··on Ask HN: How do you manage skills files?
"Don't know why the below comment by killix got flagged"

Because it's obviously written by AI.

rahimnathwani··on US Reading Scores Plummet on Global Exam
What's your source for the 2022 difference?

PISA itself doesn't publish data broken down by racial subgroup. Back in 2018, the NCES public download included a RACETHC variable (https://nces.ed.gov/pubs2021/data/2021019_rawdata.zip).

But the most recent file on the NCES PISA 'data products' page is from 2018: https://nces.ed.gov/surveys/pisa/publications.asp

(At the time of writing if you expand 'Data Products' the first file is 'Program for the International Student Assessment (PISA) 2018 Public Use File (PUF) NCES Number: 2021019 Release Date: July 8, 2021')

rahimnathwani··on Learn Programming with OCaml
How old were you?
rahimnathwani··on .name Termination
He didn't say his IoT devices relied on 'cloud' accounts. He said there are IoT that use services on this domain.

Imagine he's set up some IoT devices at his parents' home, and those devices use services that he hosts somewhere on the internet. It would be silly to hard code the IP addresses in there, right (unless he operates his own ASN)? So he would use DNS to allow those devices to find his server(s). This would be the case whether the servers are at his home, at his office, or in a rack at a data centre.

rahimnathwani··on Gemini 3.8 Flash and 3.8 Flash Cyber
Right, I mentioned that here: https://news.ycombinator.com/item?id=49539339

jampa was saying that the Gemini 3.7 (a model) ranks higher on real-world knowledge. The point I was originally making is that I would trust any model by itself to answer real-world knowledge problems. If I want to know opening hours, then probably any of the models could find the answer with a web search tool.

Perhaps the Gemini API makes this easier, as the web search tool is built in.

rahimnathwani··on A third of Perplexity's citations don't contain the number they're cited for
I did something similar for structured text extraction. I added markers throughout each source document and then, for each piece of info I wanted, I asked the LLM to provide two separate fields:

  xyz
  xyz_citation
The latter was just the node number. So then my code could extract the exact snippet, instead of trusting the LLM to quote something verbatim.
rahimnathwani··on Gemini 3.8 Flash and 3.8 Flash Cyber
You can opt in on a per-request basis.
rahimnathwani··on Gemini 3.8 Flash and 3.8 Flash Cyber
When you called the Gemini API, did you opt in to using search grounding:

  tools=[{"type": "google_search"}]
I'm curious whether in fact you were getting answers from the model weights (which is what I had assumed) or whether your API calls were resulting in web search tool calls.
rahimnathwani··on Gemini 3.8 Flash and 3.8 Flash Cyber
One thing in your comment surprised me: "when a thing opens and closes"

Why you would rely on the model's weights to know opening hours, instead of having the model call a web search tool to verify it on the official site?

← PreviousPage 3 of 34Next →