Hacker News: dmvaldman

New comment by dmvaldman in "Ask HN: What Are You Working On? (April 2026)"

dmvaldman — Sun, 12 Apr 2026 21:28:53 +0000

A ring you can talk into and it controls an agent on your phone. Eg say "pick me up" and an Uber arrives.

Looking for people who know hardware well. Let's get to know one another on a flight to Shenzhen :P

New comment by dmvaldman in "Training my smartwatch to track intelligence"

dmvaldman — Fri, 16 Jan 2026 19:27:52 +0000

may the REM sleep gods light your path

New comment by dmvaldman in "Training my smartwatch to track intelligence"

dmvaldman — Fri, 16 Jan 2026 19:27:06 +0000

thanks for catching this!

New comment by dmvaldman in "Training my smartwatch to track intelligence"

dmvaldman — Fri, 16 Jan 2026 19:24:53 +0000

1000%

The Alignment Game (2023)

dmvaldman — Thu, 15 Jan 2026 23:56:26 +0000

https://docs.google.com/spreadsheets/d/1BYh9ZtEv4k7xoSXmtf1q...

Comments URL: https://news.ycombinator.com/item?id=46641216

Points: 55

# Comments: 9

New comment by dmvaldman in "Training my smartwatch to track intelligence"

dmvaldman — Wed, 14 Jan 2026 23:32:55 +0000

it's a lot of work, but something you could do is track how you feel (manually or some other way) and do a similar statistical analysis. chess elo was just convenient and aligned for me.

New comment by dmvaldman in "Training my smartwatch to track intelligence"

dmvaldman — Wed, 14 Jan 2026 23:30:06 +0000

thank you! i think it's ridiculous how little they invest in their developer ecosystem. i have been thinking about jumping ship to oura or whoop simply because of this.

Training my smartwatch to track intelligence

dmvaldman — Wed, 14 Jan 2026 22:19:36 +0000

Article URL: https://dmvaldman.github.io/rooklift/

Comments URL: https://news.ycombinator.com/item?id=46624563

Points: 156

# Comments: 68

New comment by dmvaldman in "How to code Claude Code in 200 lines of code"

dmvaldman — Fri, 09 Jan 2026 01:55:43 +0000

This misses that agentic LLMs are trained via RL to use specific tools. Adding custom tools is subpar to those the model has been trained with. That's why Claude Code has an advantage, over say, Cursor, by being vertically integrated.

A Game to Align People and Priorities

dmvaldman — Thu, 28 Sep 2023 05:25:15 +0000

Article URL: https://docs.google.com/spreadsheets/d/1BYh9ZtEv4k7xoSXmtf1qCP8bYHBCZLEuTVHsTDPQM1M/edit

Comments URL: https://news.ycombinator.com/item?id=37685547

Points: 3

# Comments: 1

Leaving the Company I Co-founded

dmvaldman — Thu, 17 Mar 2022 14:11:19 +0000

Article URL: https://dmvaldman.github.io/

Comments URL: https://news.ycombinator.com/item?id=30711353

Points: 3

# Comments: 0

New comment by dmvaldman in "In defense of flat earthers (2020)"

dmvaldman — Tue, 18 Jan 2022 04:56:29 +0000

i have a different take. i'm very glad flat earthers exist. in general i would hope the population of people who believe an idea be proportional to the probability of its truth. so even the wildest ideas should have some modicum of support. consider a world without this. i would imagine it would necessarily have to be thought-policed. i believe this is how we should frame this discussion.

what i think is the issue is that we have a broadcasting machine (social media, news, etc) that works on sensationalism. so you are always hearing about fringe ideas and given no signaling to the size of the population that supports it.

New comment by dmvaldman in "StyleCLIP: Text-Driven Manipulation of StyleGAN Imagery"

dmvaldman — Sun, 04 Apr 2021 15:52:29 +0000

language will be the next interface to software. to get software to do something, you will simply ask it. this work is an example.

i've been documenting this theme in a twitter thread here https://twitter.com/dmvaldman/status/1358916558857269250

New comment by dmvaldman in "OpenAI API"

dmvaldman — Thu, 11 Jun 2020 20:57:23 +0000

I think this is generally a good answer, but keep in mind I said AGI "in text". My forecasting is that within 3 years you will be able to give arbitrary text commands and get the textual output of the equivalents of "clean my house, take care of my kids, ..." like problems.

I also would contend that there is reasoning happening and that zero-shot demonstrates this. Specifically, reasoning about the intent of the prompt. The fact that you get this simply by building a general-purpose text model is a surprise to me.

Something I haven't seen yet is a model simulate the mind of the questioner, the way humans do, over time (minutes, days, years).

In 3 years, I'll ping you :) Already made a calendar reminder

New comment by dmvaldman in "OpenAI API"

dmvaldman — Thu, 11 Jun 2020 19:19:37 +0000

what is the difference between zero-shot learning in text and AGI? not saying there isn't one, but can you state what it is?you can express any intent in text (unlike other media). to solve zero-shot in text is equivalent to the model responding to all intents.

many people have different definitions for AGI though. for me it clicked when i realized that text has this universality property of capturing any intent.

New comment by dmvaldman in "OpenAI API"

dmvaldman — Thu, 11 Jun 2020 19:13:00 +0000

i think you are assuming that what is happening under the hood is that a human-inputted sentence is being parsed into a grammar. it is not.

New comment by dmvaldman in "OpenAI API"

dmvaldman — Thu, 11 Jun 2020 19:06:01 +0000

Zero shot and few-shot learning in GPT-3 and lack of significant diminishing returns in scaling text models. Zero-shot learning is equivalent to saying "i'm just going to ask the model something that it was not trained to do"

New comment by dmvaldman in "OpenAI API"

dmvaldman — Thu, 11 Jun 2020 17:16:11 +0000

AGI in text is < 3yrs away.

New comment by dmvaldman in "A 2020 Vision of Linear Algebra"

dmvaldman — Tue, 12 May 2020 13:49:31 +0000

I think these are good examples, but to me "linear algebra thinking" lies in it's generality. For example, the derivative is a linear operator, so how do you write it down as a matrix? Google's PageRank is a solution of a matrix equation, what does that matrix represent? Etc.

New comment by dmvaldman in "Keys.pub – Manage cryptographic keys and user identities"

dmvaldman — Mon, 27 Apr 2020 21:19:04 +0000

80/20 rule? Make the problem simpler and deliver a better solution for it. Revisit and grow the problem space as needed.