HNHacker News
TopNewBestAskShowJobs

troelsSteegin

821 karma · joined February 12, 2017

submissionscomments
troelsSteegin··on The young, inexperienced engineers aiding DOGE
https://doge.gov does not say anything about what the DOGE plan is, and https://www.usds.gov/ is not apparently up to date wrt DOGE. Is there something other than the Executive Order [0] that lays out concretely what DOGE intends to do? This group of engineers is doubtless skilled, but I don't seem them as the decision makers and planners here.

[0] https://www.whitehouse.gov/presidential-actions/2025/01/esta...

troelsSteegin··on The Sonnenschein-Mantel-Debreu Results After Thirty Years
It's an economic theory for characterizing market dynamics. The paper's conclusion: "Thus many of the problematic outcomes from SMD theory remain entrenched."
troelsSteegin··on The Untouched Goldmine of F#
This is really about making stack traces easier to understand.
troelsSteegin··on Show HN: Interactive graphs in Rerun with a Rust port of D3-force
Thank you for your reply. I am thinking about variable dataflows through a fixed set of nodes -- so, the latter, growing and shrinking edges. The application would be visualizing system dynamics. The stages in the Sankey would be organized by degree of distance from an "intervention" node.
troelsSteegin··on Show HN: Interactive graphs in Rerun with a Rust port of D3-force
Is there a public way to add custom Blueprints (https://rerun.io/docs/concepts/blueprint) and Visualizers (https://rerun.io/docs/concepts/visualizers-and-overrides)? It looks like yes for Blueprints, reading through https://rerun.io/examples. I am guessing "not yet" for Visualizers? For example, I'd like to be able to render a Sankey diagram from a "dataflows" Blueprint through the Viewer (https://rerun.io/viewer). I realize that non-spatial is not Rerun's focus.
troelsSteegin··on Principles of Educational Programming Language Design
I wanted the author to pick a winner among approaches, but he's surveying and inviting new research. Most recently, Kölling has supervised the development of Strype, a frame-based editor for python. https://strype.org/
troelsSteegin··on AI Does NYT Connections
This is nicely presented. I would like to see the prompts to the respective services, however. Did I miss them? The "side peek" would be a natural place for them.
troelsSteegin··on Ferrari exec foils deepfake attempt by asking a question only CEO could answer
The book is the develpoment of a lecture by De Toni: https://www.youtube.com/watch?v=bnTjgI14K4c

via auto translate, the 10 points are: Complexity has always existed. Complexity always increases. There is a light side and a dark side of complexity. The signature of complexity is the power law. Complexity manifests itself according to models that can be reconstructed ex post. There is the dilemma of complexity. The complexity of the social world is the highest. Complexity, paradoxes and metamorphoses: the hidden plot. Autonomy and cooperation: the mix for navigating complexity. Complex thought and simple action.

troelsSteegin··on Theory of Constraints
If I may ask, what prompted you to move on from that?
troelsSteegin··on Proposed Oklahoma skyscraper gets redesign to become USA's new tallest building
That looks like potentially a lot of people coming into and out of one city block, like a 10X spike in density relative to everything else. There's public transit in OKC, https://www.embarkok.com/ and https://www.odot.org/transit/s5307/okc.htm, so maybe parking is not an issue, but the "flow" perspective will be interesting....
troelsSteegin··on The Hold of the Dead over the Living: A Conversation with Jill Lepore
This below is a long quote from the article. I'll summarize the quote as ... our belief in data is making us less smart:

You also write that “[t]he era of the fact is coming to an end: the place once held by ‘facts’ is being taken over by ‘data.’” How so?

For years, I have taught a class at Harvard Law School called the history of evidence. In the course of teaching that class, I developed a theory with my students of the transformation of the elemental unit of knowledge across the history of the West, from “mystery” to “the fact” to “numbers” to “data.” “Mystery” is what God knows but humans can’t know. And for a lot of human history, much of how things happen in the natural world was mysterious to us (like where does life come from? What happens when we die?), leaving us with great humility and incredible vulnerability. Many of these things are still mysteries, but a lot of things have been revealed, like how disease is transmitted. The emergence of what historians call “the cult of the fact” is associated with the beginnings of trial by jury, which starts in the 13th century when ordinary people are charged with hearing evidence given by two sides in a dispute and determining the verdict, which means the truth is based on the presentation of facts, or observed acts.

The “fact” transformed all kinds of realms of knowledge: the scientific revolution depends on the idea of the fact, journalism comes to depend on the idea of the fact, and histories tend to understand themselves as compilations and interpretations of facts. The “number” begins to vie alongside the fact in the 18th and 19th centuries with the development of the social sciences. All kinds of things can be quantified, and by counting things you get a whole new way of knowing. And in the 19th and 20th centuries, you start to see the emergence of the preference for the elemental unit of knowledge being “data”—by which, in this context, I mean quantification that requires a machine to do it. Data becomes preferable to facts because facts require the human acts of observation, deduction, reasoning, and conclusion. Data doesn’t require any of those things. It doesn’t require humans. It just requires the accumulation and analysis of numbers. And the objective of the cult of data is not truth, the objective of the cult of the fact, but prediction, because the work is all about pattern detection. If you reorganize all of humanity around prediction, and yet people can’t actually undertake those predictions—those predictions are all offered by machines—we have returned, in a sense, to the age of mystery, because in an age of data, only computers can really know things. So we are as vulnerable as we were in a prescientific, pre-Enlightenment world. It’s a really weird kind of reversal.

troelsSteegin··on Behind the scenes scaling ChatGPT and the OpenAI APIs [video] - Eng Mgr @ OpenAI
See the Show HN: https://news.ycombinator.com/item?id=38555629
troelsSteegin··on Making OKRs more playful using hill charts
> Progress is more like a hill than a straight line. Uphill: figuring things out (uncertainty, unknowns, problem-solving). Downhill: making things happen (certainty, confidence, execution)

This is the first time I've seen a "hill chart" and I found it a little confusing to look at - maybe because the shape looks like a normal distribution, where "the sides" are more uncertain and not less uncertain. If one turned a hill chart inside out, you'd get a chart like a dartboard, where as things are more certain you are closer to a bullseye in the middle, and less certain you are further away. I could see looking at OKRs that way.

troelsSteegin··on Artificial intelligence systems found to excel at imitation, but not innovation
... and our judgments of composition favor recall over precision. Which is to say that as human judges we're good at finding sense in something generated if it's plausible. We'll take a proposition (eg in a composition) and construct a justification for it. "I can see how that makes sense". I am conflating plausible combination with high recall. Precision, in this sense, is that no proposition would be acceptable without evidence or precedent. With all precision, no new idea is ok. So a "knob" for recall vs precision on a generative system would nice. Is "temperature" that knob?
troelsSteegin··on Nokia Bell Labs Is Moving to Helix in New Brunswick
A shout out for the former Bell Labs in Holmdel, now perhaps better known as a location for "Severance". https://en.wikipedia.org/wiki/Bell_Labs_Holmdel_Complex
troelsSteegin··on A Cultural Critique of the Tesla Cybertruck
To my read, the author sets up owning a Cybertruck up as the vehicular equivalent of owning an assault weapon: "A bulletproof three-and-a-half ton stainless-steel truck equipped with “Bioweapon Defense Mode” designed to slam through other cars is the perfect vehicle for a society where over a third of people are scared to walk around at night." Hard to support that without hearing from buyers, but I thought that was rather pointy.
troelsSteegin··on Show HN: Taipy – Turns Data and AI algorithms into full web applications
Thank you. So they are or would be complements to somebody elses commercial platform.
troelsSteegin··on Show HN: Taipy – Turns Data and AI algorithms into full web applications
What is the business model for https://www.taipy.io/, https://streamlit.io/, or https://www.gradio.app/? These are nice tools - but how will the sponsoring businesses support themselves? I didn't see any mention of enterprise plans, etc. Is the answer simply that "we've not announced our revenue model yet"? What should one expect?
troelsSteegin··on The birth and death of search engine optimization
Content farms aren't new - but now they are cheaper and better. I'd think that SEO would have already made a market beneficial to "qualified" content sources - like traditional or branded media. So, a site like vox.com, which has relative to GPT chum, high quality content, asks for donations, but is nominally free. I assume they make money from ads. If a qualified publisher doen't make enough money on ads and donations/subscriptions to cover their editorial costs, and if they do generate enough ad revenue for Google, is there an incentive for Google to help fund the publisher? Is there a case where Google pays for high trust content against which to serve higher margin ads? I clearly don't know how this business works...
troelsSteegin··on AI system self-organises to develop features of brains of complex organisms
I'd look at the underlying paper, https://www.nature.com/articles/s42256-023-00748-9. There's neuroscience references there (50-52) that support their hypothesis. Overall, the corresponding RNN work is meticulously designed and reported, it was a good read.
troelsSteegin··on The NASA engineer who made the James Webb Space Telescope work (2022)
The related [0] NYTimes article mentions one pattern for fixes, which was finding gaps between design and execution:

"For the screws that popped off during the shake test, it turned out that the engineering drawings did not specify how much torque needed to be applied. That was left to the contractor, Northrop Grumman, to decide, and they were not tight enough.

“You should have a specification to make sure it’s right,” Mr. Robinson said.

The review board released its report, noting a series of issues, and made 32 recommendations. NASA followed all of them, Mr. Robinson said.

One of the recommendations was performing an audit of the entire spacecraft to identify “embedded problems” — mistakes that occurred without anyone noticing.

The engineers checked the drawings and specifications. They looked at the purchase requests to make sure that what was ordered matched the specifications and that the suppliers provided the correct items.

“There were multiple teams set up, led by the most experienced people,” Mr. Robinson said. “They really dug into the paperwork.”"

[0] https://news.ycombinator.com/item?id=32109760

troelsSteegin··on I disagree with Geoff Hinton regarding "glorified autocomplete"
From the article, Gelman's money quote is this: " So I’m not knocking auto-complete; I’m just disagreeing with Hinton’s statement that “by training something to be really good at predicting the next word, you’re actually forcing it to understand.” As a person who does a lot of useful associative reasoning and also a bit of logical understanding, I think they’re different, both in how they feel and also in what they do."
troelsSteegin··on Don't Run into the Woods
The Wikipedia article on "walking simulators", https://en.wikipedia.org/wiki/Walking_simulator, suggests it's not just you.
troelsSteegin··on Don't Run into the Woods
I interpreted this advice to just start simple - to describe the experience you intend before designing the state machine to support it.
troelsSteegin··on Open, rigorous and reproducible research: A practitioner’s handbook (2021)
The page embeds https://stanforddatascience.github.io/best-practices/index.h...
troelsSteegin··on Evaluating embedding based retrieval beyond historical search results [video]
Via [0], this is a talk from Haystack, the European Search Relevance Conference [1]. [0] has good bits on this year's conference, which was (not surprisingly) vector-heavy. [0] also links to a nice ebay blog post [2] on an application of Yu Cao's work from the video.

[0] https://www.algolia.com/blog/ai/haystack-eu-2023-learnings-a... [1] https://haystackconf.com/eu2023/ [2] https://innovation.ebayinc.com/tech/engineering/how-ebays-ne...

troelsSteegin··on We’ll call it AI to sell it, machine learning to build it
So, "We'll call it AI to sell it, Machine Learning to Build it", and regression to make it work.
troelsSteegin··on A Theory on Adam Instability in Large-Scale Machine Learning
"The name Adam is derived from adaptive moment estimation.", https://arxiv.org/abs/1412.6980
troelsSteegin··on Guy Kawasaki
What about him? It's that he's a model tech evangelist. Who else? When I was paying attention, I thought Addy Osmani was good, https://addyosmani.com/blog/. Eric J. Ma, https://ericmjl.github.io/, is terrific, though I don't know that he would describe him self that way. Hadley Wickham, https://hadley.nz/, for me would be the gold standard.
troelsSteegin··on It's weird that people get mocked for changing their minds
It's that it's annoying to have to consider changing your mind because they changed theirs. And, it challenges your sense of their authority. Were they wrong in the first place? Should I not trust them? Who should I trust instead? Influential people are valuable for their "insight" and for their "stability" -- both.
← PreviousPage 2 of 9Next →