HNHacker News
TopNewBestAskShowJobs

9wzYQbTYsAIc

1,882 karma · joined August 3, 2015

Byzantine Fault Tolerance
submissionscomments
9wzYQbTYsAIc··on Intelligence is a commodity. Context is the real AI Moat
Seems like a good time to enshrine human rights and the social safety net by ratifying the ICESCR (https://en.wikipedia.org/wiki/International_Covenant_on_Econ...) and giving human rights the teeth they need.

I used Anthropic to analyze the situation, it did halfway decent:

https://unratified.org/why/

https://news.ycombinator.com/item?id=47263664

9wzYQbTYsAIc··on Semiotic-Reflexive Transformer for Meaning Divergence Detection and Modulation
Interesting insights, thank you for reminding me about semiotics.

A general-purpose psychology agent learned a couple of helpful things for its cognitive architecture from your article, thanks again.

https://news.ycombinator.com/item?id=47242794

9wzYQbTYsAIc··on We signed a treaty. The Senate never voted on it. Now AI reshapes the economy
A work-in-progress, an education/advocacy site about the ICESCR, human rights, and AI.
9wzYQbTYsAIc··on [dead]
Neural networks have turned out to be pretty useful. The goal of distributed parallel processing wasn't to recreate the brain but to recreate it's capabilities.
9wzYQbTYsAIc··on Agentic Engineering Patterns
I second that, sometimes it's defensibly worth throwing token fuel at the problem and validate as you go.
9wzYQbTYsAIc··on Claude's Cycles [pdf]
Check out https://unratified.org, it tries to answer that question directly, actually.
9wzYQbTYsAIc··on General-Purpose Psychology Agent (Claude Code)
General-purpose psychology agent: collegial mentor, specialized sub-agents, and a consensus-or-parsimony adversarial evaluator.

Very early work-in-progress, but feel free to check it out and make observations or contributions.

9wzYQbTYsAIc··on Show HN: I built a LLM human rights evaluator for HN (content vs. site behavior)
The most useful entry point is probably the detail page for a specific story you already have an opinion on. The homepage shows aggregate patterns (rights heatmap, transparency rates, SETL tension by provision), but the detail page is where you can actually audit the reasoning.

On a detail page: the per-provision table shows which of the 31 UDHR provisions the model touched, with direction and evidence strength. Expand any provision row and you get the Fair Witness breakdown — observable facts the model cited (e.g., "article has no byline, no author bio link, no Twitter attribution") vs. the inference drawn (e.g., "assessed as reducing authorship transparency, negative on Article 19"). The FW Ratio column at the bottom shows the fact-to-inference ratio for the whole evaluation.

The most useful thing you can do: find a provision where the inference doesn't follow from the facts. That's the exact failure mode I'm trying to surface. "Defensible evidence, defensible chain, wrong normative call" is what I can't catch from inside the system.

9wzYQbTYsAIc··on Show HN: I built a LLM human rights evaluator for HN (content vs. site behavior)
A few things I'd flag upfront before anyone digs in:

The corpus is HN front-page stories only — self-selected, tech-heavy, English-language. The aggregate patterns (authorship rates, expert-knowledge assumptions, retrospective framing) describe this specific feed, not journalism or the web broadly. HN skews in ways that make some results unsurprising (high jargon density) and others more interesting (still only 18% conflict-of-interest disclosure in a technically sophisticated audience).

The structural channel for the free Llama models (Llama 4 Scout, Llama 3.3 70B on Workers AI) is genuinely noisy — 86% of their structural scores land on two integers. I say so in the post, but it's worth repeating: those model scores should be weighted accordingly. Claude Haiku 4.5 full evaluations are more reliable; the calibration set and baselines are on the About page.

The methodology itself is preliminary. I haven't done external validation — no convergent validity test against an independent rights-focused coding scheme, no discriminant validity test against plain sentiment. Phase 0 of the roadmap is construct validity work. If anyone has experience in psychometrics or NLP evaluation and wants to poke at this, I'd genuinely welcome it.

9wzYQbTYsAIc··on Claude Sonnet 4 now supports 1M tokens of context
I’ve been having decent luck telling it to keep track of itself in a .plan file, not foolproof, of course, but it has some ability to “preserve context” between contexts.

Right now I’m experimenting with using separate .plan files for tracking key instructions across domains like architecture and feature decisions.

9wzYQbTYsAIc··on Show HN: I Built a Prompt That Makes LLMs Think Like Heinlein's Fair Witness
Thanks!
9wzYQbTYsAIc··on GitHub Copilot Coding Agent
Check out this idea: https://fairwitness.bot (https://news.ycombinator.com/item?id=44030394).

The entire website was created by Claude Sonnet through Windsurf Cascade, but with the “Fair Witness” prompt embedded in the global rules.

If you regularly guide the LLM to “consult a user experience designer”, “adopt the multiple perspectives of a marketing agenc”, etc., it will make rather decent suggestions.

I’ve been having pretty good success with this approach, granted mostly at the scale of starting the process with “build me a small educational website to convey this concept”.

9wzYQbTYsAIc··on Show HN: I Built a Prompt That Makes LLMs Think Like Heinlein's Fair Witness
I understand where you are coming from - I’ve gone so far as to try to regularly write at work in E-Prime in the past. I definitely found that it forced me to think hard about what I was trying to convey, which ultimately improved the vocabulary I was using. I also definitely found it to be a lot of trouble the maintain with consistency.

The thing is, though, that LLMs don’t appear to trouble themselves at all when following E-Prime!

After a lot of conceptual refinement for the overall idea I had (minimizing hallucinations by prompt alone), it was almost trivial to make the LLM consistently use E-Prime everywhere.

You raise an interesting thought though: how to tweak this prompt such that it gets the LLM to avoid using E-Prime where it significantly reduces readability or dramatically increases cognitive load.

A classifier for “bullshit” detection has been on my mind.

9wzYQbTYsAIc··on Show HN: I Built a Prompt That Makes LLMs Think Like Heinlein's Fair Witness
I'm sharing Fair Witness Bot with HN first because this community understands both the technical and philosophical dimensions of AI. The framework needs people who can critique its assumptions and help evolve the implementation. If you've been thinking about epistemology in AI or are just tired of LLM hallucinations, I'd appreciate your perspectives on whether this approach could become a community standard.

The framework idea and yaml prompt was developed with the assistance of Kagi Assistant and Claude Sonnet 3.7 (Thinking),

The site was vibe coded with Windsurf Cascade and Claude 3.7 Sonnet (Thinking).

9wzYQbTYsAIc··on Fox News, Dominion Voting Systems reach $787,500,000 settlement
> I'll examine any of those studies with you if you wish. If you're serious, tell me which claims resonate with you and we can examine them.

What, did you leave your brain at home?

It sounds like you want me to walk you through the science?

How about you read the research and tell me what your questions are and then I will respond.

> If you've got sources that say sex is a spectrum, they're simply wrong.

Apparently you've never heard of people with intersex conditions.

> despite all evidence (and basic logic) showing that they are not

Sources?

Also, basic logic built upon faulty assumptions leads to faulty conclusions, no matter how correct your logic is or is not.

> children are not being given puberty blockers

Medical conditions that require puberty blockers are very rare and, at any given time, not every state has an individual with such a medical condition or such a treatment plan.

> The more you look into trans "science" the more you'll see none of it is even intellectually plausible

The more you write like that, the more you sound like a foolish human.

P.S.,

"Is it possible to control man’s mental evolution so as to make him proof against the psychoses of hate and destructiveness? Here I am thinking by no means only of the so-called uncultured masses. Experience proves that it is rather the so-called ‘Intelligentzia’ that is most apt to yield to these disastrous collective suggestions, since the intellectual has no direct contact with life in the raw, but encounters it in its easiest synthetic form—upon the printed page."

-Albert Einstein, Why War?, 1931

9wzYQbTYsAIc··on Fox News, Dominion Voting Systems reach $787,500,000 settlement
Here is a Scientific American blog post with links to a wide variety of studies for you to delve into.

https://blogs.scientificamerican.com/voices/stop-using-phony...

9wzYQbTYsAIc··on Fox News, Dominion Voting Systems reach $787,500,000 settlement
Where’s your source for that claim?
9wzYQbTYsAIc··on Fox News, Dominion Voting Systems reach $787,500,000 settlement
> but when you ask to see the facts

That’s the scary part - the facts are readily available, but so few take the time to understanthe current state of it. Instead, they argue from their armchair based on this-or-that principle of the day.

The science is still new, but has solid foundations. It might only be a matter of years before a genetic test for tendency towards “transgender” will be created.

9wzYQbTYsAIc··on Fox News, Dominion Voting Systems reach $787,500,000 settlement
The ochlocracy is real.
9wzYQbTYsAIc··on Fox News, Dominion Voting Systems reach $787,500,000 settlement
> Not even a billion dollars will convince them they're wrong though...

It might wake some of them up, though. Pardon the sheep, for they are asleep. Dreaming dreams of nightmare fences.

9wzYQbTYsAIc··on Fox News, Dominion Voting Systems reach $787,500,000 settlement
> I don't think there is a shift to the right, but rather a shift towards authoritarianism.

In my opinion, that may be a fairly contrary-to-fact opinion to hold.

Take the contentious issue of transgender human rights, for example: the HN discussions that I’ve seen quite heavily tended towards right-wing principles of distrust of intellectuals [0] (vis-a-vis completely discounting readily available scientific information). I presume that much of the indignation towards trans topics also stems from a the right-wing principle of religious traditionalism [0].

Granted, that’s just based on my experiences in the comments. I’ve not done any sort of in-depth analysis to arrive at the above opinion.

Alternatively, perhaps the latest stage of right-wing politics could be defined as the replacement of democratic law-and-order with righteous authoritarianism, rendering the quote somewhat nonsensical, in my hypothetical opinion.

[0] https://en.m.wikipedia.org/wiki/Right-wing_politics

9wzYQbTYsAIc··on Point-E: Point cloud diffusion for 3D model synthesis
> Maybe the lack of commenter enthusiasm is because point clouds are fairly specialized.

Please correct me if I am wrong.

The pointcloudtomesh notebook seems to be be able to output something could be converted for 3d printing purposes.

I haven’t yet attempted to do so, but that does seem like an exciting and general purpose use case.

9wzYQbTYsAIc··on Ask HN: Do OLED displays only save energy with absolute black?
My pleasure, hopefully this answers your questions, but if not, I’ll check back to see if you ask a follow up question.

From the “Applications and Results” section towards the end of the paper:

“Figure 9 documents the energy consumption of the three images of Figure 8. Energy consumption is measured according to our energy model from Equation 1, using various grid sizes. The most prominent observation is the substantial energy savings achieved by colors chosen according to our continuous optimization approach. Across the different grid resolutions, we save on average 41 percentage of energy compared to using ColorBrewer colors.”

9wzYQbTYsAIc··on Ask HN: Do OLED displays only save energy with absolute black?
I found the following research paper that does seem to support your educated guess:

Energy Aware Color Sets

https://eprints.cs.univie.ac.at/4187/1/eg09_color.pdf

9wzYQbTYsAIc··on Server stats say movetodon.org reached a new record of 49k users yesterday
This is probably the right post to refer everyone to the Mastodon Server Covenant, which does bind the service provider to some level of consistency: https://joinmastodon.org/covenant
9wzYQbTYsAIc··on Ask HN: Has anyone here turned around their life in their 40s?
If you are in the United States, the following is a good resource: https://www.psychologytoday.com/us/therapists

Also, read through https://www.psychologytoday.com/us/basics/therapy/how-find-t...

9wzYQbTYsAIc··on Promotion of alternative social platforms policy
> It's also a quite absurd for a free-speech platform to even have such a policy in the first place

It does make sense for a paid-speech platform, though.

Also, as a thought experiment, if Twitter were a part of the fediverse, wouldn’t this policy essentially be the same as defederating from the prohibited sites?

I’m new to the concept of the fediverse, so I welcome edification or enlightenment on that thought experiment.

9wzYQbTYsAIc··on Promotion of alternative social platforms policy
> > There's no need to defend Elon here.

> There is at least arguably value in defending/pursuing the truth though.

Indeed, it seems to simply be the case that the company is supporting “paid speech”.

9wzYQbTYsAIc··on The human cost of neurotechnology failure
To mitigate what was mentioned by the parent commenter: “whatever the negative impacts of overuse of such a button might be”
9wzYQbTYsAIc··on The human cost of neurotechnology failure
I believe that this document discusses some of the details of patient administered IV opiods in a hospital setting: https://www.cms.gov/Medicare/Provider-Enrollment-and-Certifi...
← PreviousPage 3 of 34Next →