rectang

Original source: Thiel’s Wall Street Journal essay:

https://www.wsj.com/articles/peter-thiel-competition-is-for-...

https://www.csun.edu/~vcact00f/497CapStone/Peter%20Thiel_%20...

The title is “Competition is for losers” and the sub-head is “If you want to create and capture lasting value, look to build a monopoly, writes Peter Thiel”.

> I still can’t work out why it wasn’t obvious from the start that UCS-2 would never be enough)

Surely certain people did know, but those people weren't in a position to do anything about it.

Specifically, there were surely people who knew that because historical Chinese place names, Japanese nicknames, and so on, were not included in the original "Unicode" (it wasn't called UCS-2 yet) it was insufficient for complete expression of Asian languages.

There were also many people who objected to Han unification, which is a different problem.

But all of these objections were discarded because of the overwhelming mandate for a fixed-width encoding. The original "Unicode" was conceived as a "16-bit" initiative. Its 16-bit-ness was an essential aspect of the design and the Unicode Consortium did what they had to do to fit all scripts and characters "in modern use" into 16 bits.

From the Wikipedia article on Han Unification[1]:

> Some of the controversy stems from the fact that the very decision of performing Han unification was made by the initial Unicode Consortium, which at the time was a consortium of North American companies and organizations (most of them in California), but included no East Asian government representatives. The initial design goal was to create a 16-bit standard, and Han unification was therefore a critical step for avoiding tens of thousands of character duplications.

[1] https://en.wikipedia.org/wiki/Han_unification

That's possible, and I hedged a bit with an "if I understand correctly" because it was unclear to me how reliable the "coal roller" assessment was.

But if the bike rider's judgement was mistaken, the only consequence seems to have been that a driver of a diesel truck had ill wished upon them from afar. There's no mention of the truck getting reported to the police, and I speculate that the bike rider, as someone capable of recognizing the truck driver's good intent, might be cautious about escalating.

Nevertheless I'm sure that there will be those who continue to see the bike rider as an "unhinged" monster.

If I understand correctly, the trucker was set up to roll coal on other people, and only made an exception for this specific kid on a bike. It's not "unhinged" to stand up for others who have been targeted even while you were spared — it's just common decency.

I understand you to be implying that I shouldn't trust my perception that there's a meaningful difference in how much different models hallucinate. I will take that under advisement, but I am still interested in the answer to my original question.

How does "outperform" translate to the propensity of an LLM to hallucinate?

In my anecdotal experience there is a huge gap between GPT-5-mini which hallucinates relentlessly and Claude Opus or the latest GPTs which are fairly reliable. I'm hoping that gap can be closed with improved approaches for small models and that good reliability is achievable for LLMs without requiring absolutely mammoth computing resources.

For what it's worth, I also used GPT-5.2 (via duck.ai) this year for questions about taxes and it was helpful — which makes sense because there's an abundance of material about taxes out there to be synthesized, so a text predictor trained in that domain should do well.

Onkalo is the best approximation so far of of a memory hole for nuclear waste — but Finland has not agreed to accept all the world’s nuclear waste, similar sites are not available for all countries, and in practice storage remains a major point of contention as a wander through this discussion will show.

> The fearmongering against nuclear was always crazy to me.

I sometimes feel similarly about pro-nuclear cheer mongering.

There are a number of pro-nuclear advocates on this page who can’t just engage in good faith but instead dismiss and deride their interlocutors as “crazy”, “paranoid”, “fear mongers“. It’s infuriating because the origins of anti-nuclear sentiments are often plain and perfectly understandable, as your poignant anecdotes illustrate.

Brazil nuts and bananas are radioactive too, compared to other foods — but do not pose risks compared to coal ash.

Similarly, while coal ash is nasty stuff that kills lots of people, it lacks many of the qualities that make spent nuclear fuel especially difficult to manage even in small amounts.

For example, a “dirty bomb” made by packing coal ash around conventional explosives would be far less effective than one made from spent nuclear fuel.

Seems to me that pro-nuclear sentiments have at least as much to do with ongoing accumulation of nuclear waste as anti-nuclear sentiments.

> Worrying about expanding nuclear and ending up putting the waste in a hole deep in the ground is such a nonissue to me.

Blithe minimization of the problem of storing nuclear waste over millenia feels like "Peak HN". :)

("Peak HN" jabs are a cheap shot, though — so let me engage more seriously...)

First, "coal vs nuclear" is a false dichotomy. Everybody I see advocating for nuclear power in this thread is advocating for it as a permanent solution rather than an interim solution — in which case there are other competitors.

Second, if nuclear waste is too dangerous for less-than-ideal storage conditions, that speaks negatively to the viability of nuclear power — because over the long term less-than-ideal storage is guaranteed by our inability to design incentive structures for responsible stewardship that persist over centuries.

I don't trust the coal industry to manage forever chemicals over the long term, and I don't trust the nuclear industry to manage spent nuclear fuel over the long term.

The question that matters for both industries is what bad things happen when their stewardship inevitably lapses and the happy path dead-ends.

I don't like either answer, so that heightens the urgency of pursuing alternatives with fewer long-lived hazardous byproducts. Neither coal nor nuclear is an acceptable long term solution.

I distrust techno-optimist promises to manage ever-growing collections of spent nuclear fuel over millennia. We can hardly trust plant operators to manage it safely over decades.

Will it actually get encased successfully, will it be stored onsite in environmentally sensitive areas because it’s too much trouble to move, will your children’s children uphold the commitments you foisted on them through the political and economic turbulence in their lifetimes, and if not what happens comparatively when those coal ash heaps and nuclear fuel dumps are left to rot…

The externalities of concentrated radioactive material are not something that our socio-economic institutions are capable of handling at scale. Tragedies of the commons are the rule and eventually all of that waste will be go through periods of mishandling at one time or another.

And perhaps meaningfully contributed to a reduction in the quantity of radioactive waste products requiring custodianship on a timescale that humans can barely conceive of let alone commit to or execute responsibly.

They don’t perceive that their principles are being violated. They asked nicely for a general explanation. How about giving them an equally nice answer in return instead of issuing a combative condemnation of their character?

I love that article, but the words "move", "fast", and "break" don't appear in it.

I agree, Claude has been pretty great at explaining code. It even does well at explaining to me old code that I wrote by hand — including both non-intuitive quirks and flaws.

Claude in its default configuration has untapped potential for explaining and documenting code because it defaults to writing no comments. I added this to my global CLAUDE.md and so far so good:

    # Comments

    Keep existing comments unless they're wrong
    or stale. Err slightly toward adding short
    explanatory comments for non-obvious code.

The default for Claude today is "no comments". Here's a conversation I had with Claude about it recently:

---

Me: I've notice that Claude is deleting my comments and tends not to insert many of its own. I don't want to go too crazy, but can we modify CLAUDE.md to dial up the comments a little? Like, try not to delete comments that are still relevant, and add a few explanatory comments.

Claude: [... proposal for modifying CLAUDE.md ...]

Claude: Want me to apply this, tweak the wording, or place it differently? Note that your global ~/.claude/CLAUDE.md-style instructions (visible to me as system guidance) currently say "default to writing no comments" — the project-level rule above will override that for this repo, which is what you want.

Me: Where is this rule that says "default to writing no comments"? That seems crazy?

Claude: That text is part of the system prompt I'm given by Claude Code itself — it's not in any file in your project or in your ~/.claude/ config. It's baked into the harness's default instructions to the model.

Claude: Specifically, two lines from my system prompt:

Claude: Under "Doing tasks": "Default to writing no comments. Only add one when the WHY is non-obvious..."

Claude: Under "Text output": "In code: default to writing no comments. Never write multi-paragraph docstrings or multi-line comment blocks — one short line max."

By default, Claude is set to avoid comments. Eventually I got tired of it deleting my own explanatory comments and overrode the behavior with an addition to CLAUDE.md.

I think Claude is just behaving like all those programmers who make a VERY BIG DEAL about how much they HATE HATE HATE comments that might reiterate what the code does so will go to extreme measures to avoid them.

My eyes aren't great and I had to pinch-zoom to read parts of this page.

I didn't either. This is an ancient debate that can never be resolved completely, though — because the articles that HN submissions point to don't follow a style guide and there are always assumptions about audience priors. Best to just resolve it and move on.

Thanks! Happy to have my comment hidden by the mods if they get around to it.

a16z = "Andreessen Horowitz", for those not in the know. (The acronym is not expanded in the article. EDIT: OP has fixed the article.)

/ponder .oO( any irrational number has "infinite nines" )

/ponder .oO( i must be one of today's lucky 10000 https://xkcd.com/1053/ )

By one metric, yes!

If you are a candidate who wants to be hired, and your target employers use LLMs to filter resumes, then an LLM-generated resume that the employer LLM-powered resume filters favor is "better" — as in "more likely to get you the job".

At some point, there will be a successful copyright infringement suit against an LLM user who redistributes infringing output generated by an LLM. It could be the NYTimes suit, or it could be another, but it's coming — after which the industry will face a Napster-style reckoning.

What comes next? Perhaps it won't be that hard to assemble a proprietary licensed corpus and get decent performance out of it. Look at all the people already willing to license their voices.

You're making me nostalgic for santorum.

https://en.wikipedia.org/wiki/Campaign_for_the_neologism_%22...

Copyright laundering is an illusion.

If the LLM generates output that a court decides is sufficiently derivative, and especially (but not necessarily) if the LLM was trained on the source material being infringed, then whoever redistributes the derivative output is going to be liable for copyright infringement.

Creation of the LLM itself is transformative, but LLM output which infringes is not.

I once thought the same about all the copyrighted works on which LLMs are currently trained. Surely they can't just hoover everything up? Haha, silly me.

I understand that creating an LLM itself is transformative, but an LLM trained on copyrighted works remains capable of generating derivative works, which eventually will result in successful copyright lawsuits against LLM users who redistribute those derivative works.

In advance of that day, the great race is to build a licensed corpus as aggressively as possible (see Github's latest decision to opt in Copilot usage). Even if Blender doesn't send your data on every save, various options can be developed, such as publishing to a Blender-controlled public channel.

Will Blender start allowing Anthropic to train on your art automatically unless you opt-out?