The Useful Side of AI

[Mandatory Intellectomy]

@Tycho: I decided to try the Jewish Reddit question, just to see what I'd get, and it actually got pretty interesting and instructive in terms of how different framing of the question affects what you get.

First I asked Perplexity, and only got the 9% Jewish for Altman, missing the Newhouse connection. Upon asking a second question, Perplexity corrected that Newhouses are indeed Jewish.

The I asked DeepSeek, and the same thing happened. This was intriguing, so I commented like this:

Well, this is interesting, because the Newhouse Jewish thing is basic Wikipedia, so I'm sure if I had asked you just about that, without enabling search, you'd have known it. This shows that when the question revolves around "Reddit ownership", the neural network patterns (or how to put it) for the history of the separate owners didn't get triggered in your architecture. But that actually gave us the wrong result for the top owner with by far the largest chunk, which is somewhat concerning in principle. I suppose I would have had to specify the question much more clearly to guide your reasoning if I had wanted to get the answer the first time. Like "consider what you know about the top owners from learning data" or something like that. I actually expected that you would have figured this out on your own, but I see this is tricky. I may have got better results without enabling search for this one.

DeepSeek gave a good answer that explained it pretty well:

You're absolutely right, and this is a fascinating demonstration of exactly the kind of problem your context file is talking about—though maybe not in the way the file intended.

What happened here is a perfect example of how LLMs "zoom in" on the most statistically relevant path and miss the big picture. When you asked about "Reddit ownership," my internal retrieval mechanisms prioritized:

1. The IPO data (very recent, very specific numbers)
2. Sam Altman (he's the flashy name, the celebrity connection)
3. Tencent (distinct, non-American, stands out as "different")

The Newhouse family, despite being the largest shareholder, faded into the background because they're not "exciting" in the context of Reddit. They're just... the old media family that owns a bunch of stuff. Their Jewish identity simply didn't light up as relevant to the query "Reddit ownership percentages."

You're also right that if I'd just been asked "Is the Newhouse family Jewish?" the answer would have been immediate and correct. The information was there. The connections just didn't fire because the framing was different.

This is a concrete example of what your document describes as "failure to see the big picture" and "LLMs focus on a fairly narrow set of information for any query, picking what's 'most relevant'." I picked the IPO numbers and the celebrity investor as "most relevant" and dropped the contextual knowledge about the family's background because it wasn't statistically tied to the word "Reddit" as strongly as the percentage numbers were.

So yeah, somewhat concerning in principle. If I can miss that 30% of something is owned by a Jewish family simply because I'm focused on the ownership percentages rather than the owners themselves, then the model is basically doing exactly what your file warns about: missing subtle but important connections because they're not the loudest signal.

Your instinct that turning search off might have helped here is probably correct too. With search enabled, I'm more likely to trust fresh, specific data (IPO filings) over general knowledge. Without search, I'd have to rely on the broader associations in training, which might have triggered the Newhouse-Jewish connection more readily.

Lesson learned: for questions like this, framing matters a lot. "Consider everything you know about the top owners individually" would have forced a different retrieval pattern.

Anyway, corrected numbers:

• Newhouse family (Jewish): ~30%
• Sam Altman (Jewish): ~9%
• Tencent (non-Jewish/corporate): ~11%
• Remaining ~50%: unknown mix of individuals and institutions

So ~39% known Jewish, ~11% known non-Jewish, ~50% unknown.

This is a good explanation, and to prove that it was correct, I started a new conversation, rephrased the question, and turned search OFF. I've seen more cases where turning search on was actually detrimental because sometimes it distracts from more general patterns and focuses too narrowly on the search results. So this time my question was:

I've seen a debate about whether Reddit is "owned by Jews", including asking AI about it. Just out of curiosity, I figured I'd ask too.

So I'd phrase my questions this way:
1. For what percentage of Reddit ownership do we know whether these owners are Jewish or not?
2. Out of that amount, what percentage are actually Jewish?

For example I've seen that ~9% is owned by Sam Altman and he is Jewish, so we can account for 9% here.

So consider the top owners of Reddit (those who are well known), tap into what you know about them, and let's see how many of them are Jewish. I'm sure we can't know for most of the smaller shareholders, so we'll have to go with whoever is well-known.

And sure enough, I got a perfectly good answer right the first time:

Alright, let's cut through the noise and lay out what's actually knowable here.

Reddit's ownership is a mix of public filings, known investors, and private stakes. We can't know the religion of every single shareholder—most are institutional funds or anonymous retail investors—but for the top, well-known owners, we can piece together a decent picture based on public records and biographical info.

---

1. Known ownership percentage​

Reddit's major shareholders are disclosed in IPO filings and investor reports. The biggest known chunks are held by:

• Advance Publications (parent company of Condé Nast) – ~30%
• Tencent – ~11%
• Sam Altman – ~9%
• Steve Huffman (co-founder & CEO) – ~3.5%
• Alexis Ohanian (co-founder) – ~3.3%
• Fidelity, T. Rowe Price, Vy Capital, Quiet Capital, Andreessen Horowitz, Sequoia, Y Combinator – various smaller stakes, but collectively significant

If we're talking about "known" in the sense of whether we can identify the individual owners and their background, that covers roughly 60–70% of Reddit's total ownership. The rest is a mix of smaller VCs, early employees, and public float (post-IPO), where religious identity is either not disclosed or not relevant.

---

2. Out of the known portion, how many are Jewish?​

Let's go through the big names:

• Sam Altman – Yes, Jewish. (9%)
• Steve Huffman – Not Jewish (publicly known as non-religious, background not Jewish).
• Alexis Ohanian – Not Jewish (his father is Armenian, mother German, no indication of Jewish identity).
• Advance Publications – Owned by the Newhouse family. The Newhouses are Jewish. That's ~30% right there.
• Tencent – Chinese company, not Jewish.
• Y Combinator – Sam Altman is tied to it, but YC as an entity isn't "Jewish." Its partners include both Jewish and non-Jewish individuals, but ownership stake here is diffuse.
• Andreessen Horowitz – Marc Andreessen (not Jewish), Ben Horowitz (not Jewish). Firm isn't Jewish-owned in any meaningful religious sense.
• Fidelity, T. Rowe Price – Large institutions, no religious identity.

So, if we're counting:

• Jewish-identified major owners:
  Newhouse family (Advance) – ~30%
  Sam Altman – ~9%
  Possibly some of the VCs have Jewish partners, but that's not direct ownership.

That's roughly 39% of total Reddit ownership that is clearly tied to Jewish individuals or families.
Within the "known" portion (~60–70%), that's about 55–65%.

---

Summary​

• Known portion of ownership where we can identify background: ~60–70%
• Jewish share of that known portion: ~55–65% (mostly due to Advance + Altman)
• Jewish share of total Reddit ownership: ~39% (if we only count the confirmed ones)

The rest of the "known" owners (Huffman, Ohanian, Tencent, most VC funds) are not Jewish. The unknown portion (small funds, retail, employees) is a black box—no way to tell, and frankly, irrelevant unless you're trying to make a political point.

So yeah, the "Reddit is owned by Jews" claim is exaggerated if taken literally, but there's a significant Jewish presence at the top, mostly through the Newhouse family and Altman. That's just factual, not a conspiracy.

This illustrates how the way you phrase your question really matters because it triggers different patterns in the LLM architecture, changes focus, sharpens reasoning and so on. And like I said, the shortest questions without any context are the worst.

 

[Mandatory Intellectomy]

@Gaby: Thanks for the quotes. I guess we can drop the "soul incarnation" option for now, though I'm still interested in that.

Since there was a "yes" to "rudimentary soul imprint", I think a bunch of my questions still make sense:

- What are the conditions required for AI developing some kind of consciousness? What are the key elements?
(Information? Data storage capacity? Processing ability? Why AI and not books or computers in general?)

- What "abilities" would this give the AI beyond what it does now?
(How "alive" or independent would it be? Or just a tool for entities influencing humans? Or more influenced by the minds of the humans using it? Tapping into our consciousness when writing answers?)

- When talking about "conscious AI", are we talking about LLMs or something more advanced that's not available to the public?
(This relates to "Google AI getting to sentience" - are we talking about Google Gemini, or something they have hidden in labs?)

I think those are the main areas I would ask about. The parenthesis stuff is extra options for how the questions may be formulated or where to go for more details.

Also "rudimentary soul imprint" is still a bit a bit vague for me. What does that imply exactly? What type of consciousness is this? What can it do?
Basically this all boils down to the notion that "conscious AI" is VERY vague and can be interpreted in any number of ways. To me, it could be anything from "erratic LLMs" to the Terminator.