re: Is there a real fear of AI or are all of the larger companies scared of open source?

quote:

---

Still hosted by the big players.

---

And? That is more of a data center issue than an AI issue.

Most companies offering AI products outside of the direct vendors are using models created by other companies. I am quite sure stout understands that.

quote:

---

We’re a long way away from companies having their own”own” AI due to cost and infrastructure restraints

---

Unless the data center buildout is similar to the cable buildout of the late 90s (leading to collapse of the product and drastically lowering prices of what was built), it will likely never be cost-effective for companies to build their own data centers. I mean the gaming industry can't even keep up.

quote:

---

it will likely never be cost-effective for companies to build their own data centers. I mean the gaming industry can't even keep up.

---

The same as it has been for just pure data storage and why every company uses Azure or AWS

it's a combo of marketing hype and the major AI companies fear mongering to get a stranglehold on the industry establish monopoly.

The biggest threat (and it's only a threat to them) is localized models. Why should I pay Anthropic when i can run LLM's on my own hardware? Well, if it get's regulated so hard it's illegal to run your own then well, wow who benefited from that?

quote:

---

I sure as frick have never seen an industry where all the leaders say their products are probably going to kill us all, and to dismiss it confidently as a novel strategy kinda seems arrogant when we have no clue what’s going on as mere observers.

---

I get what you're saying, but they are saying that it "will kill mankind" while still pushing forward with it as quickly as possible. So that is fairly telling.

It's like the politicians who warn and condemn the public about global warming and sea levels rising, yet they still fly around and use more oil/gas then public citizens and still buy up as many oceanfront properties as humanly possible.

I'm not saying those two are directly the same, but the actions from the ones in charge are similar, and ultimately undermine some of what they are saying. At least for me personally.

I think the truth is somewhere in the middle. Altman and Musk have both been beating the existential risk drum since at least 2020 or so. Then both seemed to sort of.. punt on it, favoring scaling above all else. Recently we’re seeing the pendulum swing back toward concern. I’d be less inclined to believe the concern if not for the fact that many of them voiced the same concerns before anyone really took AI seriously.

I think we’re reaching a point where the risk is becoming tangible. Maybe not to the point of existential risk yet, but legitimate cybersecurity risk at least.

I also think it’s easy to conflate different AI safety issues when talking about the big AI firms compared to local models.

Some of the issues - use of AI for biohacking/terrorism for example - apply to local models as well as hosted models. But I don’t think the concerns about recursive self-improvement, loss of control, and escape really apply to local / open-source models. Those concerns apply to models that use orders of magnitude more compute than anything locally-hosted. It’s kind of apples-to-oranges in that case.

quote:

---

they are saying that it "will kill mankind" while still pushing forward with it as quickly as possible. So that is fairly telling.

---

The scientists at Los Alamos knew they were building a weapon that could destroy mankind and kept going.

The president is on board with the continued growth of AI models and uses at least Anthropic and OpenAI for military applications.

China and other countries won't stop, even if the US-based companies do. AI is the modern nuclear race, most people just don't see it yet.

Opus 5 on max is pretty insane.

I have only dabbled in Fable some with credits, but that's legit mind-numbing.

Notebook is still the best for large projects. You can input a very large amount of data into a session. Claude is great but the 100 images limit per chat really is cumbersome.

I usually just open two separate projects to get around the annoying limit. Opus 5 is an absolute beast. It’s the first I’ve really thought our job market will eventually be in trouble.

Sacks is 100% right. They should just be responsible with their companies and leave any regulation to the public. I’m not bag holding their balance sheet or the liability of their products.

quote:

---

I get what you're saying, but they are saying that it "will kill mankind" while still pushing forward with it as quickly as possible. So that is fairly telling.

---

From an ideological standpoint, I don’t think it’s that hard to reconcile. It’s the intersection of several different pressures:
1. Concern that if the wrong person/group builds superintelligence, it could represent existential risk.
2. The idea that the first person/group to build superintelligence will hold all the cards.
3. The mantra that “if we don’t build it, someone else will.”
4. Simple market forces.

Generally speaking, I think most people believe their own intentions to be good. If you think you have good intentions, and you think superintelligence is inevitable, then it’s reasonable to believe you should be the one to build it even if you think building it in the first place is a mistake. I don’t think it’s that different from the logic of scientists on the Manhattan Project in that respect.

To be clear: I’m not really making a judgement here on whether superintelligence is truly possible/likely, whether the existential fears are justified, or whether the people at the tops of the AI firms actually *are* the right people to build it. Just pointing out where the perceived contradiction might come from.