How would you even use slaves in drilling a kilometers-deep borehole?

Yep, the guardrails are also improving.

Well, the LLMs don't really consist of "code" per se, but I think the AI labs are already using them to work on improving their own design. How far are they from being able to do this without human oversight and supervision? I have no idea.

In this case, it stays deep in the ground, where it came from.

The dry rock method described in the article significantly reduces this problem, because it doesn't rely on groundwater steam that has had millennia to dissolve high mineral loads. Instead, it injects low-mineral surface water into pressure-created cracks. That water does pick up some minerals from the rocks, of course, but the result is far less corrosive than natural groundwater. In addition, super-hot rocks flash all of the water to steam, and H20 in gaseous form cannot carry any dissolved minerals (this is how distillation works), so as long as they can find ways to keep the steam from pushing liquid water up into the generation equipment, there are no salts or minerals to cause problems. This is actually easier with higher temperatures, because no pockets are cool enough to avoid flashing to steam.

I'm not saying that the problem you cite isn't relevant to hot rock geothermal, but the difficulties are hugely reduced.

Grr. "you're", not "your". If only there were a way to preview /. posts...

What sources? And when? I think the conclusions your citing predate the reasoning overlay models... and reasoning seems like exactly the right solution to both of the problems you cite.

I think you're fooling yourself. Confirmation bias at work.

This is a common perception among techies, but is not true. Google search usage and profitability has been rising steadily. It's actually still rising, though LLM usage is slowing it. I'm not sure whether the numbers I saw were public, so I won't cite them. If you'd like to verify, I suggesting looking at the annual reports.

Which draws the AC's like flies. Thanks for playing.

Say hi to Xi for me.

Why? SIMPLE. USA is SPREAD OUT! Japan is smaller than California in land area. That is just ONE state, not to mention the other 47 in the continental USA. You can travel from France to Germany by car in a day, but to travel from El Paso to Texarkana takes a couple days by car. Unless you develop "warp drive" for trains, it just won't work in the USA.

I think they're a bit more than that, but assume you're right... have you considered that they're less than three years old? ChatGPT launched November 30, 2022. And the reasoning models that have made them massively more effective in many areas (especially software development) are barely a year old?

If you reason about what will happen in the next decade or two based on where the technology is right now, a technology that didn't even exist five years ago and is still obviously in its infancy, you're clearly missing the most important point, which is that the pace of improvement has been and continues to be incredibly rapid. You need to base your reasoning on what the models will be capable of five years from now, ten years from now. Unless we suddenly hit a wall, they'll be vastly better. How much better? No one knows, but it seems safe to expect that they'll be orders of magnitude better.

I'd say they still haven't fixed their mistake, which was to create a shitty web site. A good web catalog will be far superior to any paper catalog, providing multiple ways to find a part, having real-time information about where the part is located among the retail stores, warehouses and suppliers, providing links to datasheets, installation guides, and lots more.

If Jaycar gets a competitor that builds a good web site, they'll go out of business. The fact that they don't have much competition has saved them so far, but they've responded by going the wrong direction.

No, they can't. They're already seeing searches decline as people move to asking LLMs instead. The LLMs actually end up using Google, but that doesn't generate any ad revenue for Google (that's something Google may have to figure out how to put a stop to). The solution so far is the addition of "AI results" to search output... but those AI results are produced by a far inferior model because Google search gets hundreds of thousands of queries per second, and it's currently infeasible to service those queries with a full-powered model. But the output of the inferior model doesn't satisfy users, so many of them are shifting to ChatGPT or Claude (I am!). If users opt for paid subscriptions to LLMs instead of free, ad-supported search, fine, but Google needs to be in that game and to get them to use Gemini rather than the competitors. For users who won't pay for a subscription, I assume that ad-supported LLM usage will become a thing, and Google clearly wants to get there first, or at least early enough. But right now they don't actually have the capacity and LLM usage is too expensive for an ad-supported model.

So... they need 1000X capacity, and they need it at something not too far above current OPEX costs. If it requires massive CAPEX, that's less problematic as long as it can be funded from revenues (including future revenues), and in fact it's actually good for Google if massive CAPEX is required, because it helps to build their moat, protecting them from competition by other companies that don't have the same resources. But OPEX can't be 1000X, it probably can't even be 10X.