If you spend time with the higher-tier (paid) reasoning models, you’ll see they already operate in ways that are effectively deductive (i.e., behaviorally indistinguishable) within the bounds of where they operate well. So not novel theorem proving. But give them scheduling constraints, warranty/return policies, travel planning, or system troubleshooting, and they’ll parse the conditions, decompose the problem, and run through intermediate steps until they land on the right conclusion. That’s not "just chained prediction". It’s structured reasoning that, in practice, outperforms what a lot of humans can do effectively.

When the domain is checkable (e.g., dates, constraints, algebraic rewrites, SAT-style logic), the outputs are effectively indistinguishable from human deduction. Outside those domains, yes it drifts into probabilistic inference or “reading between the lines.” But to dismiss it all as “not deduction at all” ignores how far beyond surface-level token prediction the good models already are. If you want to dismiss all that by saying “but it’s just prediction,” you’re basically saying deduction doesn’t count unless it’s done by a human. That’s just redefining words to try and win an Internet argument.

They do quite a bit more than that. There's a good bit of reasoning that comes into play and newer models (really beginning with o3 on the ChatGPT side) can do multi-step reasoning where it'll first determine what the user is actually seeking, then determine what it needs to provide that, then begin the process of response generation based on all of that.

This is not a surprise, just one more data point that LLMs fundamentally suck and cannot be trusted.

Huh? LLMs are not perfect and are not expert-level in every single thing ever. But that doesn't mean they suck. Nothing does everything. A great LLM can fail to produce a perfect original proof but still be excellent at helping people adjust the tone of their writing or understanding interactions with others or developing communication skills, developing coping skills, or learning new subjects quickly. I've used ChatGPT for everything from landscaping to plumbing successfully. Right now it's helping to guide my diet, tracking macros and suggesting strategies and recipes to remain on target.

LLMs are a tool with use cases where they work well and use cases where they don't. They actually have a very wide set of use cases. A hammer doesn't suck just because I can't use it to cut my grass. That's not a use case where it excels. But a hammer is a perfect tool for hammering nails into wood and it's pretty decent at putting holes in drywall. Let's not throw out LLMs just because they don't do everything everywhere perfectly at all times. They're a brand new novel tool that's suddenly been put into millions of peoples' hands. And it's been massively improved over the past few years to expand its usefulness. But it's still just a tool.

This is what they should do in response to the WH request.

He's probably taking a cue from someone who might say something like this:

"The best people, the smartest people, are known. That kind of intelligence is needed to make America great. The economy and the stock market are all about being smart, having an incredible brain. I can tell you, there's a great brain, a tremendous brain. Everybody knows it, everybody says it. And you know, I went to the best schools, and was a very smart person. So we're going to keep bringing in smart people, the smartest people, to get things done, and make America win again. Nobody does it better, believe me. Nobody."

(This is an amusing AI fabrication.)

This seems a little bit more like the towel trick for the Xbox 360.

Yeah, I don't see any problems with that.

> Not sure how well geofences work in a GNSS denied environment.

If you shield your GPS antenna to only see signals from above, it will make spoofing rather difficult, unless the spoofer can afford constant aerial missions in just the right places.
If you want to get really fancy, use phased-array antennas that can "aim" at the known satellites and ignore other sources.

Keep your public keys private!

> processing speeds in the petahertz range — over 1,000 times faster than modern computer chips

I'm not aware of any computer chips that operate in the terahertz range.

In the first line, it mentions a factor of 1 million, which makes sense, since current chips operate in the gigahertz range.

If/when true AGI is achieved, only a fool would announce it. What would announcing it do for you? Make you famous? Rich? Cool. Know what's better than all that?

Not telling a damn soul and using the AGI quietly to do whatever the Hell you want. If you want to be rich, the AGI will tell you how to become rich. If you want to be famous, the AGI will tell you how to become famous. You can do both. And you don't have to stop there. A real, vastly superior AGI enables the person controlling it to do anything. The second you tell people about it, you'll lose control over it and then you're the famous idiot who did a cool thing one time. Kids in elementary school will recite your name back on a test. And you could have had everything.

Anyone smart enough to crack AGI can't also be stupid enough to advertise when they do it.

> AI lies with its hallucinations. And does not like to be told it's wrong.

You know, if you substitute in certain politicians names for "AI", the sentence is still completely correct.
Unfortunately, the politicians are immune to swearing, unless it comes from major donors.

I read it as 4.5 didn't end up performing as well as hoped in the real world and 4.1 being a direct iteration of 4o rather than 4.5.

They're operating with parallel code lines because each is targeting different use cases.

Now they just need (to make) more kids to buy them.

You can invoke Technology Connections' special trick: the magic of buying two of them.