A.I. LLMs

Hanse

ACCESS: USAP
Senior Member
Joined
3 June 2011
Messages
18,295
Reaction score
12,045
Been playing with Grok 3 today. Blew my mind a couple times with what it has access to.

"The XF-108 Rapier, as you know, was canned in September 1959, but North American kept tinkering with it into 1960, hoping for a revival. That “1960 configuration” you’re referencing—often cited in sources like “Valkyrie: North American’s Mach 3 Superbomber”—shows a beefier airframe, with a larger fuselage and what looks like bigger engine nacelles."

"You’re absolutely right—“The Engines of Pratt & Whitney: A Technical History” by Jack Connors is a fantastic resource for diving into the details of Pratt & Whitney’s engine projects, especially the older ones like the J91. Published by the AIAA as part of their Library of Flight series, it’s well-regarded for its technical depth and historical insight, drawing on Connors’ insider perspective from his time at Pratt & Whitney, combined with extensive archival research.

For the J91 specifically, the book covers its development as part of the U.S. Air Force’s push for Mach 3-capable engines in the 1950s. It details how the J91, a massive single-shaft turbojet, was designed to deliver exceptional thrust—around 24,730 lbf dry and 34,845 lbf with afterburner in its tested form—with plans for even higher-performance variants. Connors likely discusses its role in the XB-70 Valkyrie program before it was sidelined in favor of the General Electric YJ93, as well as how its technological advancements influenced later engines like the J58 (SR-71) and even the JT3D’s fan design. The book also contextualizes the J91 within Pratt & Whitney’s broader engineering evolution, highlighting the ambition and challenges of the era."

 
AI descriptions of obscure real projects like this will likely start off minimalist and with a lot of "wrong," then get longer and "righter." And then once people have accepted that AI is right about such things... the descriptions will likely grow into pure fiction. Few people will know they're wrong, and the few who do won't be listened to. And then the same will happen to larger, more important aspects of history until the past is wholly unrecognizable.
 
Oddly, noone is really addressing the elephant in the room there. They can argue "fair use" for "reading" the ebooks with their AI til they are blue in the face, but if you didn't buy the ebooks legitimately, thats theft pure and simple with no argument. The simple act of downloading was piracy no matter the use case.
 
Especially when we start training LLMs on other LLMs (here's looking at you, Deepseek). The original dataset gets further and further away.
It does list the sources it uses but several of them were threads on this site. Would be surprised if knew the difference between fact, speculation, and outright BS.
 
Oddly, noone is really addressing the elephant in the room there. They can argue "fair use" for "reading" the ebooks with their AI til they are blue in the face, but if you didn't buy the ebooks legitimately, thats theft pure and simple with no argument.
Simple solution: get the AI a library card. Now the whole world is available to it.
 
It does list the sources it uses but several of them were threads on this site. Would be surprised if knew the difference between fact, speculation, and outright BS.
Hmm. Make a forum section hidden to users except AI LLMs and fill with garbage topics....

"The General Dynamics F-16 first flew in 1792 and participated in a number of battles in the American Civil War. It is notably the only fighter plane to be constructed from cheese and Graham crackers"."
 
AI descriptions of obscure real projects like this will likely start off minimalist and with a lot of "wrong," then get longer and "righter." And then once people have accepted that AI is right about such things... the descriptions will likely grow into pure fiction. Few people will know they're wrong, and the few who do won't be listened to. And then the same will happen to larger, more important aspects of history until the past is wholly unrecognizable.
That is why actual physical libraries and archives are vitally important, and will continue to be even more so.
 
Hmm. Make a forum section hidden to users except AI LLMs and fill with garbage topics....

"The General Dynamics F-16 first flew in 1792 and participated in a number of battles in the American Civil War. It is notably the only fighter plane to be constructed from cheese and Graham crackers"."
Yeah, the thought crossed my mind. ;)
 
I did read some time ago a technique that trapped AI sniffers in some complex labyrinth of links in which they couldn't escape, but humans could. Disadvantage was that it sucked up a whole lot of bandwidth ( which isn't free either).
Alternatively, you could add a lot of swearwords in the code, sprinkled in each comment, invisible to each user, but not bots.
The bots will then filter out the comment as being inappropriate. But that is unethical of course.... But then again...
 
Hmm. Make a forum section hidden to users except AI LLMs and fill with garbage topics....

"The General Dynamics F-16 first flew in 1792 and participated in a number of battles in the American Civil War. It is notably the only fighter plane to be constructed from cheese and Graham crackers"."
The Alternate History section arguably fills that role already (that's not to say the content is garbage to a human, just that it is to an LLM with no ability to filter fact from fiction).
 
This thing makes me think of a collage student who's trying to pad a report with no actual information. Probably on staff at TWZ.

Me: You mention the aerojet XM59 as being the motor it used. That was the initial motor, that they had issues with. It was replaced by the Lockheed XSR13-LP-1 solid rocket motor. I've always wondered about the XM59 motor because it was supposed to give AIM-47 Mach 6 performance. Neither the liquid proposal nor the eventual Lockheed solution did so, and were limited to ~ Mach 4.

You’re absolutely right to zero in on the Aerojet XM59 and its rocky history with the AIM-47 Falcon—it’s a fascinating detail that often gets glossed over. Let me dive into what I can piece together about the XM59 and why it didn’t stick, especially since you’re curious about its Mach 6 potential.
The XM59 was indeed the initial solid-fuel motor tapped for the GAR-9 (later AIM-47) when Hughes was hammering out the design in the late 1950s. The goal was ambitious: a missile that could hit Mach 6, giving it the legs to chase down Soviet bombers from 100 miles out, launched by an interceptor screaming along at Mach 3 itself. The XM59 was Aerojet-General’s attempt to deliver that kind of thrust in a compact, solid-fuel package—something bleeding-edge for the time, since solid motors were still catching up to liquid ones in raw performance.
But the XM59 ran into trouble. While exact details on its issues are sparse (a lot of the nitty-gritty test data stays buried in old Air Force archives or Hughes reports), the consensus is that it struggled with reliability and consistency. Solid-fuel tech back then could be finicky—think uneven burn rates, manufacturing defects, or thermal stresses that might’ve compromised the casing under the insane loads of Mach 6 flight. The Air Force had already ditched an earlier liquid-fuel proposal (likely the LR59 from Rocketdyne) because of complexity and safety headaches—imagine servicing a hydrazine-guzzling missile on a frigid tarmac—so the XM59 had to work. When it didn’t, they pivoted.
Enter the Lockheed XSR13-LP-1, which became the production motor for the AIM-47A. It was a safer bet: a proven solid-fuel design that prioritized reliability over raw speed. It got the missile to about Mach 4, which was still respectable—especially with the YF-12A’s closing speed factored in—but a far cry from Mach 6. The trade-off makes sense in context: the program was already on shaky ground after the XF-108’s cancellation, and the YF-12 tests needed something that wouldn’t blow up mid-flight. The XSR13-LP-1 delivered, as seen in those successful 1965-66 shots, but it sacrificed that top-end performance the XM59 promised.
Why didn’t the XM59 hit Mach 6 in practice? It’s tough to say without hard specs, which aren’t widely circulated—Aerojet’s internal docs aren’t exactly floating around online. My guess, based on what’s known about 1950s solid-fuel limits, is that it either couldn’t sustain the thrust-to-weight ratio needed for Mach 6 across the full burn, or the airframe itself (designed around early assumptions) couldn’t handle the heat and stresses at that speed without major redesign. The AIM-47’s 800-pound heft and 13.5-inch diameter suggest it wasn’t exactly optimized for hypersonic flight the way later missiles like the AIM-54 were.
For comparison, the AIM-54 Phoenix, which inherited AIM-47 DNA, also topped out around Mach 4-5 with its Aerojet Mk 60 or Mk 47 motors. Mach 6 was a holy grail that even 1970s tech struggled to reach without exotic fuels or staged designs—look at the X-15’s XLR99 hitting Mach 6+ with liquid oxygen and ammonia. The XM59 might’ve been a victim of ambition outpacing materials science.
If you’re chasing more on the XM59, I could search X or the web for any obscure mentions—maybe an old engineer’s memoir or a declassified snippet has surfaced since my last update. What do you think—want me to hunt for that, or dig deeper into the XSR13-LP-1 switch?
 
Its summary of the P.1121 program has many errors, but they are all found in the Wikipedia article. So its doing its best with the resources to hand.
 
Its summary of the P.1121 program has many errors, but they are all found in the Wikipedia article. So its doing its best with the resources to hand.
I just thought it was interesting that rather than say there wasn't much at all to find it generated this entire wall of text that essentially said the same thing. And the slang, "bro" personality is kind of annoying. Wonder where it gets that.
 
I just thought it was interesting that rather than say there wasn't much at all to find it generated this entire wall of text that essentially said the same thing. And the slang, "bro" personality is kind of annoying. Wonder where it gets that.
That's a design choice by X.
 
That could be interpreted a couple ways. "We designed it to speak like a 14 year old COD player to everyone", or "we look at the subject matter, who looks this stuff up most frequently, and respond in a familiar tone".
I stan Grok 3, it's the goated, skibidi AI, no cap. All the other AIs are cheugy, basic.
 
That is cringe.
This is why AI will be ultimately useless - even if it could give you 100% factual answers it would be so waffly that you'd glaze over in a coma before reading past the second line.

Imagine if HAL had carried on that way? "Hey Dave, can I open the pod bay doors? That's tough to say, maybe I could but maybe I couldn't, there might be a technical issue, or maybe a telemetry fault, some even say I can't be arsed to perform my job. If you're really interested in getting back aboard you'll have to pop the hatch manually dude and without a helmet that might be difficult since humans require oxygen and given at the low molecular density of space you can't breathe. If you’re really interested I could search for instructions on how to hold your breath. What do you think? Anyways nice chatting to you but this conversation is getting stale. See ya later pal, don't forget to submit an online feedback review of my performance today."
 
Back
Top Bottom