> Would we? What for? Why would we need reams of iPads on our desks?
To use like we'd use paper.
> Which LLMs do not. They fake it really well but it’s still an illusion. No understanding is going on, they don’t really know what you mean
That is very much up to debate at this point. But for practical purposes in context described here, they do.
> and don’t know what the right answer is.
They're not supposed to. This is LLM use 101 - the model itself is behaving much like a person's inner monologue, or like a person who just speaks their thoughts out loud, without filtering. It's very much not a database lookup.
> I also disagree that was an underappreciated featured of the Star Trek computers, it’s one of their defining features.
What I meant is, people remember and refer to Star Trek's ship computer for its ability to control music, lights or shoot weapons, etc. with voice commands. People noticed the generality, the shamelessness of interaction, lack of structured command language - but rarely I saw anyone paying deeper attention to the latter, enough to realize the subtle magic that made it work on the show. It wasn't just some fuzzy matching allowing for synonyms and filler words, but more human-like understanding of the language.
(Related observation: if you pay attention to sliding doors on Star Trek vs. reality, you eventually realize that Starfleet doesn't just put a 24th century PIR into the door frame; for it to work like it does on the show, the computer has to track approaching people and predict, in real time, whether or not they want to walk through the door, vs, just passing by, or standing next to them, etc. That's another subtle detail that turns into general AI-level challenge.)
> The ship’s computer on Star Trek could run diagnostics on itself, the ships, strange life forms and even strange pieces of technology.
That's obviously tool calls :). I don't get where this assumption comes from, that a computer must be a single, uniform blob of compute? It's probably because people think people are like this, but in fact, even our brains have function-specific hardware components.
(I do imagine the scans involve a lot of machine learning and sensor fusion, though. That's actually how "life signs" can stop being a bullshit shorthand.)
> The most advanced LLMs frequently fail at even identifying themselves.
They'll stop when run with a "who am I?" tool.
> When the Start Trek computers behaved inconsistently like that, they would (rightly) be considered to be malfunctioning. Yet you’re defending this monumental gap as being effectively the same thing. Gene Roddenberry must be spinning in his grave.
All I'm saying is, LLMs solved the "understand natural language" problem, which solves the language and intent recognition part of Star Trek voice interfaces (and obviously a host of other aspects of computer's tasks that require dealing with semantics). Obviously, they're a very new development and have tons of issues that need solving, but I'm claiming the qualitative breakthrough already happened.
Obviously, Star Trek's computer isn't just one big LLM. That would be a stupid design.
> To use like we'd use paper.
How we use paper derives not only from our own practical needs, but also from the intrinsic limitations of paper. Stacks of paper are used because it's not possible to put several pages worth of text onto a single page of paper while maintaining a legible font size. The idiosyncratic way that tablets were used in Star Trek isn't how people would actually do things, it merely reflects the limitations of the writers to imagine all of the practical implications of technology such as they were depicting. It would be like somebody in the 1800s speculating about motor vehicles, supposing that teams of a dozen or more motor vehicles might be connected using ropes and used to tow a single carriage, because that's how they did it with horses.
> To use like we'd use horses.
> Stacks of paper are used because it's not possible to put several pages worth of text onto a single page of paper while maintaining a legible font size.
Right. And trying to replace a stack of paper with one paper sheet-sized screen is a significant downgrade. Which is why tablets are used primarily for entertainment, not for work.
Having lots of sheets of paper you can spread out around you is an advantage, not a limitation, of the paper-based workflow.
No, a single screen is a massive upgrade over using stacks of paper.
People vastly prefer digital dictionaries over paper dictionaries because you can more quickly find stuff. And that’s with dictionaries in alphabetical order.
Stacks of paper suck, there’s some potential utility in a space ship for all the redundancy around independent tablets you can hand someone. That’s something that regularly happens on the show and kind of makes sense, but is more a visual reference for the audience. Which is where stacks of tablets shine, the viewer can easily follow what their doing even if you can’t see the screen.
It can be true that stacks of paper are better than a single screen in some ways and worse in others. Other people like to be able to spread out multiple sheets of paper in front of them, even if you do not. You are correct that digital search is a huge plus of having a digital interface.
If we’re talking 3-5 pieces of one sided paper for say homework you can spread them out nicely, but scale that to multiple stacks of loose paper and it invariably becomes a mess.
Thus, in practice almost everyone is using multiple screens at work when they can even if printing stuff is trivial.
> They're not supposed to. This is LLM use 101
> (…)
> Obviously, Star Trek's computer isn't just one big LLM. That would be a stupid design.
Or, in other words, we don’t have Star Trek’s computer like originally claimed, and our current closest solution isn’t the way to get it.
Your takeaway presumes that all LLM interactions are monolithic, which is the opposite of what was being claimed if you take the other poster's comments about tool use into consideration. I have no real investment in this conversation though, so your proclamation of winning can stand as far as I'm concerned.
> Or, in other words, we don’t have Star Trek’s computer like originally claimed, and our current closest solution isn’t the way to get it.
My computer can both run an LLM (albeit a bad one, only has 16 GB of RAM) and also run other things at the same time.
> That's obviously tool calls :). I don't get where this assumption comes from, that a computer must be a single, uniform blob of compute? It's probably because people think people are like this, but in fact, even our brains have function-specific hardware components.
In fairness, half the time the Trek computer does something weird, it only makes sense if there's no memory/process isolation and it's all one uniform blob of compute. Made sense in the 60s where Spock's chess app losing to him was useful evidence that the CCTV recordings had been faked, not so sensible in 2025 when the ship stops being able to navigate due to the excess system demand from the experimental holodeck.