[go: up one dir, main page]

This is a post from Robin Sloan’s lab blog & notebook. You can visit the blog’s homepage, or learn more about me.

Science fiction

February 12, 2025

Baldur Bjar­nason blogs about my post on the foun­da­tional ques­tion of lan­guage models — to which my main reac­tion is mainly, hey, blogs still work! I remember indi­rect cor­re­spon­dence like this from the good old days.

So: I think Baldur mis­reads me a bit, but/and on bal­ance his con­sid­er­a­tion is sharp, and I’m very grateful for it. Now, there’s a part of his post that I want to emphasize.


Regarding the vision of super sci­ence empow­ered by advanced lan­guage models, Baldur writes:

This is sci­ence-fiction. There is no path that can take the cur­rent text syn­thesis models and turn them into super-sci­en­tists.

You don’t get super-sci­ence by mod­el­ling text that is itself an inac­cu­rate model of reality. That’s not how any of it works.

First: yes, it is pre­cisely sci­ence fiction. Three years ago, the vision of a fluent, for­mi­dable soft­ware cor­re­spon­dent was sci­ence fiction, too. Even if these sys­tems are flawed — indeed, even if their use is not permissible — I don’t think you can argue that their capa­bil­i­ties are any­thing other than stunning. Things com­puters could obviously — obviously! — not do, com­puters can now do.

Second: I don’t think it’s pos­sible to say, with air­less certainty, that “there is no path” from lan­guage models as we know them to super sci­en­tists or super scientist-enablers. Nei­ther Baldur nor anyone else — including the people devel­oping these models — could have pre­dicted how readily they’ve now been coaxed into general-purpose reasoners. That failure to see the path to the present ought to make us cautious, and curious, about the paths ahead.

It’s totally pos­sible these models won’t ever be useful for sci­ence, at least of the super variety; I can think of many obstacles, some having to do with phys­ical reality, others with the social process of sci­ence itself. But none of them seem to me to rise to the level of like, epis­te­mo­log­ical showstopper. I don’t know, man! If Paul Dirac can sit down and think his way into the positron … 

In my post, I laid out my counter-scenario: if, in a few years, there aren’t any signs of mean­ingful con­tri­bu­tion to sci­ence from Claude’s successors, I’ll recon­sider my position. It would be inter­esting to know if there’s any poten­tial evi­dence that might make Baldur recon­sider his.

To the blog home page