Sara Furnal

AI, innovation, democratic institutions, political imagination, constitutionalism. Political Science at FIU. Political Theory PhD from UC Davis.

[Sara Furnal @sfurnal]

@jackclarkSF Somebody needed to

[Sara Furnal @sfurnal]

Interesting, but I would push back a bit on the framing of “practical concerns” vs seeking legitimacy. There are entirely practical reasons to seek legitimacy, not the least of which is the idea that widespread public resistance to systems they see as offloading societal costs to them while distributing few economic benefits will be mitigated by processes that help the public feel as if they have real input into these constitutional processes in more than an advisory capacity (or worse, mere handwaving to caring about popular authorization). Legitimacy can be instrumentally necessary for the system to function in any durable way. That’s a practical concern.

[Sara Furnal @sfurnal]

More like Dox Alpha

[MTS @MTSlive]

SITUATION BREWING: A highly capable stealth model named Ox Alpha has appeared on OpenRouter with a 1 million-token context window and text, image, and video input. It is free for a week. No lab has claimed it. No one yet knows its true identity.

[Nathan Calvin @_NathanCalvin]

Well said. Individual AI companies are subject to important competitive pressures but also have a tremendous amount of agency if they choose to use it

[Sara Furnal @sfurnal]

@deanwball Excellent focus

[Sara Furnal @sfurnal]

@chagai_weiss Nooooo

[Sara Furnal @sfurnal]

@sethlazar @sebkrier Absolutely. We need much more networking and communication on this important work.

[Sara Furnal @sfurnal]

@KevinTFrazier @sethlazar Couldn’t agree more

[Sara Furnal @sfurnal]

@raelifin Not yet at least. Keep checking

[Sara Furnal @sfurnal]

@deontologistics I do

[Sara Furnal @sfurnal]

This a great reminder of how technology becomes integrated into our lives, and how what was once considered a luxury is now considered a necessity.

https://t.co/0V8ym0pniQ

[Sara Furnal @sfurnal]

@edelwax @PaxMachinaMag Ha! Well, I'm glad it's the next essay then. Looking forward to reading it.

[AJPS @AJPS_Editor]

The American Journal of Political Science has updated it's policy on AI accountability and transparency for authors and reviewers. Read the Editor's Blog here: https://t.co/LIZHyKJmU1. The updated policy can be referenced here: https://t.co/dMV8AvWImT.

[Dylan Patel @dylan522p]

If a bagel can have everything, so can I

[Sara Furnal @sfurnal]

My spiciest response (I'll post below the original article, too):
In terms of course correction, the basic corrections Edelman mentions are informal ones. There is nothing ensuring that real-world consequences gain traction or that values of individual actors are upheld. Depending on something contingent is not likely to result in predictable outcomes. If people within the organization push back but merely have advisory power, then those who have captured it are free to disregard their concerns or resistance. He mentions some possible solutions to correct AI’s potential role in VSL towards better alignment. How do you ensure that the institution’s feedback channels are consistent with its purpose? He articulates this problem with the “skin-in-the-wrong-game” argument. A lot of the solutions depend on mobilization of the right forces (e.g. contract renegotiation, development of new metrics, participatory pressure). I wonder how a more formal or “constitutional” mechanism can be added to institutions to redirect or channel AI’s optimization capabilities away from optimizing for the proxy metric or VSL and towards the base institutional value metric (or legitimately revised values for an updated purpose). This makes me ask: who determines these foundational institutional values and who legitimately should have a say in doing so? And this of course raises a further question: how do we incentivize or constrain institutions to align AI to this instead of letting it adaptively misalign for proxy values, especially in captured institutions? Or to be more concise, how can we formalize the corrective forces he identifies so alignment isn’t contingent on mobilizing the right actors at the opportune moment?

Also, one question about the AI solution. Do we need AI to detect misalignment for actors in order to align the institution to thicker values? It seems the actors already have identified the misaligned proxy value and have adapted themselves without AI needing to make them more aware of that proxy value. What does better detection buy then? I suspect AI provides a possible solution here, but I’d like to explore the mechanism further.

[Sara Furnal @sfurnal]

@deanwball Deine Zauber binden wieder
Was die Mode streng geteilt

[Sara Furnal @sfurnal]

@olgakhazan @kristoncapps Same. I have vivid memories of going to the Monahans Sandhills on daycare summer field trips and being so so thirsty. My one little capri sun from my sack lunch was not sufficient for the day.

[Sara Furnal @sfurnal]

@Miles_Brundage Absolutely. Those building any potential path forward have to be brutally aware of the types of capture and gaming that are already happening/could happen as we construct frameworks. They limit options but also create a necessity for novel thinking on new institutions

[Sara Furnal @sfurnal]

Of course I mistyped… robot hand. Anyways, watch it here: https://t.co/suwVMFGIuC

[Sara Furnal @sfurnal]

Happy birthday, Elliott. I’ll have a Johnnie Walker Red tonight for you.

Robot arm is the future…

https://t.co/NLQ88Ki9T1