RT @emollick: Many firms built around the limitations & cost assumptions of GPT-3.5 class models, and are now stuck with complex solutions….

RT @random_walker: The origin story of “AI as Normal Technology”, and lessons learned. Many people have asked how the “AI as Normal Technol….

So exciting to see an Australian show win at the Tonys for the performance of the inestimable Sarah Snook. Go see @DorianGrayPlay on Broadway if you haven’t already (I caught it in Melbourne—10/10).

RT @emollick: The fact that Gemini Deep Research can’t read Google Books is frustrating. Google is sitting on the largest inaccessible sour….

Net, there's a lot of rhetoric, that in my mind does not bear water. There is a reason the question of the computability of the mind has stymied us for nearly a century!. I'm no philosopher, so take a grain of salt. 9/9.

On computability, the authors rely on an argument claiming non-computability of a certain process, and then disclaim a result showing that indeed that process can be simulated in lambda calculus! Dismissing it, they don't meet a burden to show what wasn't simulated. 8/9.

The paper then points to Godel incompleteness–which speaks deeply to the formal-system limits (no system can prove all true things) to claims that are quite unrelated (that biological organisms would be practically limited were they formalizable) 7/9.

Likewise, there's no reason to believe that similar precision cannot be attained for all the other external inputs to a biological system. 6/9.

Yet, our ability to construct image-processing models of arbitrary resolution (current hardware constraints not withstanding) are a direct refutation. "Algorithms" can process visual at the same numerical precision that can be sent by a biological eye. 5/9.

This points to a deeper problem: the paper conflates formalizability and computability. wrt the former, an argument is given from the apparent intractability of formalizing the inputs that an organism receives–with an implication that non-discreteness is an issue. 4/9.

The biggest underlying issue is that the work is stuck in the computer science of the 1980s!. The work assumes an ontology of computers that focusses on symbol manipulation on hand-tailored input features and well defined outputs. The rise of large models stands in contrast. 3/9.

Among its arguments: 1) reality is continuous, computers are discrete . 2) complex systems cannot be captured in the framework of Turing machines . 2a) biological systems are complex in the way that formalizing them is intractable 2/9.

Reading by @yoginho I was unhappy with its critique of computationalism. 🧵1/9.

RT @patio11: The “Huh, that’s weird” from Money Stuff in 2021 becomes the investigation in 2022 becomes the sentence in 2025.

RT @EigenGender: “the levers of power available to the executive branch are complicated to operate and require nontrivial administration ca….

RT @ben_golub: Macroeconomics is the natural predator of foolish regimes.

Not sure if to be depressed or buoyed that my broadway reviews get more reads than my academic papers. Perhaps a career change is in order?.

RT @hlntnr: Great post from Nicholas Carlini on a related theme: ."I have very wide error bars on the potential future of large language mo….

Super thoughtful piece from @daskeel with plenty to hook CS folks too!.

Life pro-tip for scientists: LLMs are quite good at LaTeX! . Saves me hours and hours of not only debugging, but initial typesetting.