RT @MarisaVeryMoe:

@rickasaurus @swyx No. It's just a paid API aggregator. IIUC, Deep Research puts o3 in a loop with some extra APIs to grab information, but it hasn't been released as an API itself yet.

@rickasaurus @swyx Poe is pretty useful for varied usage, I.e. just dabbling in everything.

@rickasaurus An LLM owns TED.

@jdegoes It seems Parkinson's law applies to every resource - time, attention, mental complexity budgets, desktop surfaces, and AI resources.

@Ngnghm @rizkidotme @BartoszMilewski @headinthebox I want FP and OO without the state-code entanglements of first-class functions or objects. (Because such entanglements are awkward for live coding, orthogonal persistence, distribution, etc.) Really has me thinking what it means to compose code and extend namespaces.

@dearmadisonblue Neel's argument relies on type equivocation, distinct meanings for (unit->bool) for TM vs LC. Thus far, no argument has been presented that these models compute distinct functions under observably equivalent input environments. To disprove the thesis would require exactly that.

@dearmadisonblue IIUC, you start with non-equivalent HoF input environments for TM vs LC. Then you used the simulation as a means of voluntarily restricting observations on HoF inputs to the TM. But this isn't demonstrating substrate dependence. It's really just an approach to a type adapter.

@dearmadisonblue It is unremarkable that we can compute different functions in context of observably distinct inputs. So to discuss equivalence of computable functions, we first hold inputs equivalent. Of some note, features such as sequences of 'random choices' can be viewed as implicit inputs.

@dearmadisonblue Similarly, for function outputs from the TM, it is sufficient to take the type description as a limitation or promise on what will be observed by the recipient. The TM only needs a convention for encoding HoF behavior, aka a language. This might reference inputs by index.

@dearmadisonblue I'll suggest instead that it's fundamental to the notion of language. So long as we're attempting to communicate ideas, we'll need a representation between realizations.

@ChShersh And the logistics challenges are mostly a matter of AI tooling and accounting, which have been improving rapidly.

@JustDeezGuy You should definitely add at least one DSL-oriented language. Perhaps something like Racket or Red.

@JustDeezGuy Self is also a very good OOP exemplar.