iiinova — where your digital self keeps evolving

I think it's likely that the process of training frontier llms does, yes, on the basis that they do stuff that seems unlikely to be doable without thinking. It's possible it doesn't and maths is just way easier than i thought.

110

dude the chinese room is a bad argument. putting something on a weird computer does not matter to me. i don't have your brain damage.

100

why does "sufficiently large neural networks can memorise random data" establish that training a not-sufficiently-large neural network on random data will generate the same internal structure we know exists because of all the studies that finds it?

100

i am not interested in your private language. the reason some computer programs are 'thinking' and some aren't is because some replicate the causal relationships of thinking and some don't, much the same way a computer adding up numbers is 'adding' even though it doesn't have paper and pencil.

100

it is pretty dead on for something searle says yeah. "What if we computed a mind. Would be pretty fucking weird right?". No? Seems normal to me? you could do the same thing to a brain? bizarre argumentative form.

100

you jumped in with the strongest version of your claim! I do think claude thinks, in a meaningful sense, but you do actually have to defend the strong claim you made that started this thread.

100

why are you so desperate to believe brains are magic?

100

this is not what functionalism means. i suggest you read the wikipedia page or my prior tweet.

100

also the original contention was that no *program* can think, which none of this is relevant for.

100

100% of these are just as applicable to brains. One of them is the chinese room. One of them is factually wrong (training on random noise will give you a model that generates random noise with no meaningful internal organisation).

100

kinda wild that this wasn't immediately obvious to everyone. exactly isomorphic to asking where in the computer the "computation" happens.

180

(also note that your argument here is *also* one searle made)

so there's this philosophical position you may not have heard of called "functionalism" which holds that mental states are defined by their causal relationships with other mental states/inputs/outputs. note that this does not mean every computation realises the same mental state.

a computer is not a mathematical object. where do you think the thought happens when a neuron is doing the thinking? (perhaps in the Special Causal Properties? i think you would really benefit from reading Searle www.cs.princeton.edu/courses/arch... so you could present your viewpoint more clearly)

cs.princeton.edu

again do you realise you are very literally recapitulating searle?

so is it metaphysically impossible or not? also why did you say you're not john searle and then immediately start doing searles can't-believe-it's-not-property-dualism?

maybe you should go argue with this dickhead then: bsky.app/profile/boxo...

You assert that *it's impossible for a computer to think*. Presumably you don't think it's impossible for you to think. What magic do you assert brains contain?

genuinely no idea how you can be so confused. my best guess is that you think we implicitly believe in whatever magic you believe in, so we have to think the vectors are doing it??? actively bizarre. like what, are we doing a penrose here?

I don't believe in Searles, either. Look dude i get that you believe in magic but we don't actually have to respect that belief or whatever word you've got for it.

i don't believe in souls, sorry!

lambda calculus is unfortunately nontrivial (but possible, obviously) to write as a type 0 grammar, is also Just Pattern Matching, and is equivalent to turing machines.

010

strictly speaking i think you have to iterate applying the type 0 grammar but basically yeah. every type 0 grammar has an equivalent turing machine that recognises the same language and vice versa.

110

type 0 grammars are "pattern matching" and they're equivalent to turing machines.

rev. howard arson @theophite.bsky.social · 03/10/2026

Reposted by jpicone

Philosopher: "In my book I invented Color Scientist Mary as a thought experiment." Tech Company: "At long last, we have created Color Scientist Mary from classic philosophical paper What Mary Didn't Know (1986)" arxiv.org/abs/2109.06129

arxiv.org

Can Language Models Encode Perceptual Structure Without Grounding? A Case Study in Color

Pretrained language models have been shown to encode relational information, such as the relations between entities or concepts in knowledge-bases -- (Paris, Capital, France). However, simple relation...

2657

the audio from the model + the dialogue from it had pretty distinct LLM Smell but this is very impressive nevertheless.

020

bsky.app/profile/cool... oh my god she doesn't know how tool calling works.

050

Unless you just mean "you could simulate an entity on a computer but LLMs lack some important but ultimately physical/mathematical component" E.g. if they couldn't do xor or something. I don't think that's true but it is an argument. 🧵2/2

000

jpicone @me.jamespicone.name · 01/10/2026

The entity being simulated? Has someone brought up the "what if we replaced a person's neurons one by one with digital ones, at what point do they stop being a 'conscious entity'?" one yet? I find it very hard to get into the headspace where there's no entity here without believing in souls. 🧵1/2

110

I haven't seen much of TOS but coincidentally I have seen the TOS version of this episode and imo the TNG version could really have benefited from something with the emotional impact of Spock getting all up in his feelings rather than being 90% Shenanigans.

000

jpicone @me.jamespicone.name · 01/10/2026

I'm once again making an attempt to actually watch all of Star Trek TNG and it is once again deeply hilarious that the second episode of the entire series - before we even really know any of the characters - is the sex pollen one.

Elsewhere in the thread is Qwen 3.8 local with reasoning on bsky.app/profile/did:... . Do you think this is doing tool calls too?

0220

Honestly I think she could just go "yeah stochastic parrots can do that, but notice that it's not very good as the digit count gets large"? surely this is not such a hole in her not-even-an-empirical-hypothesis hypothesis.

160

Notable at what point in the thread she stopped replying: bsky.app/profile/coli...

2593

Thinking of becoming a new type of guy.

The word "computer" used to be a human job te, but no one today would say it's anthropomorphizing a laptop to say it "computes."

1416

Genuinely unsure if she just doesn't know how tool calling works or if she's developing a conspiracy theory to explain away models doing things she doesn't think are possible.

370

I commented to my spouse the other night that as a practicing programmer you really only need to know about P and NP and the entire rest of the complexity zoo just isn't relevant 😛.

010

My AI course in uni taught me beam search and expert systems 😛. I guess the graph search stuff is legitimately useful

jpicone @me.jamespicone.name · 29/09/2026

very funny to see bender run into actual empirical data on what llms can do (see context of thread), but I think it might be funnier that she suggests openai et al are running a classifier on prompts and handing them to a calculator if they're a maths problem???

3323

I have been on the internet for over twenty years but tonight is the first time I have had an IPv6 address. Please clap, I guess?

001

Reposted by jpicone

Why @why.bsky.world · 29/09/2026

This fuckin dork “in 2026 people told me this company was profitable! Behold! Their 2025 numbers say otherwise!”

2431

jpicone @me.jamespicone.name · 28/09/2026

How long will you keep fighting? Maybe only the x-buster on your hand knows for sure... Anyway time to start the farming sim / PTSD simulator gameplay now.

010

jpicone @me.jamespicone.name · 28/09/2026

game idea: start with the "huh war is kinda bad" epilogue screen from games like megaman x, and then follow your efforts to reintegrate into society after your short burst of incredible violence.