I think of language as a way to determining the generation of mental representations, which are causally interactive code. LLMs demonstrate the ability to interact with reality, eg write meaningful computer programs, because the text tokens they analyze are not meaningless.
RESEARCH
-
AI Forecasting Extreme Weather for Climate Innovation
By
–
Bill Collins from @BerkeleyLab talking about extreme weather forecasting using #AI at @BerkeleyDataSci AI and physical sciences for climate innovation symposium
-
DBRX Open-Source Model Emphasizes Data Intelligence Over General Intelligence
By
–
“It’s not general intelligence; it’s data intelligence on their data.” Learn more about the release of DBRX, a new open-source general-purpose model, as shared by CEO @alighodsi on @technology
-
Wittgenstein’s Language Games and LLM Capabilities Debate
By
–
If you somehow managed to stick to Wittgenstein's beliefs without being swayed by the 80 intermittent years, the observable performance of LLMs should disprove them, but your notion that language games are unavailable to AI suggests that you are profoundly uncurious?
-
GPT-4 Multimodal Training from Original Paper
By
–
It's GPT-4 itself – the GPT-4 model was trained to be multi-modal with images from the start, see the original paper https://
arxiv.org/html/2303.0877
4v6#S4
… -
Wittgenstein’s Tractatus and Programming Language Semantics
By
–
Tractatus tries to establish a direct correspondence between symbols and meaning, akin to programming languages, and programming languages explain how meaning is constructed over automata, but later Wittgenstein did not see a path from function execution to linguistic behavior.
-
Wittgenstein’s Tractatus and Early Computationalism: Turing’s Influence
By
–
The Tractatus is brilliant, even if it ultimately fails. Wittgenstein's thinking deeply interfaced with early computationalism, he influenced Turing and already understood universality. I am not sure if Tarski could have learned much from Wittgenstein about semantics, though
-
Image Resolution Impact on AI Model Hallucination Risks
By
–
OK, this seems to be key: if the text is illegible it's VERY prone to hallucination. In my case I'd tried turning a PDF into a single long image concatenating pages together, which clearly got resized down to where it could'nt make out individual letters
-
Wittgenstein’s Observational Limits in Understanding Algorithms
By
–
The opposite: Because Wittgenstein resorts to purely reporting his observables during generation and is neither experimenting nor taking an engineering perspective (how does my mind have to be constructed to produce observed phenomenology and function?) the algorithm eludes him.
-
Tribute to Daniel Kahneman Great Mind AI Workshop
By
–
I have had the incredible honor and joy of spending a few days with Daniel Kahnemann during a small AI workshop. He was sharp, sincere, acerbic, funny, deep and curious. One of the great intellects of our time. You are missed