AI = Deep Learning Causal Inference Symbol Manipulation A I = ∇ w P ( Y = y | d o ( X = x ) ) λ #AI #GenAI #CausalAI #KnowledgeGraphs #NeuroSymbolics

Food for thought: #LLM are black boxes, so are humans, but we deal with them seeing them as non-/cooperative player #markovblankets and using our perception/action loop to probe them. Let’s do that with #GenAI too. There’s your path to explainability.

We have to support ‘biodiversity’ of programming languages.We can’t let #agentic #coding solidify 2 or 3 top languages just because they make up the majority of training data and neglect all other promising ones.This stifles progress of the very substrate of executable formal statements. #SAP #ABAP

Simply retrieving a reasoning trace looks a lot like human reasoning, until it's time to navigate uncharted territory. If you memorized all reasoning traces of humans from 10,000 BC, you could automate their lives but you could not invent modern civilization.

The reason symmetry is so important in physics is because symmetry is a highly effective compression operator. If a system is invariant under some symmetry, you only need to explain one axis of it. Scientific models represent the systematic exploitation of the universe's internal redundancies through symbolic logic.

Users converse in natural language with #AI. Vibe coders too. Skilled developers should talk in a less ambiguous, concise language in an agentic coding session. First approaches are emerging: DSPy, sudolang, DSLs. I have been advocating for this for long now. Saves tokens too. #SAP #ABAP

Enterprise software is not ready for #agenticworkflow. Mutable #OOP is problematic. Critical systems are too big and have too many dependencies to be pulled locally. Therefore a #functionalprogramming based on immutable values is needed and a local lang runtime independent of the data. #SAP #ABAP

As auto-regressive token predictors #LLM predict syntax patterns and have zero senantic understanding: esolang-bench.vercel.app/ #SAP #ABAP

Current AI is a librarian of existing knowledge. Science requires an explorer of the unknown. You don't win a Nobel Prize by staying in the library.

The time to learn how to think for yourself was before genAI, if you missed your chance, good luck

The persisting importance of prompt engineering -- and now harness engineering -- is one of the best indicators of how far we are from AGI. A general system doesn't need a task-specific harness. And when provided with instructions, it is robust to phrasing variations.

We have to come to terms with the fact that the rate of tool calling by an #agent is inversely proportional to the capabilities of the underlying core which is the #LLM. If it could do it on it’s own - no need to call a tool. Progress nowadays happens mostly in the scaffolding. #SAP #ABAP

Sufficiently advanced agentic coding is essentially machine learning: the engineer sets up the optimization goal as well as some constraints on the search space (the spec and its tests), then an optimization process (coding agents) iterates until the goal is reached. The result is a blackbox model (the generated codebase): an artifact that performs the task, that you deploy without ever inspecting its internal logic, just as we ignore individual weights in a neural network. This implies that all classic issues encountered in ML will soon become problems for agentic coding: overfitting to the spec, Clever Hans shortcuts that don't generalize outside the tests, data leakage, concept drift, etc. I would also ask: what will be the Keras of agentic coding? What will be the optimal set of high-level abstractions that allow humans to steer codebase 'training' with minimal cognitive overhead?

Natural language will not stay the pinnacle language of prompting. There must and will be an evolution towards a 6GL formal, declarative, homoiconic meta language serving as the deterministic and formally verifiable substrate to prompt #AI paradigms of the future. #SAP #ABAP