RESEARCH

→ View original post on X — @garymarcus

AGI Discussion: Contrasting Financial Losses with Lack of True AGI and Questioning Benchmarks

By

@garymarcus

–

19 March 2026 21h09

now plot losses, which have grown even faster but anyway that is $, not AGI, of which there is nothing to plot, because there is none. (just benchmarks, which often appear to be gamed).

→ View original post on X — @jiqizhixin

Parallel-Probe: Faster AI Reasoning Without Performance Loss

By

@jiqizhixin

–

19 March 2026 20h58

Can we make AI reasoning much faster and more efficient without losing its smarts? Researchers from the University of Maryland, Washington University in St. Louis, and UNC Chapel Hill introduce Parallel-Probe. This innovative, training-free controller uses "2D probing" to

Reasoning versus Pattern Matching: Causal and Correlative Models

By

–

19 March 2026 20h35

To make it very short: reasoning generates causal models of the data, pattern matching uses associative/correlative models of the data.

Model Limitations: Why AGI Needs True Metalearning Capabilities

By

–

19 March 2026 20h27

The fact that you need to provide a specialized harness clearly shows the model *does not* encode the kind of metalearning knowledge and problem-solving strategies that humans use. Humans solve novel problems without being told how to proceed step by step. AGI would *not* need a

Defining Reasoning and Pattern Matching in AI Systems

By

–

19 March 2026 19h24

"all reasoning is pattern matching" is a useless statement if you don't define "reasoning" and "pattern matching" first. You might as well say "all information processing is information processing." With grounded definitions of both, reasoning and pattern matching are *very*

Encoding Changes Degrade Frontier Model Performance on ARC

By

–

19 March 2026 19h12

This is similar to how applying basic changes to how ARC tasks are encoded considerably degrades frontier model performance. If you're looking at the test for the first time, it really shouldn't matter what the encoding is. Unless you've studied specifically for the test, using a

→ View original post on X — @alexjc

Language Models Struggle with Esoteric Programming Languages

By

@alexjc

–

19 March 2026 19h10

All models struggle in this benchmark because languages are: Brainfuck, Whitespace, Unlambda, Shakespeare. 😅

If you actually pick a useful but still esoteric language like Joy, the frontier models do great (they *can* reason), but the open source ones struggle (they memorize). https://t.co/E5Mozy0yEB
— Alex J. Champandard 🌱 (@alexjc) 19 mars 2026

All models struggle in this benchmark because languages are: Brainfuck, Whitespace, Unlambda, Shakespeare. If you actually pick a useful but still esoteric language like Joy, the frontier models do great (they *can* reason), but the open source ones struggle (they memorize).