This is a fundamental flaw of sequential (auto-regressive) symbol prediction.
The fix is simple: don't do auto-regressive symbol prediction.
RESEARCH
-
Beyond Autoregressive: Rethinking Sequential Symbol Prediction in AI
By
–
-
Symbol Manipulation Limits in AI and Animal Reasoning
By
–
Among all forms of reasoning, the subset for which a manipulation of discrete symbols helps the reasoning is tiny.
For human reasoning, it's basically just math and (to some extent) code.
For animal reasoning, it's an empty set. Communication is a separate issue from reasoning. -
Continuous Embedding Space Superior for Reasoning: Theoretical Proof
By
–
It is intuitively obvious that reasoning in continuous embedding space is dramatically more powerful than reasoning in discrete token space.
This paper from @tydsh and team show that it is the case theoretically. -
Yann LeCun keynote on AI at VivaTech 2025
By
–
of my keynote/fireside chat with Melissa Heikkilä (Financial Time) at Vivatech on 2025-06-11 . https://
linkedin.com/posts/yann-lec
un_live-metas-chief-ai-scientist-delivers-activity-7341012454140698625-mXyM
… -

II-Medical-8B-1706: Efficient Open Medical Model Outperforms Larger Competitors
By
–
Intelligent Internet introduced II-Medical-8B-1706, an updated version of its open medical model Capable of running on <8GB RAM, the AI outperformed Google’s MedGemma 27B across benchmarks despite 70% fewer parameters
-
Visual Cortex Architecture: Contralateral Processing in Neural Systems
By
–
the visual cortex is on the other side
-
USA vs China: The AGI Race and AI Infrastructure Leadership
By
–
I believe your intention is good. But you’re not thinking ahead enough steps. The USA is not the only country building this tech. So is China and their stated goal is to be the leaders by 2030. Whoever nails self improving AGI first and has the best AI infrastructure will be
-
Pope Leo Embraces AI Technology in Religious Context
By
–
Pope Leo takes on #AI #RiseoftheRobots https://
wsj.com/tech/ai/pope-l
eo-ai-tech-771cca48?st=RJJ5DA
… via @WSJ -

Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, Gauges
By
–
Geometric Deep Learning: Grids, Groups, Graphs, Geodesics, and Gauges Bronstein et al.: https://
arxiv.org/abs/2104.13478 #ArtificialIntelligence #DeepLearning #MachineLearning -
Non-pro o3 finds valid solutions faster than pro
By
–
Didn’t try pro because non-pro o3 can find many valid solutions (and is faster)