It needs to learn in a new way beyond statistics. LLMs rely on stats. So the path that they are going down isn't one that is likely to bear fruit.
AGI
-
Generative AI: A Foreshock to AI Singularity
By
–
#GenerativeAI may only be a foreshock to #AIsingularity https://
venturebeat.com/ai/generative-
ai-may-only-be-a-foreshock-to-ai-singularity/
… via @VentureBeat -
Current AI Limitations: Human Performance Remains the Benchmark
By
–
So does humans, it’s the best that AI can do as of now!!
-
LLMs and Chatbots Understanding Human Intelligence Aspects
By
–
This is nice thread on LLMs and chatbots by @fchollet
. This is as much about understanding LLMs as it is about understanding different aspects of what we think of as human intelligence -
Large Models Unlock Compound Improvement Threshold for LLMs
By
–
agree – the recent string of works using large models to improve LLM (CAI, tools…) is mind-blowing. i feels like we've unlocked something that was tried many times before in AI but never worked – passing a compound-improvement-threshold that open a wild west of new capabilities
-
What new properties will future LLMs bring?
By
–
This is quite remarkable! What new properties will future LLMs bring? 2/2
-
Theory of Mind Benchmark: Emerging Cognitive Properties in LLMs
By
–
Theory of mind (ToM) benchmark shows the performance of different LLMs on their ability to understand people's mental states. It would be great to understand what changed (for example, from GPT-3-davinci-002 to GPT-3.5-davinci-003) to make these cognitive properties emerge. 1/2
-
Humanity as Minor Component in Complex AI Systems
By
–
Dans cette hypothèse, j’aurais élaboré un système infiniment complexe, dans lequel les 13,8 milliards de sapiens sapiens ne formeraient qu’un infime maillon.
-
Gary Marcus Counterexamples Part 2: AI Systems Performance Testing
By
–
Contraejemplos de Gary Marcus, parte 2. Aquí de nuevo, parece que ambos sistemas ya no fallan en estos ejemplos. Pero Bing es tan chulo que hasta te explica el tipo de test que le estamos haciendo.
-
LLMs Still Hallucinate: Major Obstacle to Large-Scale Deployment
By
–
Fundamentally, LLMs are still not ready for the large-scale deployment both companies envision — because they still ‘hallucinate.’ ChatGPT, BARD et al basically still get s*** wrong, and they make s*** up. Those kinks need to be worked out fast.