ChatGPT / Emergent world représentation
These People fail to realize that an AI outscoring a human on a test does not mean it can actually do the thing the test measures. - GPT-4 Is Not Getting Worse / HN
We were conflating a lot of things with the word intelligence. - OpenVoice
ChatGPT is a development limité of human knowledge, you can get a good initial point, but when conversing with it, you are derivating farther away from the a valid human interaction. - yduf
see also
- AI to replace 2.4 million jobs in the US by 2030, many fewer than other forms of automation
- C++hatGPT & AI Tools’ Impact - explores the inner workings of large language models with an eye for leveraging them effectively for coding tasks.
- Good art is skill + taste + creativy
- With chatGPT skill is out of the table, because it’s brought by chatGPT at Human scale
- taste: chatGPT has some from existing code
- creativity: chatGPT has some by pattern combination, but not there yet. - see for future plan
- success stories of using chatGPT : use your taste&creativity + feedback from chatGPT to complete your lack of knowledge of particular techhnology.
- UnitTest - test framework shown in the presentation - chatGPT has no planning, no leaping to unknown, not got to what was not done.
- chatGDB - when a program failed it tell you why.
- Good art is skill + taste + creativy
- ChatGPT rêve-t-il de cavaliers électriques ? (Mr Phi) - ChatGPT3.5 reach 1800ELO playing chess / not just a stochastic parrots.
- chatGPT 4 is less good, for unkwnown reason.
- Emergent world representation (like chessboard)
- game history play a lot in ability to play in a good way
- random play history lead to pour play
- good play history lead to better play
- La preuve définitive que ChatGPT ne comprend rien (Mr Phi) - la chambre chinoise
- le fonctionnalisme / le computationisme
- la seule execution d’un algorithme ne suffit pas a générer de la compréhension - subjectif vs objectif (l’experience de la chambre chinoise) - “la vraie comprehension” ? / la conscience
- une comprehension objective sans comprehension subjective
- agrument enormément critiqué (de part et d’autre)
- l’homme processeur ne comprend rien,
- mais le systeme (la chambre chinoise) est le detenteur de la comprehension.
- l’exceptionalité humaine sans justification (les machines (toutes) ne pensent pas (toujours))
- prediction fausse de Yann Lecun - le telephone sur la table
- model entrainée sur des textes - capable de generé des images (via programmes) et dessine un mouton / une licorne.
- le fonctionnalisme / le computationisme
- la seule execution d’un algorithme ne suffit pas a générer de la compréhension - subjectif vs objectif (l’experience de la chambre chinoise) - “la vraie comprehension” ? / la conscience
Written on September 17, 2023, Last update on June 24, 2024
LLM
test
c++
mental-model