4 ms·

LeCun is one of the rare voices of reason in the AI community, especially among the celebrities. He’s been consistently voicing realistic statements with regar

LeCun is one of the rare voices of reason in the AI community, especially among the celebrities. He’s been consistently voicing realistic statements with regards to AGI (he doesn’t believe it’s a thing), AI safety and doomerism, limitations of LLMs, etc.

jarpschop 3y ago
A year ago he said that language models would never be able to explain why if you move a table, the bottle on top of it would not fall. He definitely had no idea what was coming with LLMs. And now he is arrogant enough to say "we all knew what capabilities LLMs would have". He has no intelectual humility or honesty at all. I at least admit openly that language models blew my mind. Lost respect for his opinions massively after seeing how he lied about the topic.
- blueblisters 3y ago
  Can you cite a reference for that quote? I can’t find it. He has definitely underestimated language model capabilities but he has been mostly right on the “language models can’t plan” stuff. He has been consistent and correct about his criticism about auto regressive model as well.
- agnosticmantis 3y ago
  He’s said it’s easy to generate this kind of questions that trick LLMs because of the lack of physical grounding in models trained solely on text. And that’s true now as ever. I also heard him say that training multimodal models on text+image/video would mitigate the grounding issue, and that’s proven to be true too. So I’m not sure exactly what your objection is.
stevenhuang 3y ago
As the other commenter mentions, lecun had a terrible track record when it came to LLMs. All his predictions at the time of what LLMs cannot do were proven wrong. His research at meta is in analytical methods of AI so I'm not surprised. LLMs were seen as anathema to him as he could not comprehend the capabilities that would come from simply scaling up a suitable architecture with compute and data. It's quite remarkable how confidently incorrect he was proven to be. Now he's going for round two on his takes with AI safety. Clearly he didn't learn his lesson.

jarpschop 3y ago

A year ago he said that language models would never be able to explain why if you move a table, the bottle on top of it would not fall. He definitely had no idea what was coming with LLMs. And now he is arrogant enough to say "we all knew what capabilities LLMs would have". He has no intelectual humility or honesty at all. I at least admit openly that language models blew my mind. Lost respect for his opinions massively after seeing how he lied about the topic.

blueblisters 3y ago
Can you cite a reference for that quote? I can’t find it. He has definitely underestimated language model capabilities but he has been mostly right on the “language models can’t plan” stuff. He has been consistent and correct about his criticism about auto regressive model as well.
agnosticmantis 3y ago
He’s said it’s easy to generate this kind of questions that trick LLMs because of the lack of physical grounding in models trained solely on text. And that’s true now as ever. I also heard him say that training multimodal models on text+image/video would mitigate the grounding issue, and that’s proven to be true too. So I’m not sure exactly what your objection is.

stevenhuang 3y ago

As the other commenter mentions, lecun had a terrible track record when it came to LLMs. All his predictions at the time of what LLMs cannot do were proven wrong. His research at meta is in analytical methods of AI so I'm not surprised. LLMs were seen as anathema to him as he could not comprehend the capabilities that would come from simply scaling up a suitable architecture with compute and data. It's quite remarkable how confidently incorrect he was proven to be. Now he's going for round two on his takes with AI safety. Clearly he didn't learn his lesson.