The authors of AI 2027 predict this will happen in March 2027, by the way. Th
Back in 23-24 when I would ask friends at OpenAI and anthropic and deepmind about how far away this sort of architecture was from being used in flagship models, they would generally say “a few years away.” Hence the prediction for AI 2027. To be clear, this isn’t exactly a large sample size scientific survey (it is neither large nor scientific). I definitely don’t feel confident that it’ll be in 2027 specifically. But I’d be curious to hear counterarguments for why we should be fairly confident it’s more than a decade away, for example.
I’m not confident neuralese is more than a decade away. That could happen by 2027 and I wouldn’t be shocked. I don’t think it’ll be a magic bullet though. I expect less of an impedance mismatch between neuralese and the model than language and the model, but reducing that impedance mismatch is the only problem being solved by neuralese.
Back in 23-24 when I would ask friends at OpenAI and anthropic and deepmind about how far away this sort of architecture was from being used in flagship models, they would generally say “a few years away.” Hence the prediction for AI 2027. To be clear, this isn’t exactly a large sample size scientific survey (it is neither large nor scientific). I definitely don’t feel confident that it’ll be in 2027 specifically. But I’d be curious to hear counterarguments for why we should be fairly confident it’s more than a decade away, for example.
I’m not confident neuralese is more than a decade away. That could happen by 2027 and I wouldn’t be shocked. I don’t think it’ll be a magic bullet though. I expect less of an impedance mismatch between neuralese and the model than language and the model, but reducing that impedance mismatch is the only problem being solved by neuralese.