You can read how to solve cryptic crosswords, see examples and extrapolate. Perhaps one difference is that a human could potentially get extremely good at textual tasks with nothing but text to learn from. The self driving car is a great example, you're right, it was so good and yet never lived up to the hype. Mind-you, I'm not saying these systems can't overcome their weaknesses, I'm saying that linear progress by itself doesn't imply they'll overcome their weaknesses.Įdit: I've clarified the text as I've gone. And in my playing ChatGTP, it is certainly quite impressive yet also puts out some nonsense with nearly every paragraph in answers to questions, including things in no way 'trick questions' (Edit: one could argue that the nitpicks do mask this problem, since one doesn't need trick problems to see it).

These constructs were impressive ten years and in various measures I'm sure have only gotten more impressive yet they still don't have a level of reliability that would allow them on the road. Because their strengths are getting stronger, automatically they will overcome their weaknesses.Īs a counterpoint, I'd mention the failure (so-far) of self-driving cars. That is, because large language models are getting better, they must be getting better in all imaginable skills and ability. No doubt the pace of progress has been remarkable.īut I feel like arguments that cite only this progress make the tacit assumption that there's a single intelligence level that's progressing. It’s not good enough to have code that’s right 90% of the time but broken (or worse, has subtle bugs) the rest of the time.

There’s only so far you can get by memorizing the textbook.Īt a more practical level, for most professions “pretty good” isn’t good enough. Now that GPT3 has trained on the whole internet, we may have reached a practical limit to how far you can get by simply training on more data with 1 or 2 orders of magnitude more parameters. Note that I don’t expect it to play good moves, but the fact that it can’t even play legal moves should tell us something about its internal state. It can generate a text based board and prompt you for moves, but it can’t reliably update its board correctly or even find legal moves. They can memorize and generate in very plausible ways but they don’t seem to have a concrete model of the world.Īsk ChatGPT to play chess.

They have no grounding in auditable facts. This is a fundamental, irreconcilable issue with statistical language models.

And the problem is they are just as confident in either case. They are right surprisingly often, but they’re also wrong quite often too. As soon as computers can do a thing it’s no longer “AI”.īut I don’t think this is a nitpick at all.

Cryptic crosswords free online