Over the last three years, companies worldwide have invested between 30 and 40 billion dollars into generative artificial intelligence projects. Yet most of these efforts have brought no real business…
Can you elaborate? How is this not reasoning? Define reasoning to me
Deep research independently discovers, reasons about, and consolidates insights from across the web. To accomplish this, it was trained on real-world tasks requiring browser and Python tool use, using the same reinforcement learning methods behind OpenAI o1, our first reasoning model. While o1 demonstrates impressive capabilities in coding, math, and other technical domains, many real-world challenges demand extensive context and information gathering from diverse online sources. Deep research builds on these reasoning capabilities to bridge that gap, allowing it to take on the types of problems people face in work and everyday life.
While that contains the word “reasoning” that does not make it such. If this is about the new “reasoning” capabilities of the new LLMS. It was if I recall correctly, found our that it’s not actually reasoning, just doing a fancy footwork appear as if it was reasoning, just like it’s doing fancy dice rolling to appear to be talking like a human being.
As in, if you just change the underlying numbers and names on a test, the models will fail more often, even though the logic of the problem stays the same. This means, it’s not actually “reasoning”, it’s just applying another pattern.
With the current technology we’ve gone so far into this brute forcing the appearance of intelligence that it is becoming quite the challenge in diagnosing what the model is even truly doing now. I personally doubt that the current approach, which is decades old and ultimately quite simple, is a viable way forwards. At least with our current computer technology, I suspect we’ll need a breakthrough of some kind.
But besides the more powerful video cards, the basic principles of the current AI craze are the same as they were in the 70s or so when they tried the connectionist approach with hardware that could not parallel process, and had only datasets made by hand and not with stolen content. So, we’re just using the same approach as we were before we tried to do “handcrafted” AI with LISP machines in the 80s. Which failed. I doubt this earlier and (very) inefficient approach can solve the problem, ultimately. If this keeps on going, we’ll get pretty convincing results, but I seriously doubt we’ll get proper reasoning with this current approach.
If we’re talking about Artificial INTELLIGENCE, then we should talk about “reasoning” as an ability to apply logic and not just match patterns. Because pure pattern matching is decidedly NOT reasoning, because if the pattern changes even a little (change the names and numbers, keeping the logic intact) all models start showing failures. So, yes, some people decided to reframe what “reasoning” means in this context (moving goalposts), but I’m pretty sure that 99% people who use the term when referring to AI don’t mean reasoning like that. Regardless, it’s not actually that of an interesting discussion, not do I actually care that much. So, sure, I’ll give you that point.
Can you elaborate? How is this not reasoning? Define reasoning to me
While that contains the word “reasoning” that does not make it such. If this is about the new “reasoning” capabilities of the new LLMS. It was if I recall correctly, found our that it’s not actually reasoning, just doing a fancy footwork appear as if it was reasoning, just like it’s doing fancy dice rolling to appear to be talking like a human being.
As in, if you just change the underlying numbers and names on a test, the models will fail more often, even though the logic of the problem stays the same. This means, it’s not actually “reasoning”, it’s just applying another pattern.
With the current technology we’ve gone so far into this brute forcing the appearance of intelligence that it is becoming quite the challenge in diagnosing what the model is even truly doing now. I personally doubt that the current approach, which is decades old and ultimately quite simple, is a viable way forwards. At least with our current computer technology, I suspect we’ll need a breakthrough of some kind.
But besides the more powerful video cards, the basic principles of the current AI craze are the same as they were in the 70s or so when they tried the connectionist approach with hardware that could not parallel process, and had only datasets made by hand and not with stolen content. So, we’re just using the same approach as we were before we tried to do “handcrafted” AI with LISP machines in the 80s. Which failed. I doubt this earlier and (very) inefficient approach can solve the problem, ultimately. If this keeps on going, we’ll get pretty convincing results, but I seriously doubt we’ll get proper reasoning with this current approach.
But pattern recognition is literally reasoning. Your argument sounds like “it reasons, but not as good as humans, therefore it does not reason”
I feel like you should take a look at this: https://en.m.wikipedia.org/wiki/Reasoning_system
If we’re talking about Artificial INTELLIGENCE, then we should talk about “reasoning” as an ability to apply logic and not just match patterns. Because pure pattern matching is decidedly NOT reasoning, because if the pattern changes even a little (change the names and numbers, keeping the logic intact) all models start showing failures. So, yes, some people decided to reframe what “reasoning” means in this context (moving goalposts), but I’m pretty sure that 99% people who use the term when referring to AI don’t mean reasoning like that. Regardless, it’s not actually that of an interesting discussion, not do I actually care that much. So, sure, I’ll give you that point.