That’s not true. What Lemmy is assuring you is that mythos is not significantly better than all the other llms that are out there. If you take a code base that hasn’t been examined by any LLM and run mythos on it, it probably will find a lot. However, if you’ve been using all the others and now start using mythos, you will find little more.
So what about what I just said isn’t true? Aren’t you now just repeating that exact narrative, that the “too powerful to release” is just marketing speech?
I can see how the confusion arises. People here have been saying that LLMs are useless. Now you read about Mythos finding these exploits. But Mythos is not any better than the previous LLMs. Which means that it is useless. So any suggestion to the contrary can only be a marketing tactic. QED
LLMs aren’t useless just that their capabilities are vastly overstated.
Finding patterns in code is what they are good at, what’s annoying with Mythos is they are scanning code that hasn’t really seen scrutiny and claiming it’s groundbreaking because it’s finding stuff.
Who knows. Ask in a few years. So far things don’t seem to be slowing down but it hasn’t been long and it seems like there can’t be too many 20 year old holes left.
But how can this be? Lemmy assured me that this capability of Mythos was just a marketing tactic.
That’s not true. What Lemmy is assuring you is that mythos is not significantly better than all the other llms that are out there. If you take a code base that hasn’t been examined by any LLM and run mythos on it, it probably will find a lot. However, if you’ve been using all the others and now start using mythos, you will find little more.
So what about what I just said isn’t true? Aren’t you now just repeating that exact narrative, that the “too powerful to release” is just marketing speech?
I can see how the confusion arises. People here have been saying that LLMs are useless. Now you read about Mythos finding these exploits. But Mythos is not any better than the previous LLMs. Which means that it is useless. So any suggestion to the contrary can only be a marketing tactic. QED
LLMs aren’t useless just that their capabilities are vastly overstated.
Finding patterns in code is what they are good at, what’s annoying with Mythos is they are scanning code that hasn’t really seen scrutiny and claiming it’s groundbreaking because it’s finding stuff.
Cool cool. How long will that stay true?
Who knows. Ask in a few years. So far things don’t seem to be slowing down but it hasn’t been long and it seems like there can’t be too many 20 year old holes left.