Searching...
11 results for “language models”
...models, the one domain where you have the most training data is probably coding. Right? And coding is where you can also have the most structure. And yet, anyone who has used,
44:15 / 45:14
from transcript44:15 – 45:14
...world models. There's growing excitement on that. Do you think there'll be any use in this coming year for world models in the LLM space? Yes. I I do think so also with LLMs, I what's an interesting thing here is I think if we unlock more LLM capabil
3:05:13 / 3:06:16
from transcript3:05:13 – 3:06:16
...And let let's go back to that matrix, analogy that I have, the matrix abstraction. So like I said, you know,
28:29 / 29:31
from transcript28:29 – 29:31
...the the the language, sequential model of LSTM. We we might be able to learn,
14:12 / 15:16
from transcript14:12 – 15:16
...of world models to LLMs again, where they and so instead of just having next token prediction and verifiable rewards, checking the answer correctness, they also make sure the intermediate variables are correct. You know, like, it's kind of like the m
3:06:43 / 3:07:48
from transcript3:06:43 – 3:07:48
...the models match the theoretically correct answer almost perfectly. But pattern matching is not intelligence. LLMs learn correlation. They don't build models of cause and effect. To get to AGI, MISRA argues, we need the ability to keep learning after
0:44 / 1:45
from transcript0:44 – 1:45
not working or not lending itself to law firms because they charge by the hour, and they want to take ages over their work. But that myth is being proved a a myth as we speak. And, actually, it's funny you touched on that because one of the things th
9:43 / 10:47
from transcript9:43 – 10:47
there's some something rustling in that bush. Don't go near. We know how to react to that data. We know how to, save ourselves. We internalize that learning, and our brain cells or our synapses remain plastic throughout our lifetime. What happens wit
24:01 / 25:00
from transcript24:01 – 25:00
...And so this is pretty substantial. They took LAMA three and they were able to increase the performance from 1% to 64%.
1:00:00 / 1:01:02
from transcript1:00:00 – 1:01:02
...free language training data, there are lots of other domains that are not language based, that do not have large datasets that are publicly available. And this is the second point that we we've, discovered along the way is that if you do a head to he
26:09 / 27:28
from transcript26:09 – 27:28
...baseline models. And so this is pretty substantial. They took LAMA three and they were able to increase the performance from 1% to 64%. The outcome of this basically is that this sort of a system can be used to train LLMs to do better reasoning and b
1:00:43 / 1:01:51
from transcript1:00:43 – 1:01:51
Have a podcast?
Run your podcast through Clypt
Get ranked clips, hooks, and ready-to-post copy from your own episodes. Free to try.