When talking about LLMs where do you draw the line for "large"? 200 million params? I think this question greatly changes the answer
December 21, 2025 at 5:08 PM
When talking about LLMs where do you draw the line for "large"? 200 million params? I think this question greatly changes the answer
That's definitely a real problem though I think it can be ameliorated. For example by feeding reviewers some fraction of negatives. Things like anti insurance fraud are already powered by ml+reviewers. It feels weird to suggest that using more accurate ML would make the overall process worse
December 19, 2025 at 2:56 PM
That's definitely a real problem though I think it can be ameliorated. For example by feeding reviewers some fraction of negatives. Things like anti insurance fraud are already powered by ml+reviewers. It feels weird to suggest that using more accurate ML would make the overall process worse
Ok fair enough. Personally I think the "will be reviewed by a human" use case is pretty broad (applications in anti fraud, structuring unstructured data, smart suggestions in UIs, drafting letters, etc)
December 19, 2025 at 2:42 PM
Ok fair enough. Personally I think the "will be reviewed by a human" use case is pretty broad (applications in anti fraud, structuring unstructured data, smart suggestions in UIs, drafting letters, etc)
Right, but llms can also serve such tasks. I mean machine translation is something llms can do (it's where the transformer was invented).
December 19, 2025 at 2:20 PM
Right, but llms can also serve such tasks. I mean machine translation is something llms can do (it's where the transformer was invented).
Doesn't this argument suggest all ML models are dead end? There are cases where errors are tolerable e.g. recommendations, games, anything going to human review, anything fed to a proof checker.
December 19, 2025 at 2:07 PM
Doesn't this argument suggest all ML models are dead end? There are cases where errors are tolerable e.g. recommendations, games, anything going to human review, anything fed to a proof checker.
That model makes bounding boxes? Or are you hand rolling that bit
December 6, 2025 at 7:44 PM
That model makes bounding boxes? Or are you hand rolling that bit
Can you link to the model?
December 6, 2025 at 11:13 AM
Can you link to the model?
Wrap it with paper towels and put it in a bag with rice overnight
December 4, 2025 at 5:19 AM
Wrap it with paper towels and put it in a bag with rice overnight
Surely they are legal on private land or racetracks
November 30, 2025 at 11:11 PM
Surely they are legal on private land or racetracks
Absorbs energy from the moon. Never eat overnight oats after a new moon! Don't ask how I know
November 29, 2025 at 3:54 PM
Absorbs energy from the moon. Never eat overnight oats after a new moon! Don't ask how I know
I don't think this is a *fundamental* objection to chatbots
November 17, 2025 at 4:49 PM
I don't think this is a *fundamental* objection to chatbots
But then you'd leave their platform! They need your eyeballs
September 13, 2025 at 2:56 AM
But then you'd leave their platform! They need your eyeballs
Wow your like my superhero
September 6, 2025 at 3:24 PM
Wow your like my superhero
And LA and Vegas and Phoenix and Atlanta
September 6, 2025 at 10:42 AM
And LA and Vegas and Phoenix and Atlanta
And iirc windows computers refuse to render it except when near other hieroglyphs
September 4, 2025 at 2:58 PM
And iirc windows computers refuse to render it except when near other hieroglyphs
Are you the person that exclaims "jinkies" when surprised?
August 4, 2025 at 4:51 AM
Are you the person that exclaims "jinkies" when surprised?
If you're good at something don't do it for free?
July 29, 2025 at 2:50 AM
If you're good at something don't do it for free?
Is this too celebrate tariffs getting struck down
May 29, 2025 at 12:40 AM
Is this too celebrate tariffs getting struck down