kabir
@kabir25.bsky.social
Logged the full summary here: www.notion.so/kabir25/LLMS...
LLMS Know More than they Show | Notion
Key Contributions
www.notion.so
April 8, 2025 at 3:07 PM
Logged the full summary here: www.notion.so/kabir25/LLMS...
• Different errors, different signals: Internal states can even help predict what kind of error a model will make — factual, reasoning, etc.
• Hidden knowledge: Sometimes, models internally know the right answer… but still generate the wrong one externally.
• Hidden knowledge: Sometimes, models internally know the right answer… but still generate the wrong one externally.
April 8, 2025 at 3:07 PM
• Different errors, different signals: Internal states can even help predict what kind of error a model will make — factual, reasoning, etc.
• Hidden knowledge: Sometimes, models internally know the right answer… but still generate the wrong one externally.
• Hidden knowledge: Sometimes, models internally know the right answer… but still generate the wrong one externally.
Takeaways:
• Truth is token-specific: Truthfulness signals are concentrated in certain tokens — probing those can significantly boost error detection.
• Generalization is tough: These probing techniques don’t generalize across datasets, which means LLMs hold multiple fragmented notions of truth.
• Truth is token-specific: Truthfulness signals are concentrated in certain tokens — probing those can significantly boost error detection.
• Generalization is tough: These probing techniques don’t generalize across datasets, which means LLMs hold multiple fragmented notions of truth.
LLMS Know More than they Show | Notion
Key Contributions
www.notion.so
April 8, 2025 at 3:07 PM
Takeaways:
• Truth is token-specific: Truthfulness signals are concentrated in certain tokens — probing those can significantly boost error detection.
• Generalization is tough: These probing techniques don’t generalize across datasets, which means LLMs hold multiple fragmented notions of truth.
• Truth is token-specific: Truthfulness signals are concentrated in certain tokens — probing those can significantly boost error detection.
• Generalization is tough: These probing techniques don’t generalize across datasets, which means LLMs hold multiple fragmented notions of truth.
which city in Maharashtra?
March 23, 2025 at 11:24 AM
which city in Maharashtra?
hello! are you hiring undergrad ml research interns?
March 20, 2025 at 12:52 PM
hello! are you hiring undergrad ml research interns?
this is insane, Huge congrats! 👏
December 19, 2024 at 12:08 PM
this is insane, Huge congrats! 👏
bad recs=zero vibes= productivity tanked!
December 6, 2024 at 9:30 AM
bad recs=zero vibes= productivity tanked!