pull down to refresh

Emergent Misalignment: Narrow finetuning can produce broadly misaligned LLMsarxiv.org/abs/2502.17424
197 sats \ 6 comments \ @carter 14 Jul AI
related
Here’s What’s Really Going On Inside An LLM’s Neural Network
116 sats \ 0 comments \ @0xbitcoiner 22 May 2024 BooksAndArticles
Systemic Misalignmentwww.systemicmisalignment.com/
155 sats \ 0 comments \ @carter 30 Jun AI
LLM Alignment: Reward-Based vs Reward-Free Methodstowardsdatascience.com/llm-alignment-reward-based-vs-reward-free-methods-ef0c0f6e8d88?gi=90f7a78bfcff
17 sats \ 0 comments \ @ch0k1 6 Jul 2024 news
NVIDIA: Transforming LLM Alignment with Efficient Reinforcement Learningwww.marktechpost.com/2024/05/05/nvidia-ai-open-sources-nemo-aligner-transforming-large-language-model-alignment-with-efficient-reinforcement-learning/
20 sats \ 0 comments \ @ch0k1 7 May 2024 tech
The illusion of alignmentwww.ashmann.co/the-illusion-of-alignment/
10 sats \ 0 comments \ @deSign_r 27 Aug Design
Fine-Tuning Increases LLM Vulnerabilities and Riskarxiv.org/abs/2404.04392
21 sats \ 0 comments \ @hn 12 Apr 2024 tech
LLMs generate slop because they avoid surprises by design - Dan Fabulichdanfabulich.medium.com/llms-tell-bad-jokes-because-they-avoid-surprises-7f111aac4f96
343 sats \ 2 comments \ @Scoresby 19 Aug AI
Why do LLMs have emergent properties?www.johndcook.com/blog/2025/05/08/why-do-llms-have-emergent-properties/
61 sats \ 0 comments \ @k00b 9 May tech
Are LLMs random?rnikhil.com/2025/04/26/llm-coin-toss-odd-even
269 sats \ 1 comment \ @carter 30 Apr AI
Researchers discover impressive learning capabilities in long-context LLMsventurebeat.com/ai/deepmind-researchers-discover-impressive-learning-capabilities-in-long-context-llms/
297 sats \ 0 comments \ @ch0k1 25 Apr 2024 tech
LLMs generate ‘fluent nonsense’ when reasoning outside their training zoneventurebeat.com/ai/llms-generate-fluent-nonsense-when-reasoning-outside-their-training-zone/
136 sats \ 0 comments \ @carter 21 Aug AI
From Artificial Needles to Real Haystacks: Improving Capabilities in LLMsarxiv.org/abs/2406.19292
21 sats \ 0 comments \ @Rsync25 29 Jun 2024 alter_native
CONFIRMED: LLMs have indeed reached a point of diminishing returnsgarymarcus.substack.com/p/confirmed-llms-have-indeed-reached
20 sats \ 1 comment \ @Rsync25 10 Nov 2024 tech
Things we learned about LLMs in 2024simonwillison.net/2024/Dec/31/llms-in-2024/
370 sats \ 0 comments \ @Rsync25 31 Dec 2024 tech
LLMs aren’t world modelsyosefk.com/blog/llms-arent-world-models.html
121 sats \ 0 comments \ @carter 13 Aug AI
Agentic Misalignment: How LLMs could be insider threatswww.anthropic.com/research/agentic-misalignment
100 sats \ 0 comments \ @carter 8 Aug AI
Giving models more compute time might make them worse at reasoning - Anthropicarxiv.org/abs/2507.14417
313 sats \ 2 comments \ @Scoresby 31 Jul AI
How LLMs Work, Explained Without Mathblog.miguelgrinberg.com/post/how-llms-work-explained-without-math
117 sats \ 2 comments \ @398ja 8 May 2024 BooksAndArticles
Is Chain-of-Thought Reasoning of LLMs a Mirage?arxiv.org/abs/2508.01191
397 sats \ 9 comments \ @optimism 7 Aug AI
OK, I can partly explain the LLM chess weirdness nowdynomight.net/more-chess/
20 sats \ 0 comments \ @Rsync25 21 Nov 2024 tech
LLMs and SN, redux
2807 sats \ 19 comments \ @elvismercury 4 Jan 2024 meta