All Articles
16056 articles total
Ask don't tell: Reducing sycophancy in large language models
Research on reducing LLM sycophancy by converting user statements into questions before generating a response.
The Rise of AI in Weather and Climate Information and its Impact on Global Inequality
Discussion on how the concentration of AI development in the Global North may exacerbate global inequality in climate information.
Predictive Speculative KV Replication for Bursty LLM Inference
A discussion on predictive speculative KV replication to optimize bursty LLM inference performance.
Rivian spinoff Also to start delivering e-bikes after months of delays
Rivian spinoff to begin deliveries of e-bikes and develop four-wheel pedal-assist cargo vehicles for Amazon.
Silicon Valley loves young founders. Until it doesn’t.
An exploration of how AI tools are lowering the barrier for young founders to start successful companies.
Train Large, Deploy Compact: Structured Compression for Compact Low-Rank Adaptation
Introduction of PrunedLoRA, a framework that uses structured pruning to create highly representative low-rank adapters for LLMs.
VideoNorms: Benchmarking Cultural Awareness of Video Language Models
Introduction of VideoNorms, a benchmark dataset for assessing the cultural awareness and norm reasoning of Video Large Language Models.
ARC-Encoder: learning compressed text representations for large language models
ARC-Encoder is a new text representation compressor that reduces inference costs for LLMs by compressing context into continuous representations.
$\texttt{AMEND++}$: Benchmarking Eligibility Criteria Amendments in Clinical Trials
AMEND++ is a benchmark suite for predicting eligibility criteria amendments in clinical trials using a new pretraining strategy called CAMLM.
How Context Shapes Truth: Geometric Transformations of Statement-level Truth Representations in LLMs
Research on how context geometrically transforms truth representations (truth vectors) within the activation space of LLMs.
DialectLLM: A Dialect-Aware Dialog[ue] Generation Framework Beyond Standard American English
DialectLLM is a framework for generating multi-dialectal conversational data to improve LLM performance across non-standard English dialects.
Structurally Separated Uncertainty in Supervised Latent Variable Models
A study on structurally separating epistemic and aleatoric uncertainty in supervised latent variable models to reduce correlation between estimates.
Loops (YC W22) Is Hiring a Product Educator
Loops (YC W22) is hiring for a Product Educator position.
Apple Will 'Watch Everything Burn' When AI Bubble Bursts
An opinion piece discussing the potential crash of the AI bubble and Apple's positioning.
Progressive Web Components
Discussion on Progressive Web Components and their implementation in modern web development.
Let's make the worst Htmx
A community project aimed at creating the worst possible implementation of Htmx.
Twenty-five years ago it was cryptography, today it's model weights
A conceptual comparison between the evolution of cryptography and the accessibility of AI model weights.
Authorize, don't authenticate
A technical discussion advocating for authorization over authentication in system design.
Just brute force your embeddings
A technical proposal for using brute force methods to optimize embedding searches in AI.
India is starting to pay for apps, not just download them
Report on the growth of paid app downloads in the Indian market.