General Intelligence
The latest in cutting edge AI research. Curated and delivered to your inbox every Monday and Thursday.
Connect
Plus RAG for tables, suicide risk detection on social media, how chain-of-thought-prompting improves LLM reasoning
Plus Meta's new research, new techniques to mitigate hallucinations, and generating images from brainwaves
Plus AI trading agents, musical AI, and better code generation
Plus Nvidia reduces the cost of inference, AI can infer what a user is asking, deepfake dataset, "unlearning" to improve AI safety
Plus detecting cheating during exams with AI, AI2 releases open-source language models, mitigating hallucinations in vision language models, and zero-shot detection of AI images
Plus Amazon innovates on product translations, Anthropic releases Contextual Retrieval, Detecting deepfakes with smartphones, and RLHF can mislead humans
Plus new models for human-like conversations and better RAG, chain-of-thought is only useful in math tasks, and a new health AI newsletter
Plus Microsoft releases a model to simulate financial markets, Meta researchers release Source2Synth, an LLM that can predict stock prices, and Google finds a better way to translate long-text with LLMs
Plus research from DeepMind, indecision can be incorporated into recommender systems, a dataset for jailbreaking LLMs, text-to-3D models,
Plus predicting if a startup will be successful, how TikTok content can be moderated, low-light image enhancement, xAI, and a new 7B parameter model from AI2
Plus PatentGPT can generate patents automatically, NVIDIA defends RAG vs long-context windows, facial recognition on pixelated images, and why AI agents perform poorly
Plus warships designed by AI, combating model collapse, and synthetic data with small LLMs
![[#28] DeepMind finds better ways to compress data using small language models](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/40479d2b-b56b-41f7-825c-87cdd6a28645/Screenshot_2024-10-10_at_3.18.55_PM.png)
![[#27] A new kind of neural net based on Fourier Transforms](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/cadda426-d92b-4ad7-a1d7-c44ee94c47ed/Oct-07-2024_21-02-18.gif)
![[#26] This method generates 3D renderings of clothes given a single image of fabric](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/56a8f2f8-a1d2-415b-bfa1-7be6ba6387d5/Screenshot_2024-10-03_at_8.59.50_AM.png)
![[#25] Generating camouflage that can fool detectors](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/52aaf0bb-dc3c-4e66-b68a-d404b470a819/speech-deepfake.png)
![[#24] Meta releases Llama 3.2, models for mobile & edge devices](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/2574666e-3235-4f66-bd81-ea518585aef1/Sep-26-2024_05-29-21.gif)
![[#23] Apple's HyperCloning: Training Using Small Models](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/f632fb7e-37ee-4db3-924e-7917a6e86f56/Screenshot_2024-09-21_at_10.28.16_PM.png)
![[#22] NVLM: Nvidia's new family of multimodal LLM](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/92fa9ca5-840e-4c80-b9c6-1c9404150ee9/jensen.png)
![[#21] Nvidia releases new SOTA reranking model](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/2db2b431-5784-4621-9b6d-791bb052340a/Screenshot_2024-09-16_at_6.17.10_AM.png)
![[#20] Are better deepfakes here? TikTok's parent company unveils enhanced dubbing framework.](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/6b93ef8e-e102-4835-8935-f1d17a57bd96/tswift.gif)
![[#19] ScaleAI improves how LLMs generate code](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/0f953b55-081d-4cb0-8f70-13553d442f91/Screenshot_2024-09-08_at_10.21.22_AM.png)
![[#18] How to detect LLM-generated code](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/0821a823-5527-447a-bbdc-5114212f7997/66293b4a-3ca9-47b5-ba9c-7ec978aa59a3.jpeg)
![[#17] Deepfake video detection](https://media.beehiiv.com/cdn-cgi/image/format=auto,width=800,height=421,fit=scale-down,onerror=redirect/uploads/asset/file/fa820109-6799-4bf1-80d2-a1d5d7c62a3e/Comparison_deepfake_blogpost_EN.jpg)