All tags
Topic: "alignment"
QwQ-32B claims to match DeepSeek R1-671B
qwen-2.5-plus qwq-32b deepseek-r1 gpt-4.5 gpt-3 davinci alibaba openai deepseek-ai reinforcement-learning math code-execution instruction-following alignment reasoning model-release model-benchmarking scaling performance inference-costs aidan_mclau sama scaling01 juberti polynoamial reach_vb
Alibaba Qwen released their QwQ-32B model, a 32 billion parameter reasoning model using a novel two-stage reinforcement learning approach: first scaling RL for math and coding tasks with accuracy verifiers and code execution servers, then applying RL for general capabilities like instruction following and alignment. Meanwhile, OpenAI rolled out GPT-4.5 to Plus users, with mixed feedback on coding performance and noted inference cost improvements. The QwQ model aims to compete with larger MoE models like DeepSeek-R1. "GPT-4.5 is unusable for coding" was a notable user critique, while others praised its reasoning improvements due to scaling pretraining.
X.ai Grok 3 and Mira Murati's Thinking Machines
grok-3 grok-3-mini gemini-2-pro gpt-4o o3-mini-high o1 deepseek-r1 anthropic openai thinking-machines benchmarking reasoning reinforcement-learning coding multimodality safety alignment research-publishing model-performance creative-ai mira-murati lmarena_ai karpathy omarsar0 ibab arankomatsuzaki iscienceluvr scaling01
Grok 3 has launched with mixed opinions but strong benchmark performance, notably outperforming models like Gemini 2 Pro and GPT-4o. The Grok-3 mini variant shows competitive and sometimes superior capabilities, especially in reasoning and coding, with reinforcement learning playing a key role. Mira Murati has publicly shared her post-OpenAI plan, founding the frontier lab Thinking Machines, focusing on collaborative, personalizable AI, multimodality, and empirical safety and alignment research, reminiscent of Anthropic's approach.
o3 solves AIME, GPQA, Codeforces, makes 11 years of progress in ARC-AGI and 25% in FrontierMath
o3 o3-mini o1-mini gpt-3 gpt-4o o1 openai benchmarking math reasoning model-performance inference-speed cost-efficiency alignment safety-testing sama eric-wallace
OpenAI announced the o3 and o3-mini models with groundbreaking benchmark results, including a jump from 2% to 25% on the FrontierMath benchmark and 87.5% on the ARC-AGI reasoning benchmark, representing about 11 years of progress on the GPT3 to GPT4o scaling curve. The o1-mini model shows superior inference efficiency compared to o3-full, promising significant cost reductions on coding tasks. The announcement was accompanied by community discussions, safety testing applications, and detailed analyses. Sama highlighted the unusual cost-performance tradeoff, and Eric Wallace shared insights on the o-series deliberative alignment strategy.
Meta Llama 3.3: 405B/Nova Pro performance at 70B price
llama-3-70b llama-3.3-70b gpt-4o gemini-exp-1206 meta-ai-fair openai google-deepmind hugging-face llamacloud reinforcement-learning fine-tuning model-performance document-processing pricing-models alignment online-rl sama steven-heidel aidan_mclau lmarena_ai oriolvinyalsml jerryjliu0
Meta AI released Llama 3.3 70B, matching the performance of the 405B model with improved efficiency using "a new alignment process and progress in online RL techniques". OpenAI announced Reinforcement Fine-Tuning (RFT) for building expert models with limited data, offering alpha access to researchers and enterprises. Google DeepMind's Gemini-Exp-1206 leads benchmarks, tying with GPT-4o in coding performance. LlamaCloud enhanced document processing with table extraction and analytics. Discussions on OpenAI's pricing plans continue in the community.
Gemini (Experimental-1114) retakes #1 LLM rank with 1344 Elo
claude-3-sonnet gpt-4 gemini-1.5 claude-3.5-sonnet anthropic openai langchain meta-ai-fair benchmarking prompt-engineering rag visuotactile-perception ai-governance theoretical-alignment ethical-alignment jailbreak-robustness model-releases alignment richardmcngo andrewyng philschmid
Anthropic released the 3.5 Sonnet benchmark for jailbreak robustness, emphasizing adaptive defenses. OpenAI enhanced GPT-4 with a new RAG technique for contiguous chunk retrieval. LangChain launched Promptim for prompt optimization. Meta AI introduced NeuralFeels with neural fields for visuotactile perception. RichardMCNgo resigned from OpenAI, highlighting concerns on AI governance and theoretical alignment. Discussions emphasized the importance of truthful public information and ethical alignment in AI deployment. The latest Gemini update marks a new #1 LLM amid alignment challenges. The AI community continues to focus on benchmarking, prompt-engineering, and alignment issues.
Not much technical happened today
whisper-v3-turbo llama-3 llamaindex openai poolside liquidai perplexity-ai meta-ai-fair cohere fujitsu mixture-of-experts context-windows model-optimization fine-tuning quantization model-training alignment synthetic-data model-architecture agentic-ai nick-turley arav-srinivas francois-fleuret finbarr-timbers lewtun francois-chollet jerry-j-liu mmitchell-ai jxnlco
OpenAI announced raising $6.6B in new funding at a $157B valuation, with ChatGPT reaching 250M weekly active users. Poolside raised $500M to advance AGI development. LiquidAI introduced three new MoE models (1B, 3B, 40B) with a 32k context window and efficient token handling. OpenAI released Whisper V3 Turbo, an open-source multilingual model with significant speed improvements. Meta AI FAIR is hiring research interns focusing on LLM reasoning, alignment, synthetic data, and novel architectures. Cohere partnered with Fujitsu to launch Takane, a custom Japanese model. Technical discussions included challenges in LoRA fine-tuning, float8 quantization in Keras, and new tools like create-llama for agent templates. Industry commentary raised concerns about AI development priorities and highlighted freelancing opportunities in AI.
OpenAI's Instruction Hierarchy for the LLM OS
phi-3-mini openelm claude-3-opus gpt-4-turbo gpt-3.5-turbo llama-3-70b rho-1 mistral-7b llama-3-8b llama-3 openai microsoft apple deepseek mistral-ai llamaindex wendys prompt-injection alignment benchmarking instruction-following context-windows model-training model-deployment inference performance-optimization ai-application career-advice drive-thru-ai
OpenAI published a paper introducing the concept of privilege levels for LLMs to address prompt injection vulnerabilities, improving defenses by 20-30%. Microsoft released the lightweight Phi-3-mini model with 4K and 128K context lengths. Apple open-sourced the OpenELM language model family with an open training and inference framework. An instruction accuracy benchmark compared 12 models, with Claude 3 Opus, GPT-4 Turbo, and Llama 3 70B performing best. The Rho-1 method enables training state-of-the-art models using only 3% of tokens, boosting models like Mistral. Wendy's deployed AI-powered drive-thru ordering, and a study found Gen Z workers prefer generative AI for career advice. Tutorials on deploying Llama 3 models on AWS EC2 highlight hardware requirements and inference server use.
Claude 3 is officially America's Next Top Model
claude-3-opus claude-3-sonnet claude-3-haiku gpt-4o-mini mistral-7b qwen-72b anthropic mistral-ai huggingface openrouter stable-diffusion automatic1111 comfyui fine-tuning model-merging alignment ai-ethics benchmarking model-performance long-context cost-efficiency model-evaluation mark_riedl ethanjperez stuhlmueller ylecun aravsrinivas
Claude 3 Opus outperforms GPT4T and Mistral Large in blind Elo rankings, with Claude 3 Haiku marking a new cost-performance frontier. Fine-tuning techniques like QLoRA on Mistral 7B and evolutionary model merging on HuggingFace models are highlighted. Public opinion shows strong opposition to ASI development. Research supervision opportunities in AI alignment are announced. The Stable Diffusion 3 (SD3) release raises workflow concerns for tools like ComfyUI and automatic1111. Opus shows a 5% performance dip on OpenRouter compared to the Anthropic API. A new benchmark stresses LLM recall at long contexts, with Mistral 7B struggling and Qwen 72b performing well.
CodeLLama 70B beats GPT4 on HumanEval
codellama miqu mistral-medium llama-2-70b aphrodite-engine mixtral flatdolphinmaid noromaid rpcal chatml mistral-7b activation-beacon eagle-7b rwkv-v5 openhermes2.5 nous-hermes-2-mixtral-8x7b-dpo imp-v1-3b bakllava moondream qwen-vl meta-ai-fair ollama nous-research mistral-ai hugging-face ai-ethics alignment gpu-optimization direct-prompt-optimization fine-tuning cuda-programming optimizer-technology quantization multimodality context-length dense-retrieval retrieval-augmented-generation multilinguality model-performance open-source code-generation classification vision
Meta AI surprised the community with the release of CodeLlama, an open-source model now available on platforms like Ollama and MLX for local use. The Miqu model sparked debate over its origins, possibly linked to Mistral Medium or a fine-tuned Llama-2-70b, alongside discussions on AI ethics and alignment risks. The Aphrodite engine showed strong performance on A6000 GPUs with specific configurations. Role-playing AI models such as Mixtral and Flatdolphinmaid faced challenges with repetitiveness, while Noromaid and Rpcal performed better, with ChatML and DPO recommended for improved responses. Learning resources like fast.ai's course were highlighted for ML/DL beginners, and fine-tuning techniques with optimizers like Paged 8bit lion and adafactor were discussed.
At Nous Research AI, the Activation Beacon project introduced a method for unlimited context length in LLMs using "global state" tokens, potentially transforming retrieval-augmented models. The Eagle-7B model, based on RWKV-v5, outperformed Mistral in benchmarks with efficiency and multilingual capabilities. OpenHermes2.5 was recommended for consumer hardware due to its quantization methods. Multimodal and domain-specific models like IMP v1-3b, Bakllava, Moondream, and Qwen-vl were explored for classification and vision-language tasks. The community emphasized centralizing AI resources for collaborative research.
12/8/2023 - Mamba v Mistral v Hyena
mistral-8x7b-moe mamba-3b stripedhyena-7b claude-2.1 gemini gpt-4 dialogrpt-human-vs-machine cybertron-7b-v2-gguf falcon-180b mistral-ai togethercompute stanford anthropic google hugging-face mixture-of-experts attention-mechanisms prompt-engineering alignment image-training model-deployment gpu-requirements cpu-performance model-inference long-context model-evaluation open-source chatbots andrej-karpathy tri-dao maxwellandrews raddka
Three new AI models are highlighted: Mistral's 8x7B MoE model (Mixtral), Mamba models up to 3B by Together, and StripedHyena 7B, a competitive subquadratic attention model from Stanford's Hazy Research. Discussions on Anthropic's Claude 2.1 focus on its prompting technique and alignment challenges. The Gemini AI from Google is noted as potentially superior to GPT-4. The community also explores Dreambooth for image training and shares resources like the DialogRPT-human-vs-machine model on Hugging Face. Deployment challenges for large language models, including CPU performance and GPU requirements, are discussed with references to Falcon 180B and transformer batching techniques. User engagement includes meme sharing and humor.