All tags
Topic: "instruction-following"
Anthropic releases Claude 4 Sonnet and Opus: Memory, Agent Capabilities, Claude Code, Redteam Drama
claude-4 claude-4-opus claude-4-sonnet claude-3.5-sonnet anthropic instruction-following token-accounting pricing-models sliding-window-attention inference-techniques open-sourcing model-accessibility agent-capabilities-api extended-context model-deployment
Anthropic has officially released Claude 4 with two variants: Claude Opus 4, a high-capability model for complex tasks priced at $15/$75 per million tokens, and Claude Sonnet 4, optimized for efficient everyday use. The release emphasizes instruction following and extended work sessions up to 7 hours. Community discussions highlight concerns about token pricing, token accounting transparency, and calls for open-sourcing Claude 3.5 Sonnet weights to support local model development. The news also covers Claude Code GA, new Agent Capabilities API, and various livestreams and reports detailing these updates. There is notable debate around sliding window attention and advanced inference techniques for local deployment.
not much happened today
kernelllm-8b gpt-4o deepseek-v3 mistral-medium-3 qwen3 blip3-o xgen-small anisora stable-audio-open-small alphaevolve meta-ai-fair mistral-ai qwen deepseek salesforce bilibili stability-ai google benchmarking model-performance multilinguality hardware-optimization multimodality image-generation video-generation text-to-audio model-parallelism chain-of-thought instruction-following reasoning mitigation-strategies reach_vb lmarena_ai theadimeline adcock_brett jxmnop dair_ai omarsar0
Meta released KernelLLM 8B, outperforming GPT-4o and DeepSeek V3 on KernelBench-Triton Level 1. Mistral Medium 3 debuted strongly in multiple benchmarks. Qwen3 models introduced a unified framework with multilingual support. DeepSeek-V3 features hardware-aware co-design. BLIP3-o family released for multimodal tasks using diffusion transformers. Salesforce launched xGen-Small models excelling in long-context and math benchmarks. Bilibili released AniSORA for anime video generation. Stability AI open-sourced Stable Audio Open Small optimized for Arm devices. Google’s AlphaEvolve coding agent improved Strassen's algorithm for the first time since 1969. Research shows chain-of-thought reasoning can harm instruction-following ability, with mitigation strategies like classifier-selective reasoning being most effective, but reasoning techniques show high variance and limited generalization. "Chain-of-thought (CoT) reasoning can harm a model’s ability to follow instructions" and "Mitigation strategies such as few-shot in-context learning, self-reflection, self-selective reasoning, and classifier-selective reasoning can counteract reasoning-induced failures".
Gemini's AlphaEvolve agent uses Gemini 2.0 to find new Math and cuts Gemini cost 1% — without RL
gemini gpt-4.1 gpt-4o-mini o3 o4-mini google-deepmind openai algorithm-discovery coding-agents matrix-multiplication optimization reinforcement-learning model-weights training-efficiency safety-evaluations instruction-following coding-tasks model-releases _philschmid scott_swingle alex_dimakis henry jason_wei kevinweil michpokrass scaling01 gdb
Deepmind's AlphaEvolve, a 2025 update to AlphaTensor and FunSearch, is a Gemini-powered coding agent for algorithm discovery that designs faster matrix multiplication algorithms, solves open math problems, and improves data center and AI training efficiency. It achieves a 23% faster kernel speedup in Gemini training and surpasses state-of-the-art on 20% of applied problems, including improvements on the Minimum Overlap Problem and Kissing number problem. Unlike Deep-RL, it optimizes code pieces rather than model weights. Meanwhile, OpenAI released GPT-4.1 in ChatGPT, specializing in coding and instruction following, with a faster alternative GPT-4.1 mini replacing GPT-4o mini for all users. OpenAI also launched the Safety Evaluations Hub and the OpenAI to Z Challenge using o3/o4 mini and GPT-4.1 models to discover archaeological sites. "Maybe midtrain + good search is all you need for AI for scientific innovation" - Jason Wei.
Granola launches team notes, while Notion launches meeting transcription
gpt-4.1 gpt-4o-mini gpt-4.1-mini claude-opus claude-sonnet claude-o3 qwen3 seed1.5-vl llama-4 am-thinking-v1 openai anthropic alibaba meta-ai-fair huggingface granola coding instruction-following benchmarking model-releases reasoning image-generation collaborative-software model-performance kevinweil scaling01 steph_palazzolo andersonbcdefg reach_vb yuchenj_uw qtnx_ _akhaliq risingsayak
GPT-4.1 is now available in ChatGPT for Plus, Pro, and Team users, focusing on coding and instruction following, with GPT 4.1 mini replacing GPT 4o mini. Anthropic is releasing new Claude models including Claude Opus and Claude Sonnet, though some criticism about hallucinations in Claude O3 was noted. Alibaba shared the Qwen3 Technical Report with strong benchmark results from Seed1.5-VL. Meta FAIR announced new models and datasets but faced criticism on Llama 4. AM-Thinking-v1 launched on Hugging Face as a 32B scale reasoning model. Granola raised $43M in Series B and launched Granola 2.0 with a Notion-like UI. The AI ecosystem shows rapid iteration and cloning of ideas, emphasizing execution and distribution.
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.
SOTA Video Gen: Veo 2 and Kling 2 are GA for developers
veo-2 gemini gpt-4.1 gpt-4o gpt-4.5-preview gpt-4.1-mini gpt-4.1-nano google openai video-generation api coding instruction-following context-window performance benchmarks model-deprecation kevinweil stevenheidel aidan_clark_
Google's Veo 2 video generation model is now available in the Gemini API with a cost of 35 cents per second of generated video, marking a significant step in accessible video generation. Meanwhile, China's Kling 2 model launched with pricing around $2 for a 10-second clip and a minimum subscription of $700 per month for 3 months, generating excitement despite some skill challenges. OpenAI announced the GPT-4.1 family release, including GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano, highlighting improvements in coding, instruction following, and a 1 million token context window. The GPT-4.1 models are 26% cheaper than GPT-4o and will replace the GPT-4.5 Preview API version by July 14. Performance benchmarks show GPT-4.1 achieving 54-55% on SWE-bench verified and a 60% improvement over GPT-4o in some internal tests, though some critiques note it underperforms compared to other models like OpenRouter and DeepSeekV3 in coding tasks. The release is API-only, with a prompting guide provided for developers.
GPT 4.1: The New OpenAI Workhorse
gpt-4.1 gpt-4.1-mini gpt-4.1-nano gpt-4o gemini-2.5-pro openai llama-index perplexity-ai google-deepmind coding instruction-following long-context benchmarks model-pricing model-integration model-deprecation sama kevinweil omarsar0 aidan_mclau danhendrycks polynoamial scaling01 aravsrinivas lmarena_ai
OpenAI released GPT-4.1, including GPT-4.1 mini and GPT-4.1 nano, highlighting improvements in coding, instruction following, and handling long contexts up to 1 million tokens. The model achieves a 54 score on SWE-bench verified and shows a 60% improvement over GPT-4o on internal benchmarks. Pricing for GPT-4.1 nano is notably low at $0.10/1M input and $0.40/1M output. GPT-4.5 Preview is being deprecated in favor of GPT-4.1. Integration support includes Llama Index with day 0 support. Some negative feedback was noted for GPT-4.1 nano. Additionally, Perplexity's Sonar API ties with Gemini-2.5 Pro for the top spot in the LM Search Arena leaderboard. New benchmarks like MRCR and GraphWalks were introduced alongside updated prompting guides and cookbooks.
not much happened today
gpt-4o deepseek-v3 claude-3.7-sonnet o3-mini gemini-2.5-pro openai deepseek anthropic google-deepmind togethercompute hypertecgroup coreweave cursor-ai windsurf-ai coding instruction-following image-generation policy-compliance long-context audio-processing video-processing gpu-clusters ai-infrastructure api-access sama kevinweil joannejang nrehiew_ giffmana _philschmid scaling01 saranormous
GPT-4o was praised for its improved coding, instruction following, and freedom, becoming the leading non-reasoning coding model surpassing DeepSeek V3 and Claude 3.7 Sonnet in coding benchmarks, though it still lags behind reasoning models like o3-mini. Concerns about policy compliance in image generation were noted, with efforts to improve adherence. Gemini 2.5 Pro was highlighted for its advanced audio and video understanding, long context capabilities, and integration with platforms like Cursor AI and Windsurf AI. AI infrastructure developments include a partnership between Together AI and Hypertec Group to deliver large-scale GPU clusters, and CoreWeave's IPO was celebrated for advancing AI infrastructure. GPU and TPU usage is expected to increase significantly. "GPT-4o's transparency and background generation feature" and "Gemini 2.5 Pro scored above 50% on Simple-Bench AI Explanation" were key highlights.
not much happened today
gpt-4o deepseek-v3-0324 gemini-2.5-pro gemini-3 claude-3.7-sonnet openai hugging-face sambanova google-cloud instruction-following image-generation content-filtering model-performance api coding model-deployment benchmarking model-release abacaj nrehiew_ sama joannejang giffmana lmarena_ai _philschmid
OpenAI announced the new GPT-4o model with enhanced instruction-following, complex problem-solving, and native image generation capabilities. The model shows improved performance in math, coding, and creativity, with features like transparent background image generation. Discussions around content filtering and policy for image generation emphasize balancing creative freedom and harm prevention. DeepSeek V3-0324 APIs, available on Hugging Face and powered by SambaNovaAI, outperform benchmarks and models like Gemini 2.0 Pro and Claude 3.7 Sonnet. Gemini 2.5 Pro is recommended for coding, and Gemini 3 can be deployed easily on Google Cloud Vertex AI via the new Model Garden SDK. The Gemma 3 Technical Report has been released on arXiv.
Gemini 2.5 Pro + 4o Native Image Gen
gemini-2.5-pro gpt-4o google-deepmind openai lmarena_ai autoregressive-models multimodality reasoning coding instruction-following model-release leaderboards noam-shazeer allan-jabri gabe-goh
Gemini 2.5 Pro from Google DeepMind has become the new top AI model, surpassing Grok 3 by 40 LMarena points, with contributions from Noam Shazeer integrating Flash Thinking techniques. It is available as a free, rate-limited experimental model. Meanwhile, OpenAI released GPT 4o Native Images, an autoregressive image generation model with detailed insights shared by Allan Jabri and credits to Gabe Goh. Gemini 2.5 Pro excels in reasoning, coding, STEM, multimodal tasks, and instruction following, topping the LMarena leaderboard significantly. It is accessible via Google AI Studio and the Gemini App.
lots of little things happened this week
llama-3-3-nemotron-super-49b-v1 claude anthropic nvidia sakana-ai meta-ai-fair reinforcement-learning reasoning benchmarks multi-turn-collaboration instruction-following dataset-release model-evaluation percy-liang
Anthropic introduced a novel 'think' tool enhancing instruction adherence and multi-step problem solving in agents, with combined reasoning and tool use demonstrated by Claude. NVIDIA's Llama-3.3-Nemotron-Super-49B-v1 ranked #14 on LMArena, noted for strong math reasoning and a 15M post-training dataset. Sakana AI launched a Sudoku-based reasoning benchmark to advance AI problem-solving capabilities. Meta AI released SWEET-RL, a reinforcement learning algorithm improving long-horizon multi-turn tasks by 6%, and introduced CollaborativeAgentBench, a benchmark for collaborative LLM agents working with humans on programming and design tasks. Percy Liang relaunched the HELM benchmark with 5 challenging datasets evaluating 22 top language models.
LLaDA: Large Language Diffusion Models
llada-8b llama-3-8b step-video-t2v-30b step-audio-chat-132b llama-2-7b stepfun-ai scale-ai cambridge llamaindex diffusion-models text-generation multimodality video-generation voice-processing benchmarking instruction-following model-scaling gpu-usage long-context multi-turn-dialogue arankomatsuzaki _akhaliq omarsar0 iscienceluvr gallabytes maximelabonne reach_vb
LLaDA (Large Language Diffusion Model) 8B is a breakthrough diffusion-based language model that rivals LLaMA 3 8B while training on 7x fewer tokens (2 trillion tokens) and using 0.13 million H800 GPU hours. It introduces a novel text generation approach by predicting uniformly masked tokens in a diffusion process, enabling multi-turn dialogue and instruction-following. Alongside, StepFun AI released two major models: Step-Video-T2V 30B, a text-to-video model generating up to 204 frames with high coherence and motion quality, and Step-Audio-Chat 132B, a voice-to-voice model. Additionally, challenging multimodal benchmarks like Scale AI's EnigmaEval and Cambridge's ZeroBench highlight current frontier models scoring zero, emphasizing the difficulty of these tasks. The community also noted the return of diffusion models in language modeling, a previously speculative architecture now scaled successfully.
Mistral Small 3 24B and Tulu 3 405B
mistral-small-3 tulu-3-405b llama-3 tiny-swallow-1.5b qwen-2.5-max deepseek-v3 claude-3.5-sonnet gemini-1.5-pro gpt4o-mini llama-3-3-70b mistral-ai ai2 sakana-ai alibaba_qwen deepseek ollama llamaindex reinforcement-learning model-fine-tuning local-inference model-performance model-optimization on-device-ai instruction-following api training-data natural-language-processing clementdelangue dchaplot reach_vb
Mistral AI released Mistral Small 3, a 24B parameter model optimized for local inference with low latency and 81% accuracy on MMLU, competing with Llama 3.3 70B, Qwen-2.5 32B, and GPT4o-mini. AI2 released Tülu 3 405B, a large finetuned model of Llama 3 using Reinforcement Learning from Verifiable Rewards (RVLR), competitive with DeepSeek v3. Sakana AI launched TinySwallow-1.5B, a Japanese language model using TAID for on-device use. Alibaba_Qwen released Qwen 2.5 Max, trained on 20 trillion tokens, with performance comparable to DeepSeek V3, Claude 3.5 Sonnet, and Gemini 1.5 Pro, and updated API pricing. These releases highlight advances in open models, efficient inference, and reinforcement learning techniques.
Llama 3.2: On-device 1B/3B, and Multimodal 11B/90B (with AI2 Molmo kicker)
llama-3-2 llama-3-1 claude-3-haiku gpt-4o-mini molmo-72b molmo-7b gemma-2 phi-3-5 llama-3-2-vision llama-3-2-3b llama-3-2-20b meta-ai-fair ai2 qualcomm mediatek arm ollama together-ai fireworks-ai weights-biases cohere weaviate multimodality vision context-windows quantization model-release tokenization model-performance model-optimization rag model-training instruction-following mira-murati daniel-han
Meta released Llama 3.2 with new multimodal versions including 3B and 20B vision adapters on a frozen Llama 3.1, showing competitive performance against Claude Haiku and GPT-4o-mini. AI2 launched multimodal Molmo 72B and 7B models outperforming Llama 3.2 in vision tasks. Meta also introduced new 128k-context 1B and 3B models competing with Gemma 2 and Phi 3.5, with collaborations hinted with Qualcomm, Mediatek, and Arm for on-device AI. The release includes a 9 trillion token count for Llama 1B and 3B. Partner launches include Ollama, Together AI offering free 11B model access, and Fireworks AI. Additionally, a new RAG++ course from Weights & Biases, Cohere, and Weaviate offers systematic evaluation and deployment guidance for retrieval-augmented generation systems based on extensive production experience.
not much happened today
o1-preview o1-mini qwen-2.5 gpt-4o deepseek-v2.5 gpt-4-turbo-2024-04-09 grin llama-3-1-405b veo kat openai qwen deepseek-ai microsoft kyutai-labs perplexity-ai together-ai meta-ai-fair google-deepmind hugging-face google anthropic benchmarking math coding instruction-following model-merging model-expressiveness moe voice voice-models generative-video competition open-source model-deployment ai-agents hyung-won-chung noam-brown bindureddy akhaliq karpathy aravsrinivas fchollet cwolferesearch philschmid labenz ylecun
OpenAI's o1-preview and o1-mini models lead benchmarks in Math, Hard Prompts, and Coding. Qwen 2.5 72B model shows strong performance close to GPT-4o. DeepSeek-V2.5 tops Chinese LLMs, rivaling GPT-4-Turbo-2024-04-09. Microsoft's GRIN MoE achieves good results with 6.6B active parameters. Moshi voice model from Kyutai Labs runs locally on Apple Silicon Macs. Perplexity app introduces voice mode with push-to-talk. LlamaCoder by Together.ai uses Llama 3.1 405B for app generation. Google DeepMind's Veo is a new generative video model for YouTube Shorts. The 2024 ARC-AGI competition increases prize money and plans a university tour. A survey on model merging covers 50+ papers for LLM alignment. The Kolmogorov–Arnold Transformer (KAT) paper proposes replacing MLP layers with KAN layers for better expressiveness. Hugging Face Hub integrates with Google Cloud Vertex AI Model Garden for easier open-source model deployment. Agent.ai is introduced as a professional network for AI agents. "Touching grass is all you need."
AIPhone 16: the Visual Intelligence Phone
reflection-70b llama-3-70b qwen-2-72b llama-3-1-405b claude gpt-4 gemini apple openai weights-biases vision video-understanding benchmarking planning model-evaluation privacy ai-integration instruction-following yann-lecun
Apple announced the new iPhone 16 lineup featuring Visual Intelligence, a new AI capability integrated with Camera Control, Apple Maps, and Siri, emphasizing privacy and default service use over third-party AI like OpenAI. Apple Photos now includes advanced video understanding with timestamp recognition. Meanwhile, Reflection-70B claims to be a top open-source model but benchmarks show it performs close to Llama 3 70B and slightly worse than Qwen 2 72B. Yann LeCun highlighted ongoing challenges with LLM planning abilities, noting models like Llama-3.1-405b and Claude show some skill, while GPT-4 and Gemini lag behind. Weights & Biases is sponsoring an event to advance LLM evaluation techniques with prizes and API access.
Reflection 70B, by Matt from IT Department
llama-3.1-70b llama-3 claude-3.5-sonnet hyperwrite glaive fine-tuning chain-of-thought instruction-following synthetic-data quantization model-evaluation prompt-engineering matt-shumer sahil-chaudhary
Reflection Tuning technique has been used by a two-person team from Hyperwrite and Glaive to finetune llama-3.1-70b, showing strong performance improvements with minimal synthetic data. The approach builds on the concept of adding
thinking and reflection steps to outputs, related to the Chain of Thought method. Despite some criticisms like contamination concerns, worse coding performance, and reliance on system prompts, the model has received positive reception and comparisons to claude-3.5-sonnet. The work highlights efficient instruction tuning and synthetic data generation for large models. Apple Intelligence Beta + Segment Anything Model 2
llama-3-405b llama-3 segment-anything-model meta-ai-fair apple image-segmentation memory-attention video-processing pretraining cloud-tpus post-training synthetic-data instruction-following reasoning writing benchmarking bindureddy maximelabonne reach_vb
Meta advanced its open source AI with a sequel to the Segment Anything Model, enhancing image segmentation with memory attention for video applications using minimal data and compute. Apple Intelligence delayed its official release to iOS 18.1 in October but launched developer previews on MacOS Sequoia, iOS 18, and iPadOS 18, accompanied by a detailed 47-page paper revealing extensive pretraining on 6.3T tokens and use of Cloud TPUs rather than Apple Silicon. The paper highlights improvements in instruction following, reasoning, and writing through post-training and synthetic data. Benchmarks show Apple’s model scores lower than Llama 3, but with trusted human evaluations. Additionally, Meta released Llama 3.1 with a 405B parameter model, marking a significant open-source frontier model release.
Mini, Nemo, Turbo, Lite - Smol models go brrr (GPT4o version)
gpt-4o-mini mistral-nemo llama-3 llama-3-400b deepseek-v2 openai nvidia mistral-ai togethercompute deepseek-ai lmsys model-quantization context-windows instruction-following model-performance cost-efficiency multimodality benchmarking open-source model-release sam-altman
GPT-4o-mini launches with a 99% price reduction compared to text-davinci-003, offering 3.5% the price of GPT-4o and matching Opus-level benchmarks. It supports 16k output tokens, is faster than previous models, and will soon support text, image, video, and audio inputs and outputs. Mistral Nemo, a 12B parameter model developed with Nvidia, features a 128k token context window, FP8 checkpoint, and strong benchmark performance. Together Lite and Turbo offer fp8/int4 quantizations of Llama 3 with up to 4x throughput and significantly reduced costs. DeepSeek V2 is now open-sourced. Upcoming releases include at least 5 unreleased models and Llama 4 leaks ahead of ICML 2024.
Microsoft AgentInstruct + Orca 3
mistral-7b orca-2.5 microsoft-research apple tencent hugging-face synthetic-data fine-tuning instruction-following transformers model-performance hallucination-detection dataset-quality flashattention mixture-of-experts philschmid sama bindureddy rohanpaul_ai zachtratar dair_ai
Microsoft Research released AgentInstruct, the third paper in its Orca series, introducing a generative teaching pipeline that produces 25.8 million synthetic instructions to fine-tune mistral-7b, achieving significant performance gains: +40% AGIEval, +19% MMLU, +54% GSM8K, +38% BBH, +45% AlpacaEval, and a 31.34% reduction in hallucinations. This synthetic data approach follows the success of FineWeb and Apple's Rephrasing research in improving dataset quality. Additionally, Tencent claims to have generated 1 billion diverse personas for synthetic data. On AI Twitter, notable discussions included a shooting incident at a Trump rally and recent ML research highlights such as FlashAttention-3, RankRAG, and Mixture of A Million Experts.
Not much happened today.
phi-3-mini gpt4all-3.0 yi-large meta-3d-gen meta perplexity-ai microsoft gpt4all langchainai qdrant-engine 3d-generation long-context instruction-following reinforcement-learning-from-human-feedback persona-driven-data-synthesis meta-tuning model-steering memory-retrieval multivector-search universal-query-api rohanpaul_ai andriy_mulyar cwolferesearch sarahookr
Meta introduced Meta 3D Gen, a system for end-to-end generation of 3D assets from text in under 1 minute, producing high-quality 3D assets with detailed textures. Perplexity AI updated Pro Search to handle deeper research with multi-step reasoning and code execution. Microsoft improved Phi-3 Mini with better long-context understanding and instruction following. GPT4All 3.0 launched with support for thousands of models and major OS compatibility, featuring local file chat. Yi-Large model launched on Fireworks AI Playground. Research highlights include the evolution of reinforcement learning from human feedback (RLHF), persona-driven data synthesis using a billion diverse personas, meta-tuning for few-shot generalization, and steering vectors for model behavior control. Tools updates include LangSmith improving memory retrieval and Qdrant Engine v1.10 adding universal query API and multivector search.
GraphRAG: The Marriage of Knowledge Graphs and RAG
gemma-2 llama-3-70b claude-3.5-sonnet nemotron-340b qwen2-72b llama-3 microsoft-research anthropic nvidia hugging-face retrieval-augmented-generation knowledge-graphs token-usage inference-time attention-mechanisms instruction-following coding math long-range-reasoning synthetic-data dataset-release fine-tuning context-windows function-calling travis-fischer rasbt alexandr-wang osanseviero rohanpaul_ai hamelhusain svpino aaaazzam omarsar0
Microsoft Research open sourced GraphRAG, a retrieval augmented generation (RAG) technique that extracts knowledge graphs from sources and clusters them for improved LLM answers, though it increases token usage and inference time. Gemma 2 models were released focusing on efficient small LLMs with innovations like sliding window attention and RMS norm, nearly matching the larger Llama 3 70B. Anthropic's Claude 3.5 Sonnet leads in instruction following and coding benchmarks, while Nvidia's Nemotron 340B model was released in June. Qwen2-72B tops the HuggingFace Open LLM leaderboard excelling in math and long-range reasoning. Discussions on RAG highlighted its limitations and improvements in context usage via function calls. A persona-driven synthetic data generation approach introduced 1 billion personas, with a fine-tuned model matching GPT-4 performance on math benchmarks at 7B scale. The 200GB AutoMathText dataset was also noted for math data synthesis.
Shall I compare thee to a Sonnet's day?
claude-3.5-sonnet claude-3.5 gpt-4o gemini-1.5-pro anthropic lmsys glif comfyui hard-prompts json json-extraction meme-generation instruction-following app-development fusion-energy nuclear-fission productivity fchollet mustafasuleyman
Claude 3.5 Sonnet from Anthropic achieves top rankings in coding and hard prompt arenas, surpassing GPT-4o and competing with Gemini 1.5 Pro at lower cost. Glif demonstrates a fully automated Wojak meme generator using Claude 3.5 for JSON generation and ComfyUI for images, showcasing new JSON extractor capabilities. Artifacts enables rapid creation of niche apps, exemplified by a dual monitor visualizer made in under 5 minutes. François Chollet highlights that fusion energy is not a near-term solution compared to existing nuclear fission plants. Mustafa Suleyman notes that 75% of desk workers now use AI, marking a shift toward AI-assisted productivity.
Claude Crushes Code - 92% HumanEval and Claude.ai Artifacts
claude-3.5-sonnet claude-3-opus gpt-4o anthropic openai cognition benchmarking model-performance coding model-optimization fine-tuning instruction-following model-efficiency model-release api performance-optimization alex-albert
Claude 3.5 Sonnet, released by Anthropic, is positioned as a Pareto improvement over Claude 3 Opus, operating at twice the speed and costing one-fifth as much. It achieves state-of-the-art results on benchmarks like GPQA, MMLU, and HumanEval, surpassing even GPT-4o and Claude 3 Opus on vision tasks. The model demonstrates significant advances in coding capabilities, passing 64% of test cases compared to 38% for Claude 3 Opus, and is capable of autonomously fixing pull requests. Anthropic also introduced the Artifacts feature, enabling users to interact with AI-generated content such as code snippets and documents in a dynamic workspace, similar to OpenAI's Code Interpreter. This release highlights improvements in performance, cost-efficiency, and coding proficiency, signaling a growing role for LLMs in software development.
Nemotron-4-340B: NVIDIA's new large open models, built on syndata, great for syndata
nemotron-4-340b mixtral llama-3 gemini-1.5 gpt-4o mamba-2-hybrid-8b samba-3.8b-instruct dolphin-2.9.3 faro-yi-9b-dpo nvidia hugging-face mistral-ai llamaindex cohere gemini mistral synthetic-data model-alignment reward-models fine-tuning long-context model-scaling inference-speed mixture-of-agents open-source-models model-training instruction-following context-windows philipp-schmid bryan-catanzaro oleksii-kuchaiev rohanpaul_ai cognitivecompai _philschmid 01ai_yi
NVIDIA has scaled up its Nemotron-4 model from 15B to a massive 340B dense model, trained on 9T tokens, achieving performance comparable to GPT-4. The model alignment process uses over 98% synthetic data, with only about 20K human-annotated samples for fine-tuning and reward model training. The synthetic data generation pipeline is open-sourced, including synthetic prompts and preference data generation. The base and instruct versions outperform Mixtral and Llama 3, while the reward model ranks better than Gemini 1.5, Cohere, and GPT-4o. Other notable models include Mamba-2-Hybrid 8B, which is up to 8x faster than Transformers and excels on long-context tasks, Samba-3.8B-instruct for infinite context length with linear complexity, Dolphin-2.9.3 tiny models optimized for low-resource devices, and Faro Yi 9B DPO with a 200K context window running efficiently on 16GB VRAM. The Mixture-of-Agents technique boosts open-source LLMs beyond GPT-4 Omni on AlpacaEval 2.0.
Qwen 2 beats Llama 3 (and we don't know how)
qwen-2 llama-3 llama-3-70b gpt-4 nllb alibaba groq meta-ai-fair multilinguality benchmarking inference-speed sparse-autoencoders scaling-laws post-training instruction-following rejection-sampling execution-feedback model-release multilingual-models model-training philschmid huybery jonathanross321 awnihannun gdb nabla_theta ylecun
Alibaba released Qwen 2 models under Apache 2.0 license, claiming to outperform Llama 3 in open models with multilingual support in 29 languages and strong benchmark scores like MMLU 82.3 and HumanEval 86.0. Groq demonstrated ultra-fast inference speed on Llama-3 70B at 40,792 tokens/s and running 4 Wikipedia articles in 200ms. Research on sparse autoencoders (SAEs) for interpreting GPT-4 neural activity showed new training methods, metrics, and scaling laws. Meta AI announced the No Language Left Behind (NLLB) model capable of high-quality translations between 200 languages, including low-resource ones. "Our post-training phase is designed with the principle of scalable training with minimal human annotation," highlighting techniques like rejection sampling for math and execution feedback for coding.
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.
Perplexity, the newest AI unicorn
llama-3-8b llama-3-70b llama-3 llava-llama-3-8b-v1_1 phi-3 gpt-3.5 perplexity-ai meta-ai-fair hugging-face groq context-length fine-tuning quantization instruction-following model-comparison multimodality benchmarking memory-optimization model-performance daniel-gross aravind-srinivas
Perplexity doubles its valuation shortly after its Series B with a Series B-1 funding round. Significant developments around Llama 3 include context length extension to 16K tokens, new multimodal LLaVA models outperforming Llama 2, and fine-tuning improvements like QDoRA surpassing QLoRA. The Llama-3-70B model is praised for instruction following and performance across quantization formats. Phi-3 models by Meta AI released in multiple sizes show competitive benchmark results, with the 14B model achieving 78% on MMLU and the 3.8B model nearing GPT-3.5 performance.
Not much happened today
claude-3 claude-3-opus claude-3-sonnet gpt-4 gemma-2b anthropic perplexity langchain llamaindex cohere accenture mistral-ai snowflake together-ai hugging-face european-space-agency google gpt4all multimodality instruction-following out-of-distribution-reasoning robustness enterprise-ai cloud-infrastructure open-datasets model-deployment model-discoverability generative-ai image-generation
Anthropic released Claude 3, replacing Claude 2.1 as the default on Perplexity AI, with Claude 3 Opus surpassing GPT-4 in capability. Debate continues on whether Claude 3's performance stems from emergent properties or pattern matching. LangChain and LlamaIndex added support for Claude 3 enabling multimodal and tool-augmented applications. Despite progress, current models still face challenges in out-of-distribution reasoning and robustness. Cohere partnered with Accenture for enterprise AI search, while Mistral AI and Snowflake collaborate to provide LLMs on Snowflake's platform. Together AI Research integrates Deepspeed innovations to accelerate generative AI infrastructure. Hugging Face and the European Space Agency released a large earth observation dataset, and Google open sourced Gemma 2B, optimized for smartphones via the MLC-LLM project. GPT4All improved model discoverability for open models. The AI community balances excitement over new models with concerns about limitations and robustness, alongside growing enterprise adoption and open-source contributions. Memes and humor continue to provide social commentary.
Claude 3 just destroyed GPT 4 (see for yourself)
claude-3 claude-3-opus claude-3-sonnet claude-3-haiku gpt-4 anthropic amazon google claude-ai multimodality vision long-context model-alignment model-evaluation synthetic-data structured-output instruction-following model-speed cost-efficiency benchmarking safety mmitchell connor-leahy
Claude 3 from Anthropic launches in three sizes: Haiku (small, unreleased), Sonnet (medium, default on claude.ai, AWS, and GCP), and Opus (large, on Claude Pro). Opus outperforms GPT-4 on key benchmarks like GPQA, impressing benchmark authors. All models support multimodality with advanced vision capabilities, including converting a 2-hour video into a blog post. Claude 3 offers improved alignment, fewer refusals, and extended context length up to 1 million tokens with near-perfect recall. Haiku is noted for speed and cost-efficiency, processing dense research papers in under three seconds. The models excel at following complex instructions and producing structured outputs like JSON. Safety improvements reduce refusal rates, though some criticism remains from experts. Claude 3 is trained on synthetic data and shows strong domain-specific evaluation results in finance, medicine, and philosophy.
Miqu confirmed to be an early Mistral-medium checkpoint
miqu-1-70b mistral-medium llama-2-70b-chat mixtral sqlcoder-70b codellama-70b bagelmistery-tour-v2 psyfighter-v2 mistral-ai hugging-face nous-research aiatmeta instruction-following sampling-methods fp16-quantization fine-tuning model-training context-length text-to-sql model-performance model-optimization intrstllrninja
Miqu, an open access model, scores 74 on MMLU and 84.5 on EQ-Bench, sparking debates about its performance compared to Mistral Medium. The CEO of Mistral confirmed these results. Discussions in the TheBloke Discord highlight Miqu's superiority in instruction-following and sampling methods like dynatemp and min-p. Developers also explore browser preferences and Discord UI themes. Role-playing with models like BagelMistery Tour v2 and Psyfighter v2 is popular, alongside technical talks on fp16 quantization of Miqu-1-70b. Training and fine-tuning tips for models like Unsloth and Mistral 7B are shared. In the Nous Research AI Discord, the Activation Beacon method is discussed for extending LLM context length from 4K to 400K tokens. SQLCoder-70B, fine-tuned on CodeLlama-70B, leads in text-to-SQL generation and is available on Hugging Face. The Miqu model also impresses with an 83.5 EQ-Bench score, fueling speculation about its capabilities.