All tags
Company: "philschmid"
not much happened today
claude-4 claude-4-opus claude-4-sonnet gemini-2.5-pro gemma-3n imagen-4-ultra anthropic google-deepmind openai codebase-understanding coding agentic-performance multimodality text-to-speech video-generation model-integration benchmarking memory-optimization cline amanrsanger ryanpgreenblatt johnschulman2 alexalbert__ nearcyan mickeyxfriedman jeremyphoward gneubig teortaxesTex scaling01 artificialanlys philschmid
Anthropic's Claude 4 models (Opus 4, Sonnet 4) demonstrate strong coding abilities, with Sonnet 4 achieving 72.7% on SWE-bench and Opus 4 at 72.5%. Claude Sonnet 4 excels in codebase understanding and is considered SOTA on large codebases. Criticism arose over Anthropic's handling of ASL-3 security requirements. Demand for Claude 4 is high, with integration into IDEs and support from Cherry Studio and FastHTML. Google DeepMind introduced Gemini 2.5 Pro Deep Think and Gemma 3n, a mobile multimodal model reducing RAM usage by nearly 3x. Google's Imagen 4 Ultra ranks third in the Artificial Analysis Image Arena, available on Vertex AI Studio. Google also promoted Google Beam, an AI video model for immersive 3D experiences, and new text-to-speech models with multi-speaker support. The GAIA benchmark shows Claude 4 Opus and Sonnet leading in agentic performance.
Google I/O: new Gemini native voice, Flash, DeepThink, AI Mode (DeepSearch+Mariner+Astra)
gemini-2.5-pro gemini-2.5 google google-deepmind ai-assistants reasoning generative-ai developer-tools ai-integration model-optimization ai-application model-updates ai-deployment model-performance demishassabis philschmid jack_w_rae
Google I/O 2024 showcased significant advancements with Gemini 2.5 Pro and Deep Think reasoning mode from google-deepmind, emphasizing AI-driven transformations and developer opportunities. GeminiApp aims to become a universal AI assistant on the path to AGI, with new features like AI Mode in Google Search expanding generative AI access. The event included multiple keynotes and updates on over a dozen models and 20+ AI products, highlighting Google's leadership in AI innovation. Influential voices like demishassabis and philschmid provided insights and recaps, while the launch of Jules as a competitor to Codex/Devin was noted.
ChatGPT Codex, OpenAI's first cloud SWE agent
codex-1 openai-o3 codex-mini gemma-3 blip3-o qwen-2.5 marigold-iid deepseek-v3 lightlab gemini-2.0 lumina-next openai runway salesforce qwen deepseek google google-deepmind j1 software-engineering parallel-processing multimodality diffusion-models depth-estimation scaling-laws reinforcement-learning fine-tuning model-performance multi-turn-conversation reasoning audio-processing sama kevinweil omarsar0 iscienceluvr akhaliq osanseviero c_valenzuelab mervenoyann arankomatsuzaki jasonwei demishassabis philschmid swyx teortaxestex jaseweston
OpenAI launched Codex, a cloud-based software engineering agent powered by codex-1 (an optimized version of OpenAI o3) available in research preview for Pro, Enterprise, and Team ChatGPT users, featuring parallel task execution like refactoring and bug fixing. The Codex CLI was enhanced with quick sign-in and a new low-latency model, codex-mini. Gemma 3 is highlighted as the best open model runnable on a single GPU. Runway released the Gen-4 References API for style transfer in generation. Salesforce introduced BLIP3-o, a unified multimodal model family using diffusion transformers for CLIP image features. The Qwen 2.5 models (1.5B and 3B versions) were integrated into the PocketPal app with various chat templates. Marigold IID, a new state-of-the-art open-source depth estimation model, was released.
In research, DeepSeek shared insights on scaling and hardware for DeepSeek-V3. Google unveiled LightLab, a diffusion-based light source control in images. Google DeepMind's AlphaEvolve uses Gemini 2.0 to discover new math and reduce costs without reinforcement learning. Omni-R1 studied audio's role in fine-tuning audio LLMs. Qwen proposed a parallel scaling law inspired by classifier-free guidance. Salesforce released Lumina-Next on the Qwen base, outperforming Janus-Pro. A study found LLM performance degrades in multi-turn conversations due to unreliability. J1 is incentivizing LLM-as-a-Judge thinking via reinforcement learning. A new Qwen study correlates question and strategy similarity to predict reasoning strategies.
not much happened today
nemotron-h nvidia-eagle-2.5 gpt-4o qwen2.5-vl-72b gemini-2.5-flash gemini-2.0-pro gemini-exp-1206 gemma-3 qwen2.5-32b deepseek-r1-zero-32b uni3c seedream-3.0 adobe-dragon kimina-prover qwen2.5-72b bitnet-b1.58-2b4t nvidia deepseek hugging-face alibaba bytedance adobe transformers model-optimization multimodality long-context reinforcement-learning torch-compile image-generation diffusion-models distributional-rewards model-efficiency model-training native-quantization sampling-techniques philschmid arankomatsuzaki osanseviero iScienceLuvr akhaliq
Nemotron-H model family introduces hybrid Mamba-Transformer models with up to 3x faster inference and variants including 8B, 56B, and a compressed 47B model. Nvidia Eagle 2.5 is a frontier VLM for long-context multimodal learning, matching GPT-4o and Qwen2.5-VL-72B on long-video understanding. Gemini 2.5 Flash shows improved dynamic thinking and cost-performance, outperforming previous Gemini versions. Gemma 3 now supports torch.compile for about 60% faster inference on consumer GPUs. SRPO using Qwen2.5-32B surpasses DeepSeek-R1-Zero-32B on benchmarks with reinforcement learning only. Alibaba's Uni3C unifies 3D-enhanced camera and human motion controls for video generation. Seedream 3.0 by ByteDance is a bilingual image generation model with high-resolution outputs up to 2K. Adobe DRAGON optimizes diffusion generative models with distributional rewards. Kimina-Prover Preview is an LLM trained with reinforcement learning from Qwen2.5-72B, achieving 80.7% pass@8192 on miniF2F. BitNet b1.58 2B4T is a native 1-bit LLM with 2B parameters trained on 4 trillion tokens, matching full-precision LLM performance with better efficiency. Antidistillation sampling counters unwanted model distillation by modifying reasoning traces from frontier models.
DeepCoder: A Fully Open-Source 14B Coder at O3-mini Level
deepcoder-14b o3-mini o1 gemini-2.5-pro kimi-vl-a3b gpt-4o llama-4-scout maverick behemoth gen-4-turbo imagen-3 together-ai agentica opena bytedance google-deepmind moonshot-ai meta-ai-fair runway open-source reinforcement-learning code-generation multimodality model-training mixture-of-experts l2-normalization image-generation model-performance context-windows philschmid lepikhin reach_vb akhaliq yuchenj_uw epochairesearch danielhanchen c_valenzuelab
Together AI and Agentica released DeepCoder-14B, an open-source 14B parameter coding model rivaling OpenAI's o3-mini and o1 on coding benchmarks, trained with an open-source RL framework from ByteDance and costing about $26,880. Google DeepMind launched Gemini 2.5 Pro with experimental "Flash" versions available to subscribers. Moonshot AI introduced Kimi-VL-A3B, a multimodal model with 128K context outperforming gpt-4o on vision and math benchmarks. Meta AI released Llama 4 Scout and Maverick, with a larger Behemoth model in training, featuring mixture-of-experts and L2 norm techniques. Runway launched Gen-4 Turbo with 10x better results than Gen-3 at the same cost. Google announced Imagen 3, a high-quality text-to-image model now in Vertex AI, enabling easier object removal. The report highlights open-source contributions, reinforcement learning training optimizations, and significant model performance improvements across coding, multimodal, and image generation domains.
not much happened today
o3 o4-mini gpt-5 sonnet-3.7 gemma-3 qwen-2.5-vl gemini-2.5-pro gemma-7b llama-3-1-405b openai deepseek anthropic google meta-ai-fair inference-scaling reward-modeling coding-models ocr model-preview rate-limiting model-pricing architectural-advantage benchmarking long-form-reasoning attention-mechanisms mixture-of-experts gpu-throughput sama akhaliq nearcyan fchollet reach_vb philschmid teortaxestex epochairesearch omarsar0
OpenAI announced that o3 and o4-mini models will be released soon, with GPT-5 expected in a few months, delayed for quality improvements and capacity planning. DeepSeek introduced Self-Principled Critique Tuning (SPCT) to enhance inference-time scalability for generalist reward models. Anthropic's Sonnet 3.7 remains a top coding model. Google's Gemma 3 is available on KerasHub, and Qwen 2.5 VL powers a new Apache 2.0 licensed OCR model. Gemini 2.5 Pro entered public preview with increased rate limits and pricing announced, becoming a preferred model for many tasks except image generation. Meta's architectural advantage and the FrontierMath benchmark challenge AI's long-form reasoning and worldview development. Research reveals LLMs focus attention on the first token as an "attention sink," preserving representation diversity, demonstrated in Gemma 7B and LLaMa 3.1 models. MegaScale-Infer offers efficient serving of large-scale Mixture-of-Experts models with up to 1.90x higher per-GPU throughput.
not much happened today
grok-3 deepseek-r1 siglip-2 o3-mini-high r1-1776 llamba-1b llamba-3b llamba-8b llama-3 alphamaze audiobox-aesthetics xai nvidia google-deepmind anthropic openai bytedance ollama meta-ai-fair benchmarking model-releases performance reasoning multimodality semantic-understanding ocr multilinguality model-distillation recurrent-neural-networks visual-reasoning audio-processing scaling01 iscienceluvr philschmid arankomatsuzaki reach_vb mervenoyann wightmanr lmarena_ai ollama akhaliq
Grok-3, a new family of LLMs from xAI using 200,000 Nvidia H100 GPUs for advanced reasoning, outperforms models from Google, Anthropic, and OpenAI on math, science, and coding benchmarks. DeepSeek-R1 from ByteDance Research achieves top accuracy on the challenging SuperGPQA dataset. SigLIP 2 from GoogleDeepMind improves semantic understanding and OCR with flexible resolutions and multilingual capabilities, available on HuggingFace. OpenAI's o3-mini-high ranks #1 in coding and math prompts. Perplexity's R1 1776, a post-trained version of DeepSeek R1, is available on Ollama. The Llamba family distills Llama-3.x into efficient recurrent models with higher throughput. AlphaMaze combines DeepSeek R1 with GRPO for visual reasoning on ARC-AGI puzzles. Audiobox Aesthetics from Meta AI offers unified quality assessment for audio. The community notes that Grok 3's compute increase yields only modest performance gains.
TinyZero: Reproduce DeepSeek R1-Zero for $30
deepseek-r1 qwen o1 claude-3-sonnet claude-3 prime ppo grpo llama-stack deepseek berkeley hugging-face meta-ai-fair openai deeplearningai reinforcement-learning fine-tuning chain-of-thought multi-modal-benchmark memory-management model-training open-source agentic-workflow-automation model-performance jiayi-pan saranormous reach_vb lmarena_ai nearcyan omarsar0 philschmid hardmaru awnihannun winglian
DeepSeek Mania continues to reshape the frontier model landscape with Jiayi Pan from Berkeley reproducing the OTHER result from the DeepSeek R1 paper, R1-Zero, in a cost-effective Qwen model fine-tune for two math tasks. A key finding is a lower bound to the distillation effect at 1.5B parameters, with RLCoT reasoning emerging as an intrinsic property. Various RL techniques like PPO, DeepSeek's GRPO, or PRIME show similar outcomes, and starting from an Instruct model speeds convergence. The Humanity’s Last Exam (HLE) Benchmark introduces a challenging multi-modal test with 3,000 expert-level questions across 100+ subjects, where models perform below 10%, with DeepSeek-R1 achieving 9.4%. DeepSeek-R1 excels in chain-of-thought reasoning, outperforming models like o1 while being 20x cheaper and MIT licensed. The WebDev Arena Leaderboard ranks DeepSeek-R1 #2 in technical domains and #1 under Style Control, closing in on Claude 3.5 Sonnet. OpenAI's Operator is deployed to 100% of Pro users in the US, enabling tasks like ordering meals and booking reservations, and functions as a research assistant for AI paper searches and summaries. Hugging Face announces a leadership change after significant growth, while Meta AI releases the first stable version of Llama Stack with streamlined upgrades and automated verification. DeepSeek-R1's open-source success is celebrated, and technical challenges like memory management on macOS 15+ are addressed with residency sets in MLX for stability.
Bespoke-Stratos + Sky-T1: The Vicuna+Alpaca moment for reasoning
sky-t1-32b-preview qwen-2.5-32b r1 o1-preview gpt-4o claude-3-sonnet bespoke-stratos-32b gemini-2.0-flash-thinking berkeley usc deepseek bespoke-labs google llmsys stanford lm-sys reasoning supervised-finetuning reinforcement-learning multimodality model-distillation context-windows code-execution model-repeatability behavioral-self-awareness rlhf teortaxestex cwolferesearch madiator chakraai philschmid abacaj omarsar0
Reasoning Distillation has emerged as a key technique, with Berkeley/USC researchers releasing Sky-T1-32B-Preview, a finetuned model of Qwen 2.5 32B using 17k reasoning traces for just $450, matching benchmarks of o1-preview. DeepSeek introduced R1, a model surpassing o1-preview and enabling distillation to smaller models like a 1.5B Qwen to match gpt-4o and claude-3-sonnet levels. Bespoke Labs further distilled R1 on Qwen, outperforming o1-preview with fewer samples. This progress suggests that "SFT is all you need" for reasoning without major architecture changes. Additionally, DeepSeek-R1 uses pure reinforcement learning with supervised finetuning to accelerate convergence and shows strong reasoning and multimodal capabilities. Google's Gemini 2.0 Flash Thinking model boasts a 1 million token context window, code execution, and excels in math, science, and multimodal reasoning. Critiques highlight challenges in model repeatability, behavioral self-awareness, and RLHF limitations in reasoning robustness.
not much happened today
helium-1 qwen-2.5 phi-4 sky-t1-32b-preview o1 codestral-25.01 phi-3 mistral llama-3 gpt-3.5 llama-3 gpt-3.5 llmquoter kyutai-labs lmstudio mistralai llamaindex huggingface langchainai hyperbolic-labs replit fchollet philschmid multilinguality token-level-distillation context-windows model-performance open-source reasoning coding retrieval-augmented-generation hybrid-retrieval multiagent-systems video large-video-language-models dynamic-ui voice-interaction gpu-rentals model-optimization semantic-deduplication model-inference reach_vb awnihannun lior_on_ai sophiamyang omarsar0 skirano yuchenj_uw fchollet philschmid
Helium-1 Preview by kyutai_labs is a 2B-parameter multilingual base LLM outperforming Qwen 2.5, trained on 2.5T tokens with a 4096 context size using token-level distillation from a 7B model. Phi-4 (4-bit) was released in lmstudio on an M4 max, noted for speed and performance. Sky-T1-32B-Preview is a $450 open-source reasoning model matching o1's performance with strong benchmark scores. Codestral 25.01 by mistralai is a new SOTA coding model supporting 80+ programming languages and offering 2x speed.
Innovations include AutoRAG for optimizing retrieval-augmented generation pipelines, Agentic RAG for autonomous query reformulation and critique, Multiagent Finetuning using societies of models like Phi-3, Mistral, LLaMA-3, and GPT-3.5 for reasoning improvements, and VideoRAG incorporating video content into RAG with LVLMs.
Applications include a dynamic UI AI chat app by skirano on Replit, LangChain tools like DocTalk for voice PDF conversations, AI travel agent tutorials, and news summarization agents. Hyperbolic Labs offers competitive GPU rentals including H100, A100, and RTX 4090. LLMQuoter enhances RAG accuracy by identifying key quotes.
Infrastructure updates include MLX export for LLM inference from Python to C++ by fchollet and SemHash semantic text deduplication by philschmid.
Moondream 2025.1.9: Structured Text, Enhanced OCR, Gaze Detection in a 2B Model
o1 vdr-2b-multi-v1 llava-mini openai llamaindex langchainai qdrant genmoai vision model-efficiency structured-output gaze-detection reasoning model-distillation multimodality embedding-models gan diffusion-models self-attention training-optimizations development-frameworks api cross-language-deployment semantic-search agentic-document-processing developer-experience philschmid saranormous jxmnop reach_vb iscienceluvr multimodalart arohan adcock_brett awnihannun russelljkaplan ajayj_
Moondream has released a new version that advances VRAM efficiency and adds structured output and gaze detection, marking a new frontier in vision model practicality. Discussions on Twitter highlighted advancements in reasoning models like OpenAI's o1, model distillation techniques, and new multimodal embedding models such as vdr-2b-multi-v1 and LLaVA-Mini, which significantly reduce computational costs. Research on GANs and decentralized diffusion models showed improved stability and performance. Development tools like MLX and vLLM received updates for better portability and developer experience, while frameworks like LangChain and Qdrant enable intelligent data workflows. Company updates include new roles and team expansions at GenmoAI. "Efficiency tricks are all you need."
not much happened this weekend
o3 o1 opus sonnet octave openai langchain hume x-ai amd nvidia meta-ai-fair hugging-face inference-time-scaling model-ensembles small-models voice-cloning fine-math-dataset llm-agent-framework benchmarking software-stack large-concept-models latent-space-reasoning mechanistic-interpretability planning speech-language-models lisa-su clementdelangue philschmid neelnanda5
o3 model gains significant attention with discussions around its capabilities and implications, including an OpenAI board member referencing "AGI." LangChain released their State of AI 2024 survey. Hume announced OCTAVE, a 3B parameter API-only speech-language model with voice cloning. x.ai secured a $6B Series C funding round. Discussions highlight inference-time scaling, model ensembles, and the surprising generalization ability of small models. New tools and datasets include FineMath, the best open math dataset on Hugging Face, and frameworks for LLM agents. Industry updates cover a 5-month benchmarking of AMD MI300X vs Nvidia H100 + H200, insights from a meeting with Lisa Su on AMD's software stack, and open AI engineering roles. Research innovations include Large Concept Models (LCM) from Meta AI, Chain of Continuous Thought (Coconut) for latent space reasoning, and mechanistic interpretability initiatives.
ModernBert: small new Retriever/Classifier workhorse, 8k context, 2T tokens,
modernbert gemini-2.0-flash-thinking o1 llama answerdotai lightonio hugging-face google-deepmind openai meta-ai-fair figure encoder-only-models long-context alternating-attention natural-language-understanding reasoning robotics-simulation physics-engine humanoid-robots model-performance model-releases jeremyphoward alec-radford philschmid drjimfan bindureddy
Answer.ai/LightOn released ModernBERT, an updated encoder-only model with 8k token context, trained on 2 trillion tokens including code, with 139M/395M parameters and state-of-the-art performance on retrieval, NLU, and code tasks. It features Alternating Attention layers mixing global and local attention. Gemini 2.0 Flash Thinking debuted as #1 in Chatbot Arena, and the O1 model scored top in reasoning benchmarks. Llama downloads surpassed 650 million, doubling in 3 months. OpenAI launched desktop app integrations with voice capabilities. Figure delivered its first humanoid robots commercially. Advances in robotics simulation and a new physics engine Genesis claiming 430,000x faster than real-time were highlighted.
Olympus has dropped (aka, Amazon Nova Micro|Lite|Pro|Premier|Canvas|Reel)
amazon-nova claude-3 llama-3-70b gemini-1.5-flash gpt-4o amazon anthropic google-deepmind sakana-ai-labs multimodality benchmarking model-merging model-performance model-architecture model-optimization population-based-learning philschmid bindureddy
Amazon announced the Amazon Nova family of multimodal foundation models at AWS Re:Invent, available immediately with no waitlist in configurations like Micro, Lite, Pro, Canvas, and Reel, with Premier and speech-to-speech coming next year. These models offer 2-4x faster token speeds and are 25%-400% cheaper than competitors like Anthropic Claude models, positioning Nova as a serious contender in AI engineering. Pricing undercuts models such as Google DeepMind Gemini Flash 8B, and some Nova models extend context length up to 300k tokens. However, benchmarking controversy exists as some evaluations show Nova scoring below Llama-3 70B in LiveBench AI metrics. Separately, CycleQD was introduced by Sakana AI Labs, using evolutionary computation for population-based model merging to develop niche LLM agents.
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.
FrontierMath: A Benchmark for Evaluating Advanced Mathematical Reasoning in AI
o1 claude-3.5-haiku gpt-4o epoch-ai openai microsoft anthropic x-ai langchainai benchmarking math moravecs-paradox mixture-of-experts chain-of-thought agent-framework financial-metrics-api pdf-processing few-shot-learning code-generation karpathy philschmid adcock_brett dylan522p
Epoch AI collaborated with over 60 leading mathematicians to create the FrontierMath benchmark, a fresh set of hundreds of original math problems with easy-to-verify answers, aiming to challenge current AI models. The benchmark reveals that all tested models, including o1, perform poorly, highlighting the difficulty of complex problem-solving and Moravec's paradox in AI. Key AI developments include the introduction of Mixture-of-Transformers (MoT), a sparse multi-modal transformer architecture reducing computational costs, and improvements in Chain-of-Thought (CoT) prompting through incorrect reasoning and explanations. Industry news covers OpenAI acquiring the chat.com domain, Microsoft launching the Magentic-One agent framework, Anthropic releasing Claude 3.5 Haiku outperforming gpt-4o on some benchmarks, and xAI securing 150MW grid power with support from Elon Musk and Trump. LangChain AI introduced new tools including a Financial Metrics API, Document GPT with PDF upload and Q&A, and LangPost AI agent for LinkedIn posts. xAI also demonstrated the Grok Engineer compatible with OpenAI and Anthropic APIs for code generation.
not much happened this weekend
claude-3.5-sonnet llama-3 llama-3-8b notebookllama min-omni-2 moondream openai anthropic hugging-face mistral-ai google-deepmind langchain deepmind microsoft pattern-recognition reinforcement-learning prompt-optimization text-to-speech model-optimization tensor-parallelism hyperparameters multimodal modal-alignment multimodal-fine-tuning ai-productivity privacy generative-ai rag retrieval-augmentation enterprise-text-to-sql amanda-askell philschmid stasbekman francois-fleuret mervenoyann reach_vb dzhng aravsrinivas sama lateinteraction andrew-y-ng bindureddy jerryjliu0
Moondream, a 1.6b vision language model, secured seed funding, highlighting a trend in moon-themed tiny models alongside Moonshine (27-61m ASR model). Claude 3.5 Sonnet was used for AI Twitter recaps. Discussions included pattern recognition vs. intelligence in LLMs, reinforcement learning for prompt optimization, and NotebookLlama, an open-source NotebookLM variant using LLaMA models for tasks like text-to-speech. Advances in model optimization with async-TP in PyTorch for tensor parallelism and hyperparameter tuning were noted. Mini-Omni 2 demonstrated multimodal capabilities across image, audio, and text for voice conversations with emphasis on modal alignment and multimodal fine-tuning. AI productivity tools like an AI email writer and LlamaCloud-based research assistants were introduced. Emphasis on practical skill development and privacy-conscious AI tool usage with Llama3-8B was highlighted. Generative AI tools such as #AIPythonforBeginners and GenAI Agents with LangGraph were shared. Business insights covered rapid execution in AI product development and emerging AI-related job roles. Challenges in enterprise-grade text-to-SQL and advanced retrieval methods were discussed with tutorials on RAG applications using LangChain and MongoDB.
Claude 3.5 Sonnet (New) gets Computer Use
claude-3.5-sonnet claude-3.5-haiku llama-3.1 nemotron anthropic zep nvidia coding benchmarks computer-use vision multimodal-memory model-updates ai-integration philschmid swyx
Anthropic announced new Claude 3.5 models: 3.5 Sonnet and 3.5 Haiku, improving coding performance significantly, with Sonnet topping several coding benchmarks like Aider and Vectara. The new Computer Use API enables controlling computers via vision, scoring notably higher than other AI systems, showcasing progress in AI-driven computer interaction. Zep launched a cloud edition for AI agents memory management, highlighting challenges in multimodal memory. The update also mentions Llama 3.1 and Nemotron models from NVIDIA.
Did Nvidia's Nemotron 70B train on test?
nemotron-70b llama-3.1-70b llama-3.1 ministral-3b ministral-8b gpt-4o claude-3.5-sonnet claude-3.5 nvidia mistral-ai hugging-face zep benchmarking reinforcement-learning reward-models temporal-knowledge-graphs memory-layers context-windows model-releases open-source reach_vb philschmid swyx
NVIDIA's Nemotron-70B model has drawn scrutiny despite strong benchmark performances on Arena Hard, AlpacaEval, and MT-Bench, with some standard benchmarks like GPQA and MMLU Pro showing no improvement over the base Llama-3.1-70B. The new HelpSteer2-Preference dataset improves some benchmarks with minimal losses elsewhere. Meanwhile, Mistral released Ministral 3B and 8B models featuring 128k context length and outperforming Llama-3.1 and GPT-4o on various benchmarks under the Mistral Commercial License. NVIDIA's Nemotron 70B also surpasses GPT-4o and Claude-3.5-Sonnet on key benchmarks using RLHF (REINFORCE) training. Additionally, Zep introduced Graphiti, an open-source temporal knowledge graph memory layer for AI agents, built on Neo4j.
not much happened today
flux-schnell meta-ai-fair anthropic togethercompute hugging-face audio-generation quantization prompt-caching long-term-memory llm-serving-framework hallucination-detection ai-safety ai-governance geoffrey-hinton john-hopfield demis-hassabis rohanpaul_ai svpino hwchase17 shreyar philschmid mmitchell_ai bindureddy
Geoffrey Hinton and John Hopfield won the Nobel Prize in Physics for foundational work on neural networks linking AI and physics. Meta AI introduced a 13B parameter audio generation model as part of Meta Movie Gen for video-synced audio. Anthropic launched the Message Batches API enabling asynchronous processing of up to 10,000 queries at half the cost. Together Compute released Flux Schnell, a free model for 3 months. New techniques like PrefixQuant quantization and Prompt Caching for low-latency inference were highlighted by rohanpaul_ai. LangGraph added long-term memory support for persistent document storage. Hex-LLM framework was introduced for TPU-based low-cost, high-throughput LLM serving from Hugging Face models. Discussions on AI safety emphasized gender equality in science, and concerns about premature AI regulation by media and Hollywood were raised.
The AI Nobel Prize
claude-3.5-sonnet reka-flash got openai anthropic reka-ai zep artificial-neural-networks nobel-prize knowledge-graphs memory-layers real-time-voice-api vision fine-tuning prompt-caching multimodality function-calling ocr open-source single-sign-on software-testing ai-assisted-coding ai-ethics geoff-hinton john-hopfield philschmid alexalbert mervenoyann clementdelangue svpino bindureddy ylecun rohanpaul_ai
Geoff Hinton and John Hopfield won the Nobel Prize in Physics for their work on Artificial Neural Networks. The award citation spans 14 pages highlighting their contributions. Zep released a new community edition of their low-latency memory layer for AI agents, emphasizing knowledge graphs for memory. At OpenAI's DevDay, new features like real-time voice API, vision model fine-tuning, and prompt caching with a 50% discount on reused tokens were introduced. Anthropic's Claude 3.5 Sonnet was recognized as the best model currently. Reka AI Labs updated their Reka Flash model with enhanced multimodal and function calling capabilities. The GOT (Generic OCR Transformer) achieved 98.79% accuracy on OCR benchmarks. Discussions on open-source AI models highlighted their role in fostering competition and decentralization. Software development insights included the importance of Single Sign-On (SSO), thorough testing, and AI-assisted coding workflows. Ethical and societal topics covered critiques of tax policies and the appointment of France's first Minister of AI.
a calm before the storm
o1 o1-mini qwen2.5 gpt-4 llama-2-70b llama-7b anthropic openai alibaba microsoft blackrock groq aramco disney eth-zurich pudu-robotics slack long-context kv-cache-quantization diffusion-models reinforcement-learning robotics ai-integration multilinguality model-benchmarking model-performance model-optimization adcock_brett philschmid rohanpaul_ai jvnixon kateclarktweets sama
Anthropic is raising funds at a valuation up to $40 billion ahead of anticipated major releases. OpenAI launched new reasoning models o1 and o1-mini, with increased rate limits and a multilingual MMLU benchmark. Alibaba released the open-source Qwen2.5 model supporting 29+ languages, showing competitive performance to gpt-4 at lower cost. Microsoft and Blackrock plan to invest $30 billion in AI data centers, with Groq partnering with Aramco to build the world's largest AI inference center. Robotics advances include Disney Research and ETH Zurich's diffusion-based motion generation for robots and Pudu Robotics' semi-humanoid robot. Slack and Microsoft introduced AI-powered agents integrated into their platforms. Research highlights include long-context scaling for llama-2-70b using Dual Chunk Attention and KV cache quantization enabling 1 million token context on llama-7b models.
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."
a quiet weekend
o1 datagemma aloha demostart firefly-ai-video-model pixtral-12b gamegen-o openai google-deepmind adobe mistral-ai tencent supermaven 11x cohere anthropic latent-space-university stanford microsoft mila notre-dame reinforcement-learning chain-of-thought reasoning robotics diffusion-models multimodality video-generation model-training reflection-tuning mathematical-reasoning model-benchmarking fine-tuning george-hotz terence-tao adcock_brett rohanpaul_ai bindureddy fchollet philschmid
OpenAI released the new o1 model, leveraging reinforcement learning and chain-of-thought prompting to excel in reasoning benchmarks, achieving an IQ-like score of 120. Google DeepMind introduced DataGemma to reduce hallucinations by connecting LLMs with real-world data, and unveiled ALOHA and DemoStart for robot dexterity using diffusion methods. Adobe previewed its Firefly AI Video Model with text-to-video and generative extend features. Mistral launched the multimodal Pixtral 12B model, and Tencent presented the GameGen-O open-world video game generation model. Several research papers from Stanford, OpenAI, Microsoft, Mila, and Notre Dame focus on advanced reasoning, self-verification, and reflection tuning techniques. Experts like Terence Tao and George Hotz have shared mixed but optimistic views on o1's capabilities. Seed funding rounds include Supermaven ($12M) and 11x ($24M).
not much happened today
llama-3-1 claude-3-5-sonnet llama-3-1-405b ltm-2-mini qwen2-vl gpt-4o-mini meta-ai-fair hugging-face magic-ai-labs lmsys alibaba openai long-context style-control multimodality ai-safety model-evaluation web-crawling pdf-processing ai-hype-cycles call-center-automation sam-altman ajeya-cotra fchollet rohanpaul_ai philschmid
Meta announced significant adoption of LLaMA 3.1 with nearly 350 million downloads on Hugging Face. Magic AI Labs introduced LTM-2-Mini, a long context model with a 100 million token context window, and a new evaluation method called HashHop. LMSys added style control to their Chatbot Arena leaderboard, improving rankings for models like Claude 3.5 Sonnet and LLaMA 3.1 405B. Alibaba released Qwen2-VL, a multimodal LLM under Apache 2.0 license, competitive with GPT-4o mini. OpenAI CEO Sam Altman announced collaboration with the US AI Safety Institute for pre-release model testing. Discussions on AI safety and potential AI takeover risks were highlighted by Ajeya Cotra. Tools like firecrawl for web crawling and challenges in PDF processing were noted. AI hype cycles and market trends were discussed by François Chollet, and potential AI disruption in call centers was shared by Rohan Paul.
CogVideoX: Zhipu's Open Source Sora
cogvideox llama-3-1 llama-3-405b moondream phi-3.5 llama-rank zhipu-ai alibaba meta-ai-fair google hugging-face nvidia togethercompute salesforce video-generation serverless-computing vision document-vqa text-vqa mixture-of-experts retrieval-augmented-generation long-context model-routing webgpu background-removal long-form-generation superposition-prompting rohanpaul_ai philschmid vikhyatk algo_diver jayalammar davidsholz
Zhipu AI, Alibaba's AI arm and China's 3rd largest AI lab, released the open 5B video generation model CogVIdeoX, which can run without GPUs via their ChatGLM web and desktop apps. Meta AI announced trust & safety research and CyberSecEval 3 alongside the release of Llama 3.1, with Llama 3 405B now available serverless on Google Cloud Vertex AI and Hugging Face x NVIDIA NIM API. Updates include Moondream, an open vision-language model improving DocVQA and TextVQA tasks, and the lightweight MoE chat model Phi-3.5 with 16x3.8B parameters. Together Compute introduced the Rerank API featuring Salesforce's LlamaRank model for document and code ranking. Research highlights include superposition prompting for RAG without fine-tuning, the AgentWrite pipeline for long-form content generation over 20,000 words, and a comparison showing Long Context methods outperform RAG at higher costs. Tools include Not Diamond, an AI model router, AI command line interfaces, and an open-source WebGPU background removal tool. "You don't even need GPUs to run it," referring to CogVIdeoX.
Llama 3.1 Leaks: big bumps to 8B, minor bumps to 70b, and SOTA OSS 405b model
llama-3-1-405b llama-3-8b llama-3-70b llama-3-1-8b gpt-4o gpt-4o-mini claude-3-5 qwen-2 meta-ai-fair openai alibaba multilinguality code-generation context-windows model-training synthetic-data benchmarking reasoning fine-tuning model-performance dataset-release swyx philschmid jjitsev lewtun teknium1 adcock_brett
Llama 3.1 leaks reveal a 405B dense model with 128k context length, trained on 39.3M GPU hours using H100-80GB GPUs, and fine-tuned with over 25M synthetic examples. The model shows significant benchmark improvements, especially for the 8B and 70B variants, with some evals suggesting the 70B outperforms GPT-4o. GPT-4o Mini launched as a cost-efficient variant with strong performance but some reasoning weaknesses. Synthetic datasets like NuminaMath enable models such as Alibaba Qwen 2 to surpass GPT-4o and Claude 3.5 in math competitions. Discussions include reasoning task benchmarks and dataset building for improved reasoning.
DataComp-LM: the best open-data 7B model/benchmark/dataset
mistral-nemo-12b gpt-4o-mini deepseek-v2-0628 mistral-7b llama-3 gemma-2 qwen-2 datacomp hugging-face openai nvidia mistral-ai deepseek dataset-design scaling-laws model-benchmarking model-performance fine-tuning multilinguality function-calling context-windows open-source-models model-optimization cost-efficiency benchmarking sam-altman guillaume-lample philschmid miramurati
DataComp team released a competitive 7B open data language model trained on only 2.5T tokens from the massive DCLM-POOL dataset of 240 trillion tokens, showing superior scaling trends compared to FineWeb. OpenAI launched GPT-4o mini, a cost-effective model with 82% MMLU and performance near GPT-4-Turbo, aimed at developers for broad applications. NVIDIA and Mistral jointly released the Mistral NeMo 12B model featuring a 128k token context window, FP8 checkpoint, multilingual support, and Apache 2.0 licensing. DeepSeek announced DeepSeek-V2-0628 as the top open-source model on the LMSYS Chatbot Arena leaderboard with strong rankings in coding, math, and hard prompts. This news highlights advances in dataset design, model efficiency, and open-source contributions in the AI community.
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.
There's Ilya!
chameleon-7b chameleon-34b deepseek-coder-v2 gpt-4-turbo claude-3-opus voco-llama safe-superintelligence-inc openai anthropic meta deepseek google-deepmind parallel-decoding code-generation quantization training-dynamics vision benchmarks datasets image-captioning reasoning memory-optimization ilya-sutskever jan-leike ylecun akhaliq philschmid rohanpaul_ai mervenoyann fchollet
Ilya Sutskever has co-founded Safe Superintelligence Inc shortly after leaving OpenAI, while Jan Leike moved to Anthropic. Meta released new models including Chameleon 7B and 34B with mixed-modal input and unified token space quantization. DeepSeek-Coder-V2 shows code capabilities comparable to GPT-4 Turbo, supporting 338 programming languages and 128K context length. Consistency Large Language Models (CLLMs) enable parallel decoding generating multiple tokens per step. Grokked Transformers demonstrate reasoning through training dynamics affecting memory formation and generalization. VoCo-LLaMA compresses vision tokens with LLMs improving video temporal correlation understanding. The BigCodeBench benchmark evaluates LLMs on 1,140 coding tasks across 139 Python libraries, topped by DeepSeek-Coder-V2 and Claude 3 Opus. PixelProse is a large 16M image-caption dataset with reduced toxicity.
Francois Chollet launches $1m ARC Prize
gpt-4 chatgpt openai apple togethercompute benchmarking agi pattern-recognition skill-acquisition privacy on-device-ai mixed-precision-quantization mixture-of-experts multimodality agentic-ai francois-chollet karpathy svpino philschmid clementdelangue sama gdb miramurati kevin-weil sarah-friar
François Chollet critiques current paths to AGI, emphasizing the importance of benchmarks that resist saturation and focus on skill acquisition and open-ended problem solving. The ARC-AGI puzzles exemplify "easy for humans, hard for AI" challenges to measure progress toward AGI. Meanwhile, Apple announces integration of ChatGPT into iOS, iPadOS, and macOS through a partnership with OpenAI, enabling AI-powered features like document summarization and photo analysis with privacy-preserving measures. Discussions highlight Apple's focus on deep AI integration and on-device models optimized with techniques like mixed-precision quantization, though some skepticism remains about their AI capabilities compared to GPT-4. Additionally, Together Compute introduces a Mixture of Agents approach achieving strong performance 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.
Skyfall
gemini-1.5-pro gemini-1.5-flash yi-1.5 kosmos-2.5 paligemma falcon-2 deepseek-v2 hunyuan-dit gemini-1.5 gemini-1.5-flash yi-1.5 google-deepmind yi-ai microsoft hugging-face langchain maven multimodality mixture-of-experts transformer model-optimization long-context model-performance model-inference fine-tuning local-ai scaling-laws causal-models hallucination-detection model-distillation model-efficiency hamel-husain dan-becker clement-delangue philschmid osanseviero arankomatsuzaki jason-wei rohanpaul_ai
Between 5/17 and 5/20/2024, key AI updates include Google DeepMind's Gemini 1.5 Pro and Flash models, featuring sparse multimodal MoE architecture with up to 10M context and a dense Transformer decoder that is 3x faster and 10x cheaper. Yi AI released Yi-1.5 models with extended context windows of 32K and 16K tokens. Other notable releases include Kosmos 2.5 (Microsoft), PaliGemma (Google), Falcon 2, DeepSeek v2 lite, and HunyuanDiT diffusion model. Research highlights feature an Observational Scaling Laws paper predicting model performance across families, a Layer-Condensed KV Cache technique boosting inference throughput by up to 26×, and the SUPRA method converting LLMs into RNNs for reduced compute costs. Hugging Face expanded local AI capabilities enabling on-device AI without cloud dependency. LangChain updated its v0.2 release with improved documentation. The community also welcomed a new LLM Finetuning Discord by Hamel Husain and Dan Becker for Maven course users. "Hugging Face is profitable, or close to profitable," enabling $10 million in free shared GPUs for developers.
Less Lazy AI
hamster-v0.2 flan-t5 miqu-1-120b-gguf qwen2 axolotl openai hugging-face nous-research h2oai apple model-merging fine-tuning quantization vram-optimization plugin-development chatbot-memory model-training bug-reporting api-compatibility philschmid
The AI Discord summaries for early 2024 cover various community discussions and developments. Highlights include 20 guilds, 308 channels, and 10449 messages analyzed, saving an estimated 780 minutes of reading time. Key topics include Polymind Plugin Puzzle integrating PubMed API, roleplay with HamSter v0.2, VRAM challenges in Axolotl training, fine-tuning tips for FLAN-T5, and innovative model merging strategies. The Nous Research AI community discussed GPT-4's lyricism issues, quantization techniques using
llama.cpp, frankenmerging with models like miqu-1-120b-GGUF, anticipation for Qwen2, and tools like text-generation-webui and ExLlamaV2. The LM Studio community reported a bug where the app continues running after UI closure, with a workaround to forcibly terminate the process. These discussions reflect ongoing challenges and innovations in AI model training, deployment, and interaction.