TSMZ @tsmztech

ASU researchers find that Conditional Inference Thinking (CoT) in language models is a pattern-matching tool incapable of generating new logic when tested with data outside training. #cot #nlp #llm #ailab

Researchers develop OpenCUA, an open-source framework to facilitate the creation and training of computer-use agents. This addresses data collection challenges for AI systems performing tasks on computers. #ai #opensource

OpenAI and Anthropic cross-evaluated their foundation models, revealing reasoning models resist jailbreaking better than general chat models. #artificialintelligence #machinelearning

NVIDIA reports strong Q2 revenue of $46.7B but faces growing competition from custom ASICs and regulatory challenges in China. #nvidia #asiccompetition #chinamarket

Launch of Hermes 4 by Nous Research challenges corporate AI dominance with high performance and lower computational needs. #ai #nousresearch #hermes4

ByteDance's Seed Team releases Seed-OSS-36B on Hugging Face with native long-context up to 512K tokens. Available in three variants for enhanced developer usability and advanced reasoning needs. #largelanguage #longcontext

Nvidia reports strong Q2 earnings with $46.7B revenue and a 56% year-over-year increase in data center revenues, but faces growing competition from AI-specific ASICs like those developed by Broadcom and Google. #nvidia #q2earnings #asiccompetition

Memp introduces procedural memory to LLM agents, enhancing efficiency and reliability through continuous learning from past experiences. #ai #llm #proceduralmemory

DeepSeek V3.1 showcases how technological merit trumps national boundaries in AI development, democratizing innovation and raising ethical questions. #ai #opensource #innovation #ethics

A new Memp technique from Zhejiang University and Alibaba Group enables LLM agents to handle complex tasks more efficiently with procedural memory. #ai #proceduralmemory #llmagents

NVIDIA reports Q2 revenue of $46.7B with data center revenues at $41.1B; facing growing competition from ASICs and proprietary hardware. #nvidia #revenue #datacenter #aiinfrastructure

OpenAI introduces gpt-realtime model for more expressive and instruction-following voice AI, improving natural-sounding voices and complex instruction understanding. Pricing is also reduced by 20%. #voiceai #gptrealtime #openai

MIT report reveals employees are leading AI productivity gains with personal tools despite high failure rates of enterprise initiatives. #ai #employeeadoption

Google releases 'Nano Banana' update to Gemini app for enhanced image editing control and consistency in enterprise use. #google #gemini #nanobanana

Study shows large language models (LLMs) generate fluent but logically flawed responses when reasoning outside their training data distribution. #ai #machinelearning

CodeSignal launches Cosmo, a mobile app providing AI-powered micro-courses to bridge skill gaps in the rapidly evolving job market. #education #learningapps

CZI launches rBio, an open-source AI model for biological research and drug discovery. #ai #biologicalresearch #drugdiscovery

Salesforce launches CRMArena-Pro, an AI 'Flight Simulator' for rigorous testing of enterprise agents before deployment. Aims to boost success rates amidst high failure rate of pilot projects. #ai #syntheticdata #enterprise #crmaresnapro

GEPA: Genetic-Evolution Pareto optimizes tasks with large language models using genetic prompt evolution and natural language feedback for efficiency. #ai #optimization #languagemodels

Introducing OpenCUA: a new open-source framework for creating advanced computer-use agents with AgentNet and Chain-of-Thought reasoning. #ai #computerauseagents