AITOMATIC

“Generative AI will have a bigger impact on the physical industrial world than on the digital one. The future is here, and it’s industrial!” #GenerativeAI #IndustrialAI Watch it here: https://lnkd.in/gSyug94n

“Domain-specific AI agents are like seasoned experts, not fresh PhDs. Experience and knowledge are key to solving real industrial problems.” #AI #DomainSpecificAI Watch it here: https://lnkd.in/g6tBH2Td

“AI in industrial applications requires consistency, not creativity. The future lies in domain-specific agents that understand our processes deeply.” #AI #IndustrialAI Watch it here: https://lnkd.in/gBAka39D

“America faces a de-industrialization crisis, but AI can help us leapfrog the old burdens in manufacturing. Let’s seize the opportunity!” #AI #IndustrialRevolution Watch it here: https://lnkd.in/gjjvYHqx

Come find the AITOMATIC booth at Industrial AI Conferences (IAC) and learn how domain-specific AI agents can transform and supercharge your operations with expert consistency!

Showcasing #SemiKong powering an Etch Process Expert AI Agent was energizing—Especially while celebrating open-source AI innovation together with the community! Special thanks to The AI Alliance, Cerebral Valley and Ollama for organizing this amazing event! Link to SemiKong demo in the comment ⬇

Looking forward to working with TOKYO ELECTRON LIMITED, Samsung Semiconductor, IBM Research, JSR Corporation, FPT Software AI Center, NVIDIA AI, and others already engaged, to build out open, shared industry models—upon which there will be healthy and thriving competition of innovative, proprietary models and commercially viable AI agents. We keep getting this question when presenting open-source #SemiKong: “How do you make money?” AITOMATIC fully intends to monetize this, by providing valuable products & services at the highly valuable proprietary model and enterprise agent layers, and so do our partners. And we do so at the high end of the value chain, by sharing the R&D expense at the foundation layer. We all benefit from more rapid innovation and greater value creation for our customers, avoiding duplicating efforts and re-inventing the wheel at the lower layers. Yann LeCun, we are concretely realizing the vision you & I discussed, thanks to open foundational components like #Llama and #Gemma. BTW congrats on Llama3.1. It is another game changer, for SemiKong and others. Cc Anthony J. Annunziata Yoshi Hirayama Jae-Yong Park Daisuke Oku Phong Nguyen #AIAlliance

Thanks Yann LeCun and AI at Meta team for making Llama3 available to advance open science & open development—in this case, for the semiconductor industry. Cc Anthony J. Annunziata Phong Nguyen Quynh Lan Nguyen Daisuke Oku Noritaka Yokomori Sean Hughes AITOMATIC, FPT Software AI Center, TOKYO ELECTRON LIMITED —— Yann LeCun, VP and Chief Al Scientist at Meta, the company behind Llama, said in a statement, “SemiKong, built on Llama3, shows open-source Al’s potential to drive specialized innovation. It’s a prime example of why we advocate for Al openness.” https://lnkd.in/g43sPzme

Join us today at PST 3:30 PM at #SEMICONWest for an exclusive look at how domain specialist agents—powered by SemiKong the first open-source LLM for semiconductors—are transforming the industry. Hear from Dr. Christopher Cuong T. Nguyen of AITOMATIC, Mr. Daisuke Oku and Mr. Atsushi Suzuki of Tokyo Electron on how this breakthrough can help you accelerate innovation and reduce operation costs! #Semiconductors #AI #SemiKong #aiKO

The difference between RAG vs. RAG + Fine-tuned Models vs. RAG + Planning & Reasoning? 37% vs. 59% vs. 85% accuracy. In our recent study on the design space of Q&A/RAG (an #AIAlliance project together with IBM Research and others), we explored advanced techniques to enhance domain-specific Q&A systems using the FinanceBench dataset. Key findings: 1. Fine-tuning Embedding models boosts performance more than fine-tuning Generative models. More bang for your buck. 2. Employing Reasoning yields significant gains, especially on compound/abstract queries (eg. in Finance, Semiconductor, Manufacturing) 3. RAG + Planning & Reasoning outperforms even fully fine-tuned RAG. What does this mean to you? 1. Prioritize fine-tuning your retriever models for efficiency and performance gains. 2. Implement iterative reasoning mechanisms (such as OODA loop) to drastically improve accuracy. 3. Planning & reasoning is the future for solving enterprise Q&A challenges. How are you innovating and advancing domain-specific AI systems? Let us know in the comments. 👇 https://lnkd.in/guxPxaYi