MythWorx, an AI startup, announced the appointment of Jason Williamson as its new chief executive officer. Concurrently, the company revealed that its Echo Ego v2 model has delivered standout performance in the MMLU-Pro benchmarking test, outperforming hundreds of larger LLMs. This success is attributed to its unique hybrid architecture and biomimetic learning, which offer a sustainable and cost-effective alternative to brute-force AI scaling.
Jason Williamson, a former Oracle and AWS executive and United States Marine Corps veteran, has been appointed CEO of MythWorx.
MythWorx's Echo Ego v2 model outperformed models hundreds of times larger in MMLU-Pro benchmarking.
The model achieved a 71.24% accuracy with zero pretraining and no chain-of-thought prompting.
It uses approximately one tenth the power of a typical LLM and reduces AI infrastructure costs by 90%.
The model has only 14 billion parameters, compared to over 600 billion in some large LLMs.
Echo Ego v2 demonstrated best-in-class accuracy of 87.64% on the math portion of the test.
MythWorx's success in MMLU-Pro benchmarking highlights a shift in the AI industry, proving that superior performance on complex reasoning tasks can be achieved through architectural ingenuity rather than massive scale. The company's Echo Ego v2 model is built on a hybrid architecture that employs biomimetic learning, which mimics human cognitive processes to self-improve and adapt without the need for extensive pretraining or massive datasets. This approach contrasts sharply with traditional LLMs that rely on brute-force scaling, which has led to significant power usage and compute limitations.
This breakthrough addresses a critical challenge for the AI industry, which is facing potential expansion stalls due to power constraints. Jason Williamson, the newly appointed CEO, emphasized this point by stating, "LLMs do not have true reasoning capabilities, and fall short of the breakthrough performance of MythWorx. Other firms run on power-hungry GPUs in a highly inefficient way that simply cannot scale at the current pace. They require massive datasets and pretraining to perform acceptably. We have a hybrid architecture that's designed to self-improve and adapt just like a human brain, without the need for any pretraining. It’s a ‘black swan’ moment for AI, that will democratize access to high-performance reasoning through elegant design, not raw power.”
This new architecture not only provides a more sustainable approach but also enhances speed, answering queries in approximately 1.2 seconds, significantly faster than traditional LLMs. For enterprises, this means a more efficient and powerful tool for complex reasoning tasks, from solving legal cases to medical diagnostics, all with minimal compute requirements.
My focus is on growing the team, safeguarding and promoting our groundbreaking AGI innovations and benchmarking results, and delivering our capabilities to market.
MythWorx is an AGI innovator introducing a new class of AI that delivers true reasoning capabilities without pretraining and with minimal compute requirements. It outperforms large-scale models across key benchmarks. MythWorx is now entering joint ventures and master licensing agreements.
TechIntelPro empowers B2B technology decision-makers with the latest industry trends, deep insights, leadership perspectives and actionable intelligence to drive both organizational and professional milestones—and thrive on the tech pulse.