Suggestions
Surya Dantuluri
founder
Professional Background
Surya Dantuluri is a prominent figure in the field of technology and entrepreneurship, with over a decade of experience in driving innovation and delivering exceptional results in various capacities. He is well-known for his leadership in technology startups and his ability to navigate complex business landscapes, demonstrating a strong entrepreneurial spirit and strategic thinking. Throughout his career, Surya has held key positions in reputable organizations where he has consistently demonstrated an aptitude for problem-solving, project management, and enhancing operational efficiencies.
Surya started his professional journey with a firm commitment to understanding the intricacies of technology and its impact on business. His work experience spans various industries, including software development, information technology, and digital marketing, which has equipped him with a diverse skill set and profound insights into technological advancements.
Education and Achievements
Surya Dantuluri boasts an impressive educational background that serves as a foundational pillar for his professional achievements. He earned a Bachelor’s degree in Computer Science Engineering from a prestigious institution, where he demonstrated exceptional academic prowess. His education provided him with a robust technical foundation that he has successfully leveraged throughout his career.
In addition to his degree, Surya has undertaken various certifications to stay ahead in the rapidly evolving tech landscape. His commitment to lifelong learning has empowered him to adapt to new challenges and seize opportunities in the digital realm.
Notable Achievements
Surya's dedication and innovative approach have garnered him several accolades and recognition in the industry. He is known for his ability to lead teams to success, having spearheaded numerous projects that have transformed businesses and pushed the envelope in technology adoption. His remarkable leadership skills are complemented by his collaborative approach, enabling him to work effectively with diverse teams and drive collective success.
Throughout his career, Surya has made significant contributions to developing cutting-edge software solutions that have enhanced user experiences and improved operational efficiencies for organizations. His projects have not only yielded tangible results but have also paved the way for future innovations in technology. Surya’s work has gained recognition from his peers and stakeholders alike, establishing him as a thought leader and influencer in the tech space.
In addition to his professional accomplishments, Surya actively participates in mentorship programs and community initiatives, sharing his expertise and insights with aspiring entrepreneurs and technology enthusiasts. He is passionate about fostering an inclusive environment for innovation and technology education, advocating for diversity in tech to drive holistic growth and development across various sectors.
Highlights
claude playing TFT -- i ran it for two games and it improved in-context from 0 to 3/30 rounds won. it figured out "3-starring" on its own (buying pairs to upgrade units) which is a core mechanic of this game and hasnt been instructed on how to play besides asking it to "play tft and win"
comparing this cua agent to OpenAI Five's result from 2017 which is a different game, model(a big LSTM), had access to Dota's API outputted a discrete action at 4.6hz
this cua agent runs at 0.15hz considering it takes anywhere from 2-15 seconds to think(0.07hz without), and is just pixels in, any mouse/keyboard out... and (potentially) wasnt trained on TFT specifically but not sure
despite this 2-15s seems to be fine for a turned based game like TFT but in prior videos like league/minecraft is considerably harder for a cua agent to do since by the time you take a screenshot the game state may have changed and you've wasted 7 seconds thinking, and another 7 seconds thinking about the new game state, and catastrophically failing like that.
the intention here is not to play TFT or league with a bot. please dont use it as such
stack more layers add more compute and spend an oom at inference time - hard to do currently apart from cot thinking tokens
energy based transformers are an expensive way of spending compute at test time but seem increasingly more effective: https://t.co/LWLUCFF4g2