- AI startup Cosine, operating as Buildt AI Inc., raised $2.5 million in seed funding.
- Funding is led by U.S.-based venture capital firms Uphonest and SOMA Capital, with additional support from Lakestar, Focal, and others.
- Cosine’s latest AI model achieved the highest score on the SWE-Bench benchmark, with a 30% score.
- This score represents a 56% improvement over the previous record and a 2,196% increase compared to OpenAI’s GPT-4.
- The SWE-Bench evaluates AI models’ ability to perform complex software engineering tasks.
- Cosine’s AI model focuses on augmenting, not replacing, human developers by mimicking human reasoning processes.
- Alistair Pullen, Yang Li, and Sam Stenner co-founded the company after they identified the potential of large language models in software development.
Main AI News:Â
In a significant stride within the artificial intelligence sector, Cosine, an AI startup dedicated to emulating human reasoning, has successfully secured $2.5 million in seed funding to advance the development of its cutting-edge AI software developer. This investment round was spearheaded by U.S.-based venture capital firms Uphonest and SOMA Capital, with additional backing from Lakestar, Focal, and others.
Operating under the official name Buildt AI Inc., the startup also announced a remarkable achievement with its latest AI model. Dubbed the most advanced in its lineup, the model has attained what Cosine claims to be the highest score ever recorded on the SWE-Bench, an industry-standard benchmark designed to assess AI software engineering capabilities. Cosine’s AI-powered software developer scored 30% on the SWE-Bench test—a figure that, while modest in isolation, represents a substantial 56% improvement over the previous record of 19%, set by The San Francisco AI Factory Inc. Notably, it also marks a staggering 2,196% increase compared to OpenAI’s GPT-4, which scored just 1.31%.
The SWE-Bench benchmark rigorously evaluates AI models’ ability to perform complex software architecture tasks, including debugging issues and implementing new features within existing codebases. The assessment gauges the models’ proficiency in understanding, generating, and modifying sophisticated code.
Cosine was co-founded by Chief Executive Alistair Pullen, Chief Operating Officer Yang Li, and Chief Information Officer Sam Stenner, who identified the potential of large language models to mimic human software developers as early as 2022. Stenner elaborated that they achieved this by codifying human reasoning processes and using that framework to train Genie’s underlying large language model.
Pullen highlighted the company’s advancements in human reasoning, which have enabled the creation of AI models capable of performing well beyond the narrow task ranges and rigid prompts typical of current AI-driven software development tools.
Stenner clarified that Cosine’s goal is not to replace human developers but to augment them with AI assistants with human-like reasoning abilities, allowing for seamless collaboration on various coding tasks.
Conclusion:
Cosine’s advancements signal a significant shift in the AI software development landscape. By achieving unprecedented benchmarks, the company is positioning itself as a leader in creating AI tools to augment human developers with advanced reasoning capabilities. This development will likely drive further investment in AI-driven software engineering solutions, potentially accelerating the adoption of AI assistants in coding environments. The market should expect increased competition as other companies strive to reach or surpass these new standards, ultimately leading to more innovative and efficient software development tools.
