More Gemini Features Coming to Google TV
Google TV integrates Gemini AI for image/video generation, enhances Google Photos, and adds a dedicated YouTube Shorts feed to the homepage.
Read moreA professor from Northeastern University believes in open scientific collaboration in the field of artificial intelligence technology. He and others have created an advanced open-source generation model called StarCoder, which provides programmers with licenses and flexible adjustments for various purposes such as gaming and industrial automation.
Over the past few years, generative AI and large-scale language models have become popular worldwide and have had a significant impact on programming. However, computer scientists, programmers, and smaller market participants have limited knowledge of the development process of these models, which prevents them from gaining a deep understanding of the technology and limits their meaningful participation in its further expansion.
Therefore, the professor and his research team actively participated in a project called BigCode, initiated by the private companies Hugging Face and ServiceNow.
Hugging Face is a company that hosts a large open-source machine learning community, while ServiceNow helps optimize technical solutions for businesses. They collaborate to support individuals with a professional background in AI research to responsibly develop and use open-source large-scale language models. To support this project, both companies have invested a significant amount of manpower and hardware resources. As a result, others can now obtain licenses for the state-of-the-art open-source generation model called StarCoder and adapt it for different purposes.
The professor stated, "You can spend a lot of money building these models, but you actually don't know their quality." He believes that the few multi-billion-dollar companies with vast resources to build these learning models and constantly amaze the world are completely closed off from sharing this technology with the community.
He said, "If you ask the people who create these models, 'What can I do with it?' they will unrealistically answer 'anything,' which is misleading." The professor believes that academic research can play a role in shaping generative AI technology.
He said, "Scholars can come in and rigorously evaluate these models, pointing out their strengths and weaknesses. They can say, 'Yes, you can use it for this, but please don't use it for other things without strict limitations.'"
He believes that a more urgent issue is people using this technology to make decisions that affect others, such as loan applications or job vacancies. He said that we should discuss in which situations it is not suitable to use these models to avoid causing more harm.
The professor has put a lot of effort into the BigCode project, which was launched in September 2022. He leads a working group focused on evaluating the two open models created by this project, StarCoder and SantaCoder.
First, creating an LLM requires determining the data on which the input model will be trained. After the model training is completed, its actual capabilities should be evaluated.
The models created by the BigCode project are trained on the Hugging Face cluster. The professor's team evaluated most of the models on the Northeastern Discovery cluster at the Massachusetts Green High-Performance Computing Center, which is a high-performance parallel computing system with leading computing technology and powerful storage solutions.
They conducted extensive evaluations on 19 different programming languages to understand the capabilities of these models.
He said, "When this project started, one of the goals was to make it usable in different programming languages to benefit multiple communities."
These models were tested for tasks such as generating code from natural language descriptions, documenting code, and predicting type annotations.
Other researchers conducted additional analyses, such as bias and toxicity analysis. The results showed that due to the models not being trained on a large amount of internet data, the content they generate has a lower toxicity.
The professor said that the StarCoder model underwent the most extensive evaluation in the history of the BigCode project, as it is a large collaborative effort.
He said, "It's a great project that brings together many researchers at different stages of their careers."
Part of the project published a paper in May of this year with nearly 70 co-authors. The professor mentioned that some doctoral and undergraduate students also contributed to the model.
Now, anyone can request to download and use StarCoder or SantaBase for research, commercial, or non-commercial purposes free of charge by signing the BigCode Open Responsible AI License Agreement and complying with applicable restrictions. They can also modify the models according to their needs.
For example, the professor has also collaborated with companies such as MathWorks and Roblox, which specialize in providing mathematical computing software and an online global gaming platform for engineers and scientists. They are exploring how to use StarCoder, introduce it internally, and customize it according to their own needs.
The model is also being used by many other researchers.
The project clearly and transparently explains the data used by its models. If people want the project to stop using their data, they can make a request. So far, only a few dozen people have done so.
The BigCode project is preparing for the next round of projects and is expected to announce further developments soon.
Google TV integrates Gemini AI for image/video generation, enhances Google Photos, and adds a dedicated YouTube Shorts feed to the homepage.
Read moreAlphabet Q1 paid subscribers reached 350M, fueled by YouTube & Google One. YouTube ad revenue slightly missed at $9.88B amid Premium growth. Total Q1 revenue hit $109.9B, beating forecasts.
Read moreAnthropic releases Claude connectors for Adobe, Blender, and Ableton, enabling AI workflow automation and app control. It also sponsors the Blender Development Foundation.
Read moreApple introduces an App Store model splitting 12-month plans into monthly installments. On iOS 26.4+, it lowers upfront costs without reducing total price, adding billing tracking.
Read moreGoogle launches Aadhaar digital IDs in Wallet for India, enabling secure QR/NFC verification. A phased partner rollout targets key sectors, while passport-based credentials expand globally.
Read moreYouTube tests "Ask YouTube," an AI tool blending text and video for complex queries. Initially for US Premium users, it supports interactive follow-ups, expanding Google's AI search.
Read moreGoogle grants US DoD classified AI access, joining OpenAI and xAI in defense deals amid Anthropic's dispute. Contractual safeguards against surveillance and lethal weapons remain unenforceable.
Read moreMistral AI launches Workflows, a Temporal-powered orchestration platform for secure, Python-driven enterprise AI automation featuring human-in-the-loop approvals.
Read moreCustom subliminal audio maker

AI Social Media Manager

AI Search Visibility

Workout and Calorie Tracker
Stock research platform

AI interview assistant

Scientific figure generator

AI courses with real code