Big tech firms like Google, OpenAI, and Anthropic, known for pioneering AI technology, are facing serious challenges as they try to develop the next generation of AI. According to sources, these companies are struggling to make breakthroughs with new AI models, citing a critical lack of fresh, high-quality data to train them effectively.
For example, OpenAI recently aimed to push boundaries with a new model, code-named “Orion,” but early tests suggest it doesn’t surpass existing models like GPT-4 or GPT-3.5. Reports indicate Orion failed at certain tasks, particularly coding challenges that were outside its training. Similar issues are affecting Google and Anthropic’s models. Google’s latest Gemini AI hasn’t met internal performance goals, while Anthropic delayed releasing its anticipated Claude 3.5 Opus model, hinting at similar setbacks.
Also read: OpenAI plans to build its first AI chip by 2026, check details
One major hurdle these companies face is the lack of new, valuable data, reports Bloomberg (via Windows Central). Generative AI models like ChatGPT rely heavily on internet content for training. However, they’re reaching a limit, especially as publishers and authors raise concerns about AI using their content without permission or compensation. Microsoft and OpenAI are even battling copyright infringement cases, with OpenAI’s CEO, Sam Altman, acknowledging that creating models without copyrighted material is nearly impossible. Recently, OpenAI won a case against publishers Raw Story and AlterNet, but copyright disputes remain a significant obstacle.
Also read: Is OpenAI violating copyright laws? Former company employee says YES
Beyond data issues, the cost of developing and maintaining these advanced AI systems is skyrocketing. OpenAI, for instance, faced potential financial trouble and projected $5 billion in losses before securing $6.6 billion in funding from Microsoft, Nvidia, and other investors. This funding helped stabilise OpenAI, now valued at $157 billion, but analysts predict more losses ahead, with profitability not expected until 2029. To avoid further financial strain, OpenAI may either need to become a for-profit company soon or return investor funds.
In addition to financial and legal challenges, OpenAI’s latest model Orion has reportedly completed its training phase, hinting at a release early next year. OpenAI is likely moving away from naming its models sequentially (e.g., GPT-5) and could give this release a different name. CEO Sam Altman believes advanced AI is achievable with current technology, but reaching “AGI” (Artificial General Intelligence) remains a costly, complex goal, potentially requiring billions of dollars in infrastructure.