Greetings, technology enthusiasts, and welcome to our dedicated AI bulletin from TechCrunch.
In the realm of AI this week, a study by Gartner has highlighted that approximately 30% of generative AI initiatives within corporations are likely to be discontinued post the proof-of-concept stage by the end of 2025. The challenges leading to this outcome are numerous, encompassing low-quality data, insufficient risk management, and rising costs of infrastructure, among others.
The report also notes that one significant hurdle to the broader uptake of generative AI is its ambiguous value proposition to businesses.
Implementing generative AI on an enterprise scale incurs substantial expenses, predicts Gartner, with figures ranging from $5 million to a staggering $20 million. It also found that a basic coding assistant might demand an initial investment of $100,000 to $200,000 plus recurring charges over $550 per user annually. Meanwhile, an AI tool for document search could have an upfront cost of $1 million and annual per-user costs between $1.3 million to $11 million, according to the same report.
Such hefty costs prove difficult for many companies to justify, especially when the tangible benefits remain elusive and potentially long off in the future—if they emerge at all.
A recent survey conducted by Upwork has thrown light on the actual impact of AI on workplace productivity, suggesting that AI might be more of a hindrance than a help to many employees. This survey, which heard from 2,500 individuals across C-suite executives, full-time staff, and freelancers, revealed that close to half (47%) of AI users are at a loss on how to realize the productivity improvements anticipated by their employers. Even more telling, over three-quarters (77%) felt that AI applications have in some way hampered their productivity and added to their workload.
This indicates that the initial allure of AI could be wearing thin, in spite of continued interest and investment from the venture capital sector. This notion is further supported by numerous stories highlighting the practical challenges and unresolved issues presented by generative AI technologies.
Just recently, Bloomberg shared a story on a Google-assisted AI tool being trialed at Florida’s HCA hospitals to analyze patient records, which has struggled to reliably deliver pertinent health information, failing in scenarios such as identifying patient drug allergies.
As expectations from AI continue to rise, it’s becoming crucial for providers to realistically manage those expectations unless significant advancements are made to overcome the technology’s current limitations.
Time will tell if they rise to the occasion with the necessary humility.
Updates
SearchGPT: Last Thursday, OpenAI unveiled SearchGPT, a new search functionality aimed at providing prompt, web-sourced answers to users’ questions.
Bing enhances AI capabilities: Microsoft is not far behind, having recently showcased Bing generative search, its own AI-enhanced search feature, currently available to a select few users, that, similar to SearchGPT, synthesizes information from the web to generate summary answers.
X involves users: X (formerly Twitter) has recently been noted for auto-enrolling user data for training its chatbot Grok, a move quickly criticized by EU regulators among others. (For those looking to opt-out, here’s how.)
EU seeks AI guidance: The European Union has initiated a consultation regarding the regulations for general-purpose AI model providers under its AI Act, a risk-based regulatory framework geared towards AI applications.
Perplexity to share ad revenue: AI search engine Perplexity is planning to share ad revenue with news publishers whose content is highlighted by its chatbot, an apparent effort to calm concerns over accusations of plagiarism and unethical data scraping.
Meta introduces AI Studio: Meta announced the widespread release of its AI Studio tool to all creators in the U.S., facilitating the creation of personalized AI-driven chatbots. This follows an initial launch and testing phase with select creators earlier in the year.
Commerce Department recommends open models: The U.S. Department of Commerce recently endorsed the use of “open-weight” generative AI models such as Meta’s Llama 3.1, while also advising the government to develop enhanced monitoring capabilities for potential risks.
$99 Friend: Avi Schiffmann, having left Harvard, is developing a $99 AI companion named Friend. This neck-worn device aims to act as a personal companion, although its effectiveness remains to be seen.
Research Highlight
The dominant method for directing AI models to follow guidelines and ensure safety, Reinforcement Learning from Human Feedback (RLHF), is being revisited by OpenAI due to its intensive demand for human evaluators and feedback. Exploring an alternative, OpenAI’s recent publication introduces Rule-Based Rewards (RBRs), a technique employing explicit rules to guide a model’s responses, promising improved safety while reducing reliance on extensive human feedback data.
Model Spotlight
DeepMind’s recent venture into solving intricate mathematical challenges with AI has yielded notable achievements with the unveiling of two AI systems capable of resolving four out of six problems from the latest International Mathematical Olympiad (IMO). These systems, AlphaProof and AlphaGeometry 2, showcased their ability to abstract and plan complexly—a frontier usually challenging for AI to breach.
In Brief
Stability AI has developed a generative AI model, Stable Video 4D, capable of transforming a single video into multiple clips from diverse angles, showing potential applications in gaming, video editing, and virtual reality. This innovative model, as per the company’s announcement, opens new vistas for realistic, multi-perspective video creation.
Compiled by Techarena.au.
Fanpage: TechArena.au
Watch more about AI – Artificial Intelligence
