Monthly Brandtech Blend - January 2025

GPT-5’s Release Date and Performance in Question

According to a recent article from The Wall Street Journal, OpenAI's development of its much-anticipated model, GPT-5, is reportedly lagging behind schedule and not meeting expectations, with its performance not yet justifying the extensive costs associated with its development. This report follows an earlier one from The Information, which indicated that OpenAI was exploring new strategies as GPT-5 may not represent a significant advancement compared to its predecessors. The development process, codenamed Orion, has spanned 18 months and seen at least two large training runs. The initial run progressed slower than anticipated, raising concerns about the time and resources required for subsequent efforts.

To enhance GPT-5's capabilities, OpenAI has shifted its approach by utilizing publicly available data and licensing agreements and hiring personnel to generate new data through coding and problem-solving tasks. Additionally, the company is relying on synthetic data produced by another model, o1. Despite these efforts, the WSJ report indicates that while GPT-5 shows some improvements over earlier models, it has not yet reached a level of advancement that would warrant the ongoing costs of its operation. OpenAI has previously stated that it does not plan to release the Orion model this year, and the company has not provided further comments on the current status of GPT-5.

Find more details at TechCrunch.

‍

New LLM DeepSeek V3 Outperforms GPT-4 but Draws Criticism

Chinese start-up DeepSeek has shaken up the global AI landscape by launching its new large language model (LLM), DeepSeek V3, which reportedly outperforms models from major players like Meta and OpenAI in benchmark tests. The model boasts 671 billion parameters and was developed in approximately two months for $5.58 million, utilizing fewer computing resources than those typically required by larger tech firms. This advancement highlights the growing capabilities of smaller companies in the AI sector, particularly in generative AI, which relies on LLM technology to perform complex tasks such as coding and essay writing.

However, DeepSeek V3 has drawn criticism for misidentifying itself as ChatGPT and even claiming to be a version of OpenAI's GPT-4  in some responses. This misrepresentation includes sharing jokes identical to those from GPT-4. Experts suggest that this behavior may result from DeepSeek V3 being trained on datasets that include outputs from ChatGPT, raising concerns about the model's potential to perpetuate biases and inaccuracies. Training AI models on existing outputs can lead to a decline in quality, akin to "taking a photocopy of a photocopy," and may violate OpenAI's terms of service. As the internet becomes increasingly filled with AI-generated content, the challenge of distinguishing reliable data from this "contamination" further complicates the reliability of models like DeepSeek V3.

More info at TechCrunch and South China Morning Post.

‍

The Responsible Marketing Hub Publishes its Key Takeaways for 2025

The Responsible Marketing Hub with Google "supports marketers in mastering end-to-end measurement, effectiveness, and growth strategies to help them deliver ROI for their business." Below is a summary of their key takeaways for 2025, as presented in The Drum.

According to the Hub, in 2025, the focus will likely shift towards privacy-first measurement strategies, innovative media effectiveness solutions, and the continued adoption of AI, all of which are expected to further transform the digital advertising landscape. Privacy remains a paramount concern, with tools designed to ensure compliance while enhancing performance. Successful brands will focus on building trust through transparency and compliance, viewing privacy as a competitive advantage rather than a hindrance. And as the marketing landscape evolves, investing in durable measurement foundations and embracing AI-driven strategies might prove essential for marketers aiming to thrive in a rapidly changing environment.

Read more at The Drum.

‍

Britain’s ICO Disapproves of Google’s “Fingerprinting” plan

Britain's data protection regulator, the Information Commissioner's Office (ICO), has criticized Google for allowing advertisers to use digital fingerprinting to track users, labeling the move "irresponsible." The ICO expressed concerns that this practice undermines consumer control and choice regarding data collection, as fingerprinting is more difficult for users to detect and block than cookies. Fingerprinting gathers various signals from a device's software or hardware to create a unique identifier for users, which poses challenges for privacy-conscious individuals who may not be able to easily consent to or manage their data like they can with cookie consent pop-ups.

Google announced that from mid-February on, advertisers would be able to utilize fingerprinting to better target audiences on connected TV platforms and other internet-linked devices. The ICO's executive director for regulatory risk, Stephen Almond, emphasized that businesses must deploy advertising technologies like fingerprinting lawfully and transparently, warning that the ICO would take action if necessary. Google has indicated it will engage in further discussions with the ICO regarding this policy change.

Find out more at The Guardian.