Feb
“Distillation” Is Shaking Up The AI Industry
jerry9789 0 comments artificial intelligence, Brandview World
Copyright: Airam Dato-on
Paradigm Shift
We’ve recently written about recent AI advancements and popularity, particularly generative AI like that of ChatGPT, driving renewed demand for data centers not seen in decades. This surging demand pushed tech investors to put $39.6 billion into data center development in 2024, which is 12 times the amount invested back in 2016.
A recent development, however, has stirred things up, especially the concept that billions of dollars needed to be spent for AI advancement. Developed by a Chinese AI research lab, an open-source large language model named DeepSeek was released and performed on par with OpenAI, but it reportedly operates for just a fraction of the cost of Western AI models. OpenAI, however, is investigating if DeepSeek utilized distillation of the former’s AI models to develop their systems.
Copyright: cottonbro studio
What Is “Distillation?”
According to Labelbox, model distillation (or knowledge distillation) is a machine learning technique involving the transfer of knowledge from a large model to a smaller one. Distillation bridges the gap between computational demand and the cost for training large models while maintaining performance. Basically, the large model learns from an enormous amount of raw data for a number of months and a huge sum of money typically in a training lab, then passes on that knowledge to its smaller counterpart primed for real-world application and production for less time and money.
Distillation has been around for some time and has been used by AI developers, but not to the same degree of success as DeepSeek. The Chinese AI developer had said that aside from their own models, they also distilled from open-source AIs released by Meta Platforms and Alibaba.
However, the terms of service for OpenAI prohibits the use of its models for developing competing applications. While OpenAI had banned suspected accounts for distillation during its investigation, US President Donald Trump’s AI czar David Sacks had called out DeepSeek for distilling from OpenAI models. Sacks added that US AI companies should take measures to protect their models or make it difficult for their models to be distilled.
Copyright: Darlene Anderson
How Does Distillation Affect AI Investments?
On the back of DeepSeek’s success, distillation might give tech giants cause to reexamine their business models and investors to question the amount of dollars they put into AI advancements. Is it worth it to be a pioneer or industry leader when the same efforts can be replicated by smaller rivals at less cost? Can an advantage still exist for tech companies that ask for huge investments to blaze a trail when others are too quick to follow and build upon the leader’s achievements?
A recent Wall Street Journal article notes that tech executives expect distillation to produce more high-quality models. The same article mentions Anthropic CEO Dario Amodei blogging that DeepSeek’s R1 model “is not a unique breakthrough or something that fundamentally changes the economics” of advanced AI systems. This is an expected development as the costs for AI operations continue to fall and models move towards being more open-source.
Perhaps that’s where the advantage for tech leaders and investors lies: the opportunity to break new ground and the understanding that you’re seeking answers from unexplored spaces while the rest limit themselves and reiterate within the same technological confines. Established tech giants continue to enjoy the prestige of their AI models being more widely used in Silicon Valley — despite DeepSeek’s economical advantage — and the expectation of being the first to bring new advancements and developments to the digital world.
And maybe, just maybe, in that space between the pursuit of new AI breakthroughs and lower-cost AI models lie solutions to help meet the increasing demand for data centers and computing power.
Copyright: panumas nikhomkhai
Featured Image Copyright: Matheus Bertelli
Top Image Copyright: Airam Dato-on
Aug
Can Synthetic Respondents Take Over Surveys?
jerry9789 0 comments artificial intelligence, Burning Questions
What Are Synthetic Respondents?
AI has increased operational efficiency by streamlining knowledge bases and shortcutting processes so it’s no surprise people and companies are looking for more ways for its application. For market research, one curious consideration is whether it could take over surveys, essentially by replacing actual respondents with synthetic respondents.
Also known as virtual respondents, digital personas, and Virtual Audiences, synthetic respondents are individual profiles constructed by Large Language Models (LLMs) from real or simulated data. Ideally, the data or descriptions used to generate these profiles come from previously conducted surveys and are combined with individual-level demographics, attitudes and behaviors.
Using these synthetic respondents over real respondents could benefit your research with speed, accuracy and cost savings, at least according to their advocates. Basically, you just need to conduct one survey and from the profile description or data you gathered from the actual respondents, you’re able to generate results from the constructed individuals over and over for succeeding studies and research.
Testing Synthetic Respondents
While synthetic respondents could accurately represent real respondents, relying exclusively on the results from these AI-based individuals may not be entirely beneficial. A webinar hosted by Radius Global took a closer look at the potential of AI-generated synthetic respondents through three case studies of quantitative concept testing, quantitative communications research, and qualitative communications research.
Aggregate results for the concept tests involving game controllers indicate somewhat strong similarities between the results of the real and synthetic respondents. This extends to the results from the quantitative communications research when it comes to the believability of statements on the benefits of milk, although there were some differences. The differences were much more pronounced though when it comes to surprise over the same statements, and there was incongruence when considering how each statement could possibly increase milk consumption.
The qualitative communications research was seeking in-depth insights into women’s needs, perceptions, and preferences for running a race or marathon, with the feedback gathered meant to be used for creative content. Personas were constructed from the profiles of six women aged between 18 and 64 years old who ran at least once in an average week. They had an LLM assume each persona to allow a comparison between findings from real participants to synthetic respondents.
They found that while both real and synthetic respondents have somewhat similar responses when it comes to functional aspects as goals for women in general pursuing fitness, the AI responses lacked emotional expressions. There are also little differences in the synthetic respondents’s responses despite having different profiles, and there was even a lack of subtle differences.
As for concerns among women who are aspirational marathon runners, the synthetic personas were consistent in their responses while the real respondents provided more nuances, variety, and perspectives more prevalent among women.
Synthetic Respondents Vs. Real Respondents
Synthetic respondents appear to be useful if you’re evaluating existing ideas and concepts; however, if you’re looking for “breakthroughs” or essentially new insights you would’ve never arrived at had you not performed the case study or research, you would need to engage with real respondents, relying exclusively on their results or combining them with that of synthetic respondents. Yes, there could be cases where synthetic respondents could be used, but the results must be extensively validated. It would also require increasing the efficiency of how data used to construct these individuals are analyzed in addition to enhancing the quality of the data and information gathered for these profiles through thorough screening, intelligent probing, and smart choice models.
There is a place for synthetic respondents in market research, but as another tool in a researcher’s disposal. They won’t be taking over surveys nor replacing actual respondents wholesale anytime soon it seems, as that elusive “Eureka” moment researchers seek are inherently tied with the nuances and perspectives of human emotion and experience you just simply couldn’t construct.
Photo courtesy of Pavel Danilyuk
info@cascadestrategies.com-
2032 Newport Way NW,
Issaquah, WA 98027 - (425) 677-7430
