Getting realistic about AI’s potential with Nick Frosst from Cohere | Equity Podcast

21 Aug 2024 (26 days ago)

Nick Frosst's Journey and Cohere's Focus

Nick Frosst became interested in neural networks around 2012-2013 after seeing research from Geoff Hinton's lab at the University of Toronto that showed neural networks could be trained to recognize images.
Frosst initially believed he had missed out on the opportunity to be involved in the early stages of neural network research.
Cohere was founded based on the realization that general-purpose AI models would outperform specialized models for tasks such as document summarization and data extraction.
Cohere is a company focused on developing language models for practical enterprise applications rather than pursuing artificial general intelligence (AGI).
Cohere focuses on addressing the specific needs of enterprises, such as data security, privacy, multilingual capabilities, accuracy, and reducing hallucinations in their models.
Cohere is focused on making large language models (LLMs) useful for enterprise, with a focus on improving multilingual capabilities, data security, and reducing hallucinations.
Nick Frosst, founder of Cohere, believes that AI is not on its way to becoming a digital God and that the technology will likely become a regular part of life.
Cohere is focused on making their language models useful for a variety of languages, including investing in Japanese language data to cater to the large Japanese developer community.

A general-purpose language model trained on a vast amount of language data, with feedback, outperforms models trained for specific tasks.
Language is inherently diverse, requiring a broader understanding than simpler tasks like image recognition.
While linguistics plays a role, language models are fundamentally statistical models of sequences, capable of learning different languages without explicit grammatical understanding.
There has been a shift in focus from simply building AI demos to creating AI solutions that provide tangible value to businesses, such as automating tasks and improving efficiency.
Many people are shifting their focus away from fears of artificial general intelligence or "digital gods" and towards the practical applications of large language models (LLMs).

The distribution of languages on the web does not reflect the distribution of languages spoken in the world, making it more difficult to develop language models for languages other than English. (00:15:48)
Concerns exist regarding the environmental impact of LLMs, particularly their energy consumption during training and use.
Data privacy is another concern, with questions arising about the source of training data and potential misuse.
Cohere's CEO would prefer to use nuclear power as the energy source for data centers.
Cohere's CEO believes that large language models will not be a major energy consumer compared to other industries.

Cohere, a company focused on providing AI solutions for enterprises, recently raised a significant funding round, enabling them to further develop their products and meet the growing demand for practical AI applications.
Cohere has raised funding and is valued at over 5 billion dollars.
Nick Frosst, from Cohere, believes that AI is not in a bubble because companies are building real value.