Cohere’s Aidan Gomez thinks enterprise is key to AI profit

🌈 Abstract

The article discusses Aidan Gomez, the CEO and co-founder of Cohere, an AI startup focused on the enterprise market. It covers Gomez's background as one of the authors of the "Attention is all you need" paper that described transformers and kicked off the LLM revolution in AI. The article explores Cohere's enterprise-focused approach, the challenges of bringing LLMs to market, the importance of competition and optionality for enterprise customers, and Gomez's views on the current state and future potential of AI technology.

🙋 Q&A

[01] Cohere's Enterprise-Focused Approach

1. What is Cohere's focus and how does it differ from other AI companies?

• Cohere is focused on the enterprise market, building AI products and models for large companies, rather than consumer-facing products.
• Unlike companies like OpenAI, Cohere is not making consumer products at all. Its focus is on providing AI capabilities that enterprises can integrate into their own products and workflows.

2. How does Cohere's enterprise-focused approach give it an advantage over competitors?

• Enterprises are risk-averse and price-sensitive, so they want to work with companies that operate in a competitive landscape rather than being locked into a single provider.
• Cohere's independence and ability to offer optionality to customers is an advantage over larger tech companies like Microsoft or Google.

3. How does Cohere structure its business and teams to support its enterprise focus?

• Cohere has a large engineering team focused on building reliable, controllable models that enterprises can safely deploy.
• The company also has a growing go-to-market team dedicated to educating enterprises on the capabilities and limitations of the technology.
• Cohere's structure is aimed at helping the market adopt the technology in a responsible way.

[02] The Challenges of Bringing LLMs to Market

1. What are the key challenges in deploying large language models (LLMs) at scale?

• The compute and infrastructure costs to run LLMs are extremely high, making it difficult for enterprises to deploy them widely.
• Cohere has focused on building models that are the "right size" for the market, rather than the largest possible models, to make them more economically viable for enterprises.

2. How does Cohere approach the issue of model reliability and controllability?

• Cohere has internal teams focused on improving model safety, reducing bias, and developing tools for enterprises to maintain control and oversight over the models.
• The company sees this as a critical part of its strategy, as enterprises are very risk-averse and need assurances around the reliability of the technology.

3. What are Gomez's views on the challenges of bridging the gap between the deterministic nature of traditional computing and the more probabilistic, nondeterministic nature of LLMs?

• Gomez acknowledges this is a significant challenge, but argues that the world is actually quite robust to nondeterministic systems, as we already rely on humans who can be unpredictable.
• He believes enterprises can adopt safeguards and controls to manage the risks, rather than expecting LLMs to be as deterministic as traditional software.

[03] The Potential and Limitations of AI

1. How does Gomez view the current capabilities and limitations of AI/LLMs?

• Gomez believes the technology has made remarkable progress, but is still not "there" in terms of being able to fully replace human intelligence and decision-making, especially in high-stakes domains like medicine.
• He sees a future where AI may surpass human knowledge in certain fields, but believes human oversight will still be critical in the near-term.

2. What are Gomez's views on the risks and societal impacts of AI, particularly around misinformation and manipulation?

• Gomez sees the potential for AI to be used for scalable manipulation and the spread of misinformation, but believes these risks can be mitigated through measures like better human verification on social media platforms.
• He argues that democracy is already vulnerable to manipulation, and that AI is not a fundamentally new threat, but rather amplifies existing risks that need to be addressed.

3. How does Gomez respond to concerns about the ability of LLMs to "bear the weight" of our expectations for AI?

• Gomez believes the technology will continue to improve and surpass expectations, even if we are perpetually dissatisfied and want more.
• He argues that LLMs can already be more trustworthy than humans for certain tasks, like medical diagnosis, and that the bar for trust will continue to rise over time.