LLM use cases? And now comes conversation.

3 innovation and digital news in 1 minute. Every Monday. Episode 373

Sierra to tackle the “difficult” in conversation AI.

The drawbacks of conversation AI vs. human are clear, and this is what Sierra tackles. Empathy and difficult situations. Sierra is training for these situations. The offer is an “AgentOS”: knowledge, integration, profile, and personality (!). And large recent fundings of $110m — Sequoia is in the lead. Opinion: Here lies the force of Large Language Models versus former models: exact understanding, context memorisation (so important in customer service), and the capabilities to adapt to a situation (memory, more training, and personality like empathy).

Exactly the first steps of an Agent AI that can replace a human.

Conversations in customer service, this is Kore

Kore.ai is the next step of a chat bot. Conversation bots based on LLM — and this is where the transformer models are excellent. Kore offers bots that are trained in verticals: banking, retail, food, service, health. A remarkable financing of $223m ($150m from FTV capital). NVIDIA is an early investor. Opinion: This reminds us of the 2017–2019 wave of bot development based on Large Language Model. This model showed its limits: low memory, finite and hard to train. Well, this is the strength of the LLM. Pre-training in specific verticals is just a huge time and resource benefit.

Kore and similar models will replace all “traditional” bots. Especially in call centres.

Summary of health conversation doctors — patients

This is Abridge, a US start-up that gathered over $200m for summarizing clinical documentation and conversations around health. There is a pro version (integrated into the leading patient record platform in the US called Epic System) and an app “Abridge for Patients” to record, transcribe, save, and summarise docs’ conversations. Opinion: A major LLM use case. A very “American” service. US where law plays a strong role in medication, proving and saving clinical and medication records is key. Abridge is a place for documentation, but also for patients’ approval in the app.

Beyond the US-specificity a very interesting case!