Zeya Yang and Kristina Shen, of Andreessen Horowitz, recently published a fantastic article in relation to B2B applications built with LLMs.
It really articulates how what Michael Irvine and I are trying to build at Delphi Labs is different to many of the other entrants to the space.
As we've gone through our fundraise, I've described the product features and workflow that we have in our roadmap. It's been key to how we differentiate ourselves in what is the hottest space for startups - this space will remain the most disruptive for years to come.
However, at times, I've struggled to articulate why what we're building is so different; that we shouldn't really be compared to some of these other entrants. The article describes two “waves” of applications, the first being “GenAI” which generates responses from prompts and the second being “SynthAI” which processes a vast quantity of information into insights.
Many of the other entrants, which are using LLMs in the data space, are Wave 1: Generative applications - they make it easier to generate new content and analyses quickly.
This does dazzle many people, but as Michael and I know from many data practitioner hours worked - trust is everything. You would rather have fewer, consistent, clear and reliable answers than many which cause confusion and erode trust.
Delphi is a Wave 2: SynthAI platform. We have instinctively designed it from the ground up to be this way. Delphi has, and will continue to develop, many steps to try and use a vast body of existing information to deliver insights. Whether this is an answer to an existing question, an existing dashboard, notebook or graph. Even helping users to ask better questions.
Delphi will work with human data teams, incorporating their validation in multiple places in our platform. This starts with our use of semantic layers, which is one of the most scalable ways to leverage validated human input.
"Today, with Wave 1 applications, the answer is frequently that we’re better off doing it ourselves."
I know that this will be the unfortunate truth for many entrants - if stakeholders can't trust the outputs of your platform, and data folks find it irritating to validate the machine-generated code it writes, they will just do it themselves. It's what I would do if asked to validate some >100 line SQL query from a machine... chuck it and write my own.
I've enjoyed meeting Barry McCardel a few times over the last year, and was an early Hex customer in the UK:
"AI is here to augment and improve humans, not replace them. When it comes to understanding the world and making decisions, you want humans in the loop. What AI can do is help us apply more of our brainwaves to valuable, creative work, so that we not only spend more hours in a day on the work that matters, but also free ourselves to do our best work."
Amen 💯
"We believe LLMs will need to focus on synthesis and analysis — SynthAI — that improves the quality and/or speed of decision-making"
When making decisions with data, quality cannot decline due to the introduction of LLMs. It must either be the same and increase speed, or increase quality. With Delphi, it's our aim for answers to be as good as you would get from an analyst, as we indeed learn from them - it's just orders of magnitude faster.
"the prize is not about who can build the AI synthesis capability; rather, it’s who can own the workflow."
This is why we've started in Slack: it's where most data workflows begin. It's much easier to integrate into an existing workflow and speed it up, than try to fight against culture and muscle memory. We aim for the entire analytics workflow to be solved by Delphi - this is our mission.
https://a16z.com/2023/03/30/b2b-generative-ai-synthai/
We’re also now expanding our beta user group, we have reached out to the first batch of 100 sign ups on our waitlist, but we’d love to hear from others too!
Please fill in this Google Form to help us understand if we are compatible with your stack and can add value to your organisation:
The second two sections are helpful for us to be able to understand our early demographics, but aren't required - you can click 'next section' on them without filling, if you don't want to answer any or all of the questions in them, and then finally click 'submit'.