Ep 6- Autonomous AI Agents in B2B SaaS, Building for RevOps, future AI interface, challenges + more

Summary

In this thought-provoking discussion, Renee engages with a guest who shares insights into the AI industry, particularly the significance of innovative user interfaces beyond mere text inputs. They delve into the potential of specialized AI tools tailored to specific use cases, exploring the integration of probabilistic systems and embracing their inherent quirks. The conversation touches on the challenges faced by sales teams in adopting AI, the need for open-mindedness, and the optimism surrounding AI's evolution towards more natural and intuitive interfaces. Additionally, they explore the guest's diverse interests, including urban planning, startups, and the captivating realms of psychology and consciousness.

Takeaways

• 🤖 AI agents like ChatGPT provide generalists with 'superpowers' to tackle complex projects and tasks beyond their expertise.
• 💬 Text input interfaces like chatboxes are limited and frustrating ways to interact with AI systems. More intuitive interfaces like buttons and visual builders are needed.
• 🧩 Specialized, task-oriented AI tools with simple user interfaces tailored to specific use cases will likely be more successful than general-purpose AI assistants.
• 🤝 Collaboration between humans and AI systems that leverage human feedback loops (e.g. reinforcement learning) will become more common.
• ⏳ Overcoming hesitancy around AI systems taking unpredictable or incorrect actions will take time as people get more comfortable with probabilistic outputs.
• 🔮 The guest believes AI assistants' capabilities will rapidly improve, increasing their accuracy and usefulness over time.
• 🏢 Enterprise sales teams are key targets for AI-powered sales intelligence tools that can streamline research and strategy.
• 🌐 Online communities like Twitter and Reddit are valuable for learning about and discussing the latest AI developments.
• 🤖 The guest is interested in eventually working more technically, learning areas like robotics engineering and woodworking.
• 🔑 Interfaces beyond just text/voice input are crucial for unlocking AI's potential across diverse use cases.

Q & A

What was the key frustration highlighted by the speaker regarding AI and text input interfaces?

The speaker expressed frustration with using text boxes and voice input as the primary interface for interacting with AI systems, finding them too limiting and unintuitive. The speaker believes that text boxes and voice input alone are not the best ways to interact with most software tools.

How did the speaker's experience with Anthropic's Omni AI agent differ from using vanilla ChatGPT or other language models?

The speaker found Omni's AI agent to be more sophisticated and capable of providing detailed and insightful outputs compared to vanilla ChatGPT or other language models. While the outputs from other models were sometimes inconsistent or off-base, the speaker saw glimpses of 'gold' in Omni's responses that showed real potential.

What were the three main objections or concerns that the speaker encountered from potential Omni customers?

The three core objections were: 1) Concern about hallucinations or inaccuracies in the AI's outputs; 2) Reluctance to change from established tech stacks and sales workflows; and 3) Skepticism from teams accustomed to traditional, relationship-based selling methods.

How did the speaker view the role of user interfaces and specialized tools in the future of AI adoption?

The speaker believes that successful AI adoption will require more specialized, purpose-built tools with intuitive user interfaces tailored to specific use cases, rather than relying solely on generalized text or voice input. Diverse interfaces beyond just text boxes will be necessary to make AI more accessible and user-friendly.

What did the speaker find intriguing about the work of Carl Jung, the psychoanalyst?

The speaker found Carl Jung's theories on consciousness, identity, and the origins of his ideas fascinating, especially in relation to discussions around generative AI and general intelligence. Jung's framework for understanding consciousness within non-organic systems was particularly compelling to the speaker.

How did the speaker's background in urban planning and affordable housing influence his career path?

After studying urban planning and working in affordable housing development, the speaker became interested in social entrepreneurship and community-driven initiatives. This led him to explore startups, which he found to be a faster way to drive impact compared to the slow pace of government policy work.

What was the speaker's perspective on the trade-offs between open-source and closed-source technologies?

The speaker acknowledged that there can be valid reasons for choosing open-source or closed-source technologies, and that neither approach is inherently better or more virtuous. The choice depends on the specific goals, problems, and use cases being addressed.

How did the speaker view the role of probabilistic systems and human feedback in the development of AI tools?

The speaker believed that more people will become comfortable with probabilistic AI systems that may not be perfect but can still provide significant value most of the time. Incorporating human feedback through techniques like reinforcement learning will be crucial for improving these systems over time.

What were the speaker's thoughts on the impact of AI on the physicality of day-to-day experiences?

The speaker expressed concern that the "software eating the world" prophecy has led to a loss of physicality in daily experiences. He wished for more innovation in hardware and tangible interfaces to complement the advancements in software and AI.

If given an extra five hours per day, how did the speaker plan to spend that time?

The speaker said he would spend three hours learning robotics engineering, specifically focusing on linear algebra, as he sees himself pursuing more technical work in the future. The remaining two hours would be dedicated to learning woodworking, an art and craft he has been interested in for a long time.

Outlines

00:00

🌐 Exploring Innovative User Interfaces for AI Interaction

The speaker expresses frustration with text-based interfaces for interacting with AI systems, as they often require learning specific prompts like 'magical spells'. The speaker advocates for more intuitive and user-friendly interfaces beyond just text input, such as buttons and graphical interfaces tailored to specific use cases. This would allow users to interact with AI agents more naturally and effectively.

05:02

💡 The Generalist's Superpower: Augmented Knowledge with AI

The speaker shares their experience as a generalist without deep expertise in any particular area. By using tools like ChatGPT, they felt empowered to take on more complex projects and tasks that would typically require specialized skills. AI agents and language models act as knowledge augmentation tools, enabling generalists to explore broader domains and tackle bigger problems. The speaker also touches on their background in urban planning and the transition towards startups.

10:04

🛠️ Embracing AI for Community-Driven Entrepreneurship

The speaker recounts their journey from affordable housing and urban planning to exploring social entrepreneurship and community-driven initiatives. This exposure led them to appreciate the potential of startups and new technologies to drive positive change more rapidly than traditional government policies. The speaker shares their optimism about the opportunities AI presents for doing good while also achieving business success.

15:06

🎯 Contextual Interfaces: The Future of AI Interaction

The speaker envisions a future where AI systems will be integrated into specialized tools with tailored interfaces specific to the user's goals and use cases. Rather than relying solely on text prompts, these tools would allow users to interact with AI agents through intuitive interfaces like buttons and drag-and-drop features. This approach would enable more natural and efficient collaboration between humans and AI, aligning the system's outputs with the user's desired outcomes.

20:06

🤖 Overcoming Objections to AI Adoption in Enterprise Sales

The speaker discusses the common objections and concerns that enterprise sales teams have about adopting AI solutions like Omni. These include concerns about the accuracy and reliability of AI-generated outputs, resistance to changing established sales processes and tech stacks, and a preference for maintaining control over every aspect of the workflow. The speaker explains how Omni addresses these concerns through ongoing model improvements, flexibility in integrating with existing systems, and probabilistic decision-making capabilities.

25:07

🤝 Human-in-the-Loop: Bridging the Gap with Reinforcement Learning

The conversation touches on the concept of reinforcement learning with human feedback, often referred to as "human-in-the-loop." This approach involves users providing feedback (e.g., thumbs up/down) to AI systems, enabling them to learn and improve continuously based on human input. The speaker acknowledges that while some users demand near-perfect accuracy, others are open to working with probabilistic systems that may occasionally make mistakes but offer significant value overall.

30:10

📐 Strong Opinions on AI Interface Design and Specialization

The speaker shares their strong opinion that text-based and voice interfaces for interacting with AI systems are limited and not the best approach for most use cases. Instead, they advocate for more specialized, purpose-built tools with intuitive interfaces tailored to specific contexts and user needs. This aligns with their belief that a diverse tech stack with specialized tools will emerge, rather than a few dominant platforms attempting to serve every use case.

35:12

🚀 Personal Interests: Robotics, Woodworking, and More

When asked about how they would spend an extra five hours per day, the speaker expresses their keen interest in learning robotics engineering, particularly the hands-on aspects like linear algebra. Additionally, they mention a desire to explore woodworking and arts and crafts, showcasing their diverse range of interests beyond their professional pursuits in the AI space.

Keywords

💡Prompting

Prompting refers to the process of providing textual input or instructions to an AI system, typically a language model, to generate a desired output. In the context of the video, the speaker expresses frustration with the limitations of text-based prompting as the sole means of interacting with AI tools. They argue that while prompting allows for flexibility, it can be difficult to accurately express one's intent, likening it to learning 'magical spells' to make the system work as desired. This highlights the need for more intuitive and user-friendly interfaces beyond text prompts.

💡Interfaces

Interfaces refer to the means by which users interact with software or systems, such as graphical user interfaces (GUIs) or command-line interfaces. The speaker believes that relying solely on text prompts as an interface for AI tools is limiting and advocates for the development of more specialized, purpose-built interfaces that better align with the specific tasks or use cases. These interfaces could incorporate elements like buttons, menus, and visual representations, making the interaction more intuitive and efficient than typing out prompts.

💡AI Agents

AI agents, in this context, refer to software systems powered by artificial intelligence that can assist users with specific tasks or workflows. The speaker discusses their experience with Omni, an AI agent focused on B2B sales, which can research prospects, generate messaging strategies, and support the sales process. AI agents aim to augment human capabilities by leveraging AI models to provide intelligent assistance, recommendations, or even autonomous decision-making within their specialized domains.

💡Hallucination

Hallucination is a term used to describe the phenomenon where language models generate outputs that are plausible but factually incorrect or inconsistent with the provided information. The speaker acknowledges concerns from potential customers about AI agents 'hallucinating' or providing inaccurate suggestions, which can undermine trust in the system. This highlights the need for ongoing improvement in model training and evaluation to reduce hallucination and increase the reliability of AI-generated outputs.

💡Generalists

Generalists refer to individuals with diverse interests and knowledge across various fields, rather than specializing in a single domain. The speaker identifies as a generalist and expresses enthusiasm for AI tools that can augment their abilities to tackle complex problems or projects without requiring deep expertise in specific areas. The AI community, such as the 'Generalist World' mentioned, attracts individuals with generalist mindsets who appreciate the potential of AI to empower their diverse pursuits.

💡Reinforcement Learning with Human Feedback (RLHF)

Reinforcement Learning with Human Feedback (RLHF) is a technique used in training AI systems, particularly language models, where human feedback is incorporated to refine and improve the model's outputs. The speaker mentions RLHF in the context of users providing feedback (e.g., thumbs up or down) on the outputs generated by AI tools, which can be used to reinforce desirable behaviors and refine the model's performance over time. This approach aims to align the AI system's outputs more closely with human preferences and expectations.

💡Probabilistic Systems

Probabilistic systems refer to AI models or algorithms that generate outputs based on probability distributions rather than deterministic rules. The speaker acknowledges that AI agents are inherently probabilistic, meaning their outputs may not be 100% accurate or consistent every time. However, they argue that even if the system is correct 80% or 95% of the time, the added value and time savings it provides can be significant. This highlights the need for users to embrace the probabilistic nature of AI systems and be comfortable with occasional inaccuracies or unexpected outputs.

💡Special-Purpose Tools

Special-purpose tools refer to software applications or platforms designed to address specific tasks or use cases, rather than being general-purpose tools. The speaker advocates for the development of more special-purpose AI tools, each with tailored interfaces and functionalities optimized for their intended use cases. This contrasts with the current trend of relying on a handful of general-purpose AI tools like ChatGPT for a wide range of tasks. Special-purpose tools can provide a more user-friendly and efficient experience by aligning closely with the user's goals and workflows.

💡Multidimensional Interaction

Multidimensional interaction refers to the ability to interact with AI systems through various modalities beyond just text or voice input. The speaker highlights the need for AI tools to support different interaction mechanisms, such as buttons, menus, drag-and-drop interfaces, or even physical controls. By allowing users to engage with AI systems through multiple dimensions, the interaction becomes more natural, intuitive, and aligned with the specific task or context, potentially improving efficiency and user experience.

💡Robotics Engineering

Robotics engineering is the discipline that involves designing, building, and programming robots, including their physical hardware and software systems. The speaker expresses a strong interest in learning robotics engineering, indicating a desire to pursue more technical work in the future. This interest aligns with their enthusiasm for exploring tangible and physical interactions with technology, beyond just software interfaces. Robotics engineering combines principles from various fields, such as computer science, mechanical engineering, and artificial intelligence, to create intelligent and autonomous robotic systems.

Highlights

When I first started using Omni it like there was just this like holy moment where um you know, you put in your your query and then at the time was just this research tool so, you would put in your query and it would, go off and do research and you know two, minutes later you'd get this whole, report that would have taken you, probably half an hour to put together by, yourself

And when then when we started, to build the sales tool so this was a, little bit uh later on um same moments, happened where the anal would it would, go off and do research on these uh you, know prospects that that you were trying, to sell to um and it would make these, inferences about how to sell to them uh, what messaging is going to be most, effective for various buyers

And it was, it was still early on and so the tech, was a little bit inconsistent sometimes, the answer would be a little bit out of, left field but sometimes there would, just be gold in there and like I knew, seeing some of these um outputs that, were so much more sophisticated and so, much more detailed than what you would, get from vanilla chat GPT or bar there's, really something here that's special

I think the thing that um and then your, prior guest who who we were talking, about earlier the account executive um, he he put it in in such a perfect way, that that uh I also really um resonate, with like when I started using chat GPT, to write code I'm somewhat technical I, would say right I have a background in, programming through undergrad and, independent projects in grad school but, like by no means am I an engineer or, data scientist or anything and when I, got my hands on chat chat GPT and, started to build out projects with it, felt like I had superpowers

My undergrad I I, didn't know really have any sort of uh, plans going out of school I didn't want, to be a consultant I didn't want to be a, banker I didn't want to be an engineer, uh but I was really interested in cities, and urban planning and fell into that, kind of career path

Ended up working internships and, through some startups with various, companies and and Founders and investors, that were in that space uh and that was, kind of like a gateway drug into, startups in general where I just felt, like so much more at home being a, startup operator than working in like a, traditional government policy setting, which is just so slow

And there are a lot of other, tools that are philosophically in in, kind of the same same boat it makes me, think of um like I've used a lot of I, want to say like no code web scraping, tools where it's like they and I've just, leared the word alignment and what that, means in AI and that's like the AI or my, definition I'm going to butcher it but, that it's doing the thing that I want it, to do in the way that I want to do it

So like that that's like the, direction that we're building at Omni, for example and there are a lot of other, tools that are philosophically in in, kind of the same same boat it makes me, think of um like I've used a lot of I, want to say like no code web scraping, tools where it's like they and I've just, leared the word alignment and what that, means in AI and that's like the AI or my, definition I'm going to butcher it but, that it's doing the thing that I want it, to do in the way that I want to do it

Interestingly I I never, really especially wanted to work at an, AI company I was more just drawn by um, the really interesting technology and, the opportunity to work with these tools, that I you that are just so fascinating, um and have so much, possibility

Um and you know actually what's really, interesting is that a lot of these teams, are the ones that are most eager to, adopt a tool like Omni or something in, our in our realm once they kind of get, comfortable with it I use it um and tral, it out because they don't even have to, learn all this super complex automation, right they don't have to have a revops, team they don't have to do all of this, it's it's literally like a plug-and, playay system that works

I think what, we're definitely encountering is like, some people are are very open to using, these systems and are comfortable to, like deal with the underlying models, which are unpredictable inherently, they're probabilistic systems and I, think some teams just require absolute, certainty that everything that gets, output is what they expect it to be but, yeah these these the technolog is always, improving and I think you know similar, to how every kind of technological era, comes with you know a lot of people who, are uncomfortable using different, technology or thinking about different, workflows or just like different ways of, solving problems and then there are, people who are very comfortable in those, modes um we'll see the same things uh, here where I think over time more and, more people will just start to, understand probabilistic systems better

I think my strongest, opinion is um around interfaces other, than um chat input uh to interact with, these systems and when I say chat I both, Voice and text I I truly think that, while that's a great way of interacting, with language models and certainly it's, it's kind of like like you know text is, the fundamental unit of interacting with, language models of course but I think, from like a user perspective you'll see, more and more interfaces that are you, know I think more comfortable and more, natural to whatever the use case and the, context is

And along with that I think, what what uh met necessitates is more um, special purpose tools so rather than, going to chat GPT for everything you, want to do I think it'll be look much, more like yeah like software prior to uh, chat gbt where you had you know your, tool that does this and your tool that, does that and and you know you you had, this like more diverse Tech stack that, you would go to

I really don't think, we'll see uh you know just like a, handful of software platforms dominate, everything um when it comes to like, where users spend their time to uh get, work done or or get play done or, whatever I I do think that we'll see, quite a bit of diversity still um and, that'll just be necessitated by like, it's impossible to build for every, single use case well that there's a lot, of value in building for specific person, and um that will just then necessitate, different design, decisions

So so actually I, wish I had way more than three hours you, can have five I'll give you five five, okay well i' spent three of those hours, I've been really interested in robotics, and I I want to learn Robotics and, specifically like robotics engineering I, think longterm I see myself doing work, that's much more technal technal than, what I do today so robotics engineering, and like I've been learning uh spending, some time self learning linear algebra

Transcripts

but it's like one of my Crusades is as

as long as I work in this space is

certainly to show that the text box is

is you know and text is just not the

only way to inter interact with these

tools Renee here at UNS superb learning

your easy listening podcast for bleeding

edge open source Tech no I think there

is something to like like physical

things so I spoke to someone the other

day that uses a stream deck to do home

automation and he said that he like

twists a thing and then the lights come

on and I was like oh that's amazing yeah

that's what I don't know what rabbit

does at all but that's like piece of

Hardware so that's one of my wondering

if Hardware is going to be a thing in

the future I don't know yeah Hardware is

I think there's so much it's one of like

those things that has really

disappointed me actually if you if you

want to say like like another hot take

about tech an AI I've been so more Tech

but like I've been so disappointed about

how little like like Innovation there's

been in Hardware you know over the last

like 20 years like yeah of course I mean

we have like VR glasses you know VR AR

stuff and we have iPhones and like

there's like some great things but those

are are like singular inventions in like

a decade right um when you think about

that compared to how many different like

typ of software we have and and you know

it's a matter of Economics it's a matter

of Supply chains and difficulty of

building new hardware um I wish there

was more because it's so something about

it just being tangible makes it so

satisfying like to hold to to interact

with I

think yeah that's one downside I think

to the sort of software eating the world

Prophecy from you know andreon like

which has probably played out to be more

or less true so far um

but we've lost something about the

physicality of our day-to-day

experiences that um I wish we could get

back you know what I'm going to title

this

video what man angry at lack of

hoverboards that shakes fists because

there is no flying cars J promised me

flying cars yeah but no I get what

you're saying so I'm sitting here today

with film star Natalie Portman I'm

kidding you know the Lonely Island what

Lonely Island like the the band like the

SNL thing yeah yeah yeah yeah yeah that

was from the Natalie Portman rap in like

2009 my God and every time I hear I'm

sitting here today I think a film with

film star Natalie Portman soya this is

the thing I've never said your last name

out loud middle ear oh like middle ear

yeah yeah like so middle ear who is

working as chief of staff at Omni an AI

agent for you still

BB sales BB sales and do you want to

kick it off maybe some background into

how you got into the AI space was Omni

your gateway drug yeah so I actually

first got into the AI space building

independently um so when I worked at the

company I was at prior to Omni uh was

seeing chat GPT launch and a lot of cool

AI tools launch and I've always sort of

been a tinkerer when it comes to new

technology and and a lot of stuff that's

been open sourced and just available

online and so was really eager to get my

hands on the open AI API um and started

to build some independent projects with

it pretty much as soon as it came out

and so that was my first foray into all

of this and built some uh little little

projects which I can talk about later on

I guess but uh no otherwise I'm like a a

kind of proper work capacity Omni was

the first time I got to work with

generative Ai and prior to Omni the only

experience that I had had with uh

machine learning and AI systems was uh

pricing optimization models um that I

worked on at at an e-commerce company

before that um yeah so we joined Omni

working with uh at the time the solo

founder David back in late May early

June um really just doing the very early

stages of um user research market

research and then eventually um you know

built it into a full company and where

are you at now with with Omni like who

do you serve so right now Omni we serve

uh customers um in B2B sales but the the

unifying thing is that these are

companies in organizations that sell to

larger midmarket Enterprise buyers so we

really support these more complex um

longer sales cycles that are very

research driven and very kind of

consultative in nature and I only found

out the other day that did you say

you're a python teachers assistant I was

yeah in grad school uh was ta for our

data analytics with python class class

ES that's really cool I remember I think

I don't know if it was a tweet that you

that you did that was like um said

sequel is unlock to I'm just going to

make up a quote that you said and just

put words in your mouth but it's

essentially like sequel is like an

unlock for Ops people and I sat I sat

with myself and I was like I don't know

sequel but it was a great point I feel

like python is something like

foundational that is great to know going

into AI you just explained how you made

the decision to to go with your ICP but

you surprised by how much these AI

agents could do or like when you joined

Omni what did you think you were joining

early on I honestly had no idea um it

was pretty amazing to see this vibrant

and like super passionate community of

users just from the beginning I mean

this was the height of the hype and

craze around at the time I think it was

GPT 3.5 which was out and maybe GP GT4

at that point um and there was just so

much excitement around it so it was

super fun to talk to users where I DM

somebody or email somebody and of course

they would want to talk to me and would

interview them uh and yeah it was a

really great experience so it started

off like I said sort of just doing

interviews and market research to try to

figure out like what direction should we

point the ship in um and then uh it

became something that was a little bit

more L and marketing oriented um and

also just the basic operations starting

up the company and getting our

accounting in order getting our legal in

order I think interestingly I I never

really especially wanted to work at an

AI company I was more just drawn by um

the really interesting technology and

the opportunity to work with these tools

that I you that are just so fascinating

um and have so much

possibility uh when I first started

using Omni it like there was just this

like holy moment where um you know

you put in your your query and then at

the time was just this research tool so

you would put in your query and it would

go off and do research and you know two

minutes later you'd get this whole

report that would have taken you

probably half an hour to put together by

yourself and when then when we started

to build the sales tool so this was a

little bit uh later on um same moments

happened where the anal would it would

go off and do research on these uh you

know prospects that that you were trying

to sell to um and it would make these

inferences about how to sell to them uh

what messaging is going to be most

effective for various buyers and it was

it was still early on and so the tech

was a little bit inconsistent sometimes

the answer would be a little bit out of

left field but sometimes there would

just be gold in there and like I knew

seeing some of these um outputs that

were so much more sophisticated and so

much more detailed than what you would

get from vanilla chat GPT or bar there's

really something here that's special

CU I know we met through generalist

World which is a community online

and I think that there is like a really

big cross-section of people interested

in Ai and very generalist mindsets or

very um potentially like neurod

Divergent people not labeling you but

definitely um yeah so I I don't know if

you want to start with urban planning or

your own projects or whatever sure uh

well I actually let me start with the

general World thing and then I can talk

more about the urban planning so I I

think the thing that um and then your

prior guest who who we were talking

about earlier the account executive um

he he put it in in such a perfect way

that that uh I also really um resonate

with like when I started using chat GPT

to write code I'm somewhat technical I

would say right I have a background in

programming through undergrad and

independent projects in grad school but

like by no means am I an engineer or

data scientist or anything and when I

got my hands on chat chat GPT and

started to build out projects with it

felt like I had superpowers like you

know here I am like it would take me

probably a month to build the things

that with chat GPT I could put together

in a weekend and so as a generalist with

a lot of different interests and you

know no super deep expertise certainly

in any technical area or any area of

like subject matter the most part um

being able to have this like

augmentation really of of knowledge um

that you know and then when you have ai

agents that layer on even more kind of

ability to S sort of support you in

workflows that are more complex than

just kind of prompting something and

getting an answer back these tools are

so exciting because it it lets me

without any of those uh specialized

skills um really take on bigger problems

and projects umly so so that's been

great and yeah the urban planning thing

I mean you know I think it's a a pretty

typical story for anybody who's got a

kind of generalist skill set and and

background where um my undergrad I I

didn't know really have any sort of uh

plans going out of school I didn't want

to be a consultant I didn't want to be a

banker I didn't want to be an engineer

uh but I was really interested in cities

and urban planning and fell into that

kind of career path uh first in the

private sector doing uh affordable

housing and mixed income housing

development and then went to grad school

to do research around affordable housing

and um Community Development and when I

was in grad school got very interested

in social entrepreneurship and a lot of

the work that was going on in Chicago

focused on kind of community-driven

Entrepreneurship local entrepreneurship

and ended up working internships and

through some startups with various

companies and and Founders and investors

that were in that space uh and that was

kind of like a gateway drug into

startups in general where I just felt

like so much more at home being a

startup operator than working in like a

traditional government policy setting

which is just so slow right like I was

like kir was slighting in policy School

seeing you know wh Mo hit the streets

with autonomous vehicles and Airbnb was

having huge impact on the property

market and city and state governments

were just like twiddling their thumbs

with no budget to do anything um and you

know as somebody who's fairly impatient

I became a fairly easy choice to say hey

there's this other career path where

perhaps it's not as like Mission driven

per se but there's a lot more

opportunity to actually have an impact I

felt and yeah that's what sort of taken

me in this direction absolutely have

like a healthy cynicism around this

stuff that like I think at the time I

was a bit naive to but I I honestly do

and you know perhaps still naively

believe that like there are a lot of

opportunities to do a lot of good in

this space and still you know make

shareholders happy and all that all that

stuff doesn't have to be just you know

all one or all the other um I think it

was I say Dan from Jan Daniel the

co-founder of Jan he mentioned it was

like in a public discussion on GitHub of

all places that there is like a little

bit of virtue signaling with the open

source space and it's like just because

something's open source not proprietary

technology doesn't mean that it's

inherently better or virtuous or there's

there's a lot of um I I think I am a

professional

offence and it's it's not for lack of

conviction it's just for openness for

whole strong opinions loosely held and

so I feel like I think it's something

that we potentially both struggle with

as generalists you're so open to so many

different things that it can seem like

oh you've got no opinions and it's like

no I do I've got Direction Just lots of

different directions all at the same

time absolutely or it could be like I

have you know there there might be a

goal that I want to achieve or a set of

outcomes that I want to achieve or

problem spaces that I'm really excited

about working on but I I think I bring a

more diverse tool set then somebody else

who might say hey look I've got a wrench

and a screwdriver like I I can only use

these two things whereas I'm coming in

with a wrench a screwdriver a set of

paints you know a computer over here all

this other random stuff and and so you

get to be a little bit more creative but

I think at the same time you always have

to like be able to answer that question

of like why do you need to bring these

tools to this particular problem like

are so yeah there's always a trade-off

that you make right and it's the same

thing there's also a trade-off that you

make with open source versus closed

Source there are good reasons why you

might choose one versus the other yeah

well you you mentioned about

frustrations and this isn't I feel like

this should be a question that I ask

people often what frustrates you most

about using like ai ai agent or any part

of the process oh what frustrates me the

most that's an interesting question I my

mind goes to prompting uh I don't want

to be what's the word suggestive though

I don't want to give you no I think

that's it I I think what really

frustrates me the most about a lot of

using a lot of these tools today is that

like text input uh text box is just not

the best interface to interact with most

software tools it's the same thing that

frustrates me about like you know voice

activated like Alexa and stuff like it's

great for some tasks right I love my

parents Al Alexa I can tell it to play

music but I can't ask it to do much else

for me right um For Worse the computer

screen with the mouse and keyboard and

buttons and and all of that uh is a

really really good user interface um and

not to say that there aren't others and

I think VR that's all really cool and

and will come up and and I just think

that like a text box is so limited um

and in particular it's not so much that

it's like um only one way of interacting

with something but there's so much it's

so difficult sometimes to actually

Express what you want uh through that

simple input um you have to learn what

are basically these magical spells that

that you need to input to this machine

to get it to work right and it's it's

not as if there's like that much logic

to it necessarily like there is a lot of

logic and their structure and like there

there are a lot of great guides on

prompt engineering out there um but

still there's like the peculiarities of

how these systems are trained and and

the fact that many of them are black

boxes means that in order to prompt them

Well there's almost this like trial and

error that reminds me of um when I used

to study history in school and would

read about how the ancient Romans would

learn all these like magical spells in

order to make sure that their crops

would grow correctly and that their

armies would win in battle they they had

to learn all these intricate little

steps and words to say just because

these things are of course very

difficult to influence um and it's the

same thing with these these models in a

kind of strange way that's how I always

feel using them like I just want to

click a few selection boxes and it'll

make my life so much easier but um as as

general purpose tools they need to be

flexible right they can't like the the

chaty BTS and Bs right if if that's all

you have is input this one text box that

you use to do everything it's on the one

hand unlimited and totally open sandbox

which is great but on the other hand

it's so difficult to interact with

sometimes if if there's a specific

output that you're looking for it's kind

of like needing um it's like if you're

working at I don't know any SAS and it's

like you need to be an expert in hub

spot and air table and whatever else

suddenly knowledge workers are all that

more valuable because like yeah you can

have the engineering tool but who's

going to drive it and this is like me

battling for revops people everywhere

because I feel like they're so valuable

um yeah but imagine like if in order to

like you know put together a particular

HubSpot uh like I don't know

customization or something like yeah it

works automatically which is great but

you have to spend half an hour trying to

figure out how do I you know prompt this

thing just to get this exactly correct

output and it's fun don't get me wrong I

love tinkering with this stuff and

exploring it and there are people I'm

sure who are much better account

Engineers than I but it's like one of my

Crusades is as as long as I work in this

space is certainly to show that the text

box is is you know and text is just not

the only way to inter interact with

these tools and neither is voice like

there there has to be some kind of

multi-dimensional interaction mechanism

like neurot Tech like no no no I just me

something super simple like like uh

buttons and interface I think and this

is why like and you can only make this

happen if if you have more special

purpose tools where you can as um a

designer and as an engineering team or

as as a product team anticipate what are

my users trying to accomplish with this

tool so that I can build an interface

that supports it right that provides

these mental shortcuts these like uh

metonyms so to speak right the these

analogies within user interface so that

a user rather than having to type out in

words every single thing that they want

to do and want to accomplish they can

simply just click here drag there

whatever and it just works uh the way

that they intend to I'm thinking about

those gesting like the almost the no

code Builders where it's like you drag a

block and it's like this block is a I'm

actually thinking of something a little

bit different which is not so much a no

code Builder so much as it is very very

simple user interfaces

where the the reason why you come to

this tool is to accomplish a specific

task right and you have let's say an AI

agent within the tool that understands

what your goals are right you're able to

tell this this software that you're

collaborating with what goals you're

trying to accomplish what data you have

access to in order to make decisions

about how to accomplish those goals and

then the AI is intelligent enough to

actually accomplish them and make

decisions about how to do it right and

so rather than having to put together as

you're describing these fairly

complicated uh flowcharts if then rules

about um how to build out these these

decision trees and and workflows it's

actually something much simpler which is

to say like hey I want to be able to

click a button here or maybe even not

take any action at all and the AI system

should be intelligent enough to use the

information it has to help me accomplish

my goal so like that that's like the

direction that we're building at Omni

for example and there are a lot of other

tools that are philosophically in in

kind of the same same boat it makes me

think of um like I've used a lot of I

want to say like no code web scraping

tools where it's like they and I've just

leared the word alignment and what that

means in AI and that's like the AI or my

definition I'm going to butcher it but

that it's doing the thing that I want it

to do in the way that I want to do it

and it know like if I'm like grab me

content of the latest Reddit posts it it

knows that I want it to go onto this

subreddit and only grab like the top

five most relevant it's not going to go

and grab the whole page and all of that

sort of thing and the other other thing

is I'm thinking

is that does that mean that process

automation is AI like not necessarily I

mean I think it's kind of the reverse

which is that the best AI systems are a

much more advanced version of process

automation than I think many people who

don't work te in technical spaces are

used to right so I think a lot of folks

who work for example in revops are very

accustomed to these workflow Builders or

they're accustomed to these spreadsheet

Integrations where there's a lot of

configuration that you need to do ahead

of time and I think what a lot of people

don't appreciate is that when you are

building those workflows and these

automations you're building in a lot of

assumptions about the decisions that one

should make using the data right but in

reality humans tend to be very bad at

integrating a ton of data and making

decisions about what filter criteria

make the most sense or what sorts of

decisions are optimal based on the

information that I have and rather than

building those assumptions into the

system I think a lot of people will see

that it's a lot better to say here's the

data I have here's the outcome that I'm

trying to achieve let's say it's you

Revenue right this as an obvious one and

here's the the tools that I have to work

with and the AI can be much more

intelligent to actually make decisions

about how how best to do that right it

can either suggest and collaborate with

a user to accomplish those goals or in

some cases where it's very low stakes

decisions it can just take those actions

by itself and like one simple example

right is um let's say you have a

research agent that's um you know

helping you for example ly does right we

we aggregate a bunch of information

about the the companies that you're

trying to sell to from a bunch of

different data sources and the internet

and our agent is smart enough to to

understand based on that information

which of those pieces of data are most

relevant for the seller to use within

let's say a discovery call or something

right we're not explicitly telling it

look for this keyword we're not explicit

explicitly telling it look for

information within these dates rather

what it's doing is saying given all this

information and the objective of selling

to this person what information actually

is best and that's just like a very

simple example but I think as these

systems get more Advan and people get

more comfortable working with them

you'll see that kind of

um uh that that kind of system become

more and more commonplace it's

interesting like I think it was a I want

to say predy pry Bas pretty Bas they're

in the UK and they had a survey that

like 220 something people and it was who

um like why people the uptake of llms

aren't isn't great in businesses and

it's like all the different reasons like

oh it hallucinates I don't want to give

it access to my data all these different

things and given that Omni is primarily

B2B sales lead focused in Enterprise and

small to medium businesses what kind of

jections do you like if any I'm assuming

um or like do you have to overcome tyal

usually there is like there really Three

core things that we hear over and over

um and they're all understandable I mean

the first one is like concern around

hallucination right where something just

the effect of um oh the AI suggested X

but you know in my opinion I would do y

right and um it works 80% of the time

that 20% of the time I would do it

differently uh and that's really

understandable I think that that

definitely mirrors a lot of the

discussion that I've heard around like

say autonomous vehicles right where I

mean obviously what we're doing is much

lower Stakes uh much um you know making

mistakes is less you know destructive in

our case luckily but it's interesting

because I think a lot of times people

get very cut up on the negative and kind

of deemphasize or underappreciate

the value that even if we're only

correct 80% of the time uh that 80% adds

a ton of value um but even so we

actually tend to be right more often and

I think typically what happens is that

people value their own experience and

they value their own like yeah they

value their own judgment uh sometimes at

the expense of data and um so it takes a

very open-minded person I think to be

okay to like start to work with one of

these systems I think some people will

demand that their system be perfect you

know 99 or 100% of the time in order to

feel like they can trust it but I think

a lot of uh people certainly our early

customers are you know perfectly fine

they can see the value with um all the

value that we add in terms of time

savings and Improvement on the accuracy

and quality of the information they get

and like the the strategies that we're

able to build for them you know and if

we sometimes uh pull slightly out of

data information or the insights aren't

exactly what they would have said like

they trust the system that it's um

either working as intended or you know

they're okay to sort of work out the

Kinks I guess and kind of let because

the systems are getting better all the

time and frankly our our AI agent like I

said 8020 we're not 8020 anymore I mean

I'd say that we get it right far more

often than we get it wrong uh and we're

only improving as the models improve and

as we just get a lot better at training

these agents so that the second thing is

um sort of related uh which we

definitely hear a lot more around um

more mature sales organizations which is

like some sales teams have these Tech

stacks of like 30 different tools you

know not really it's probably like eight

different tools but these things are all

wild wired together and they have like

their way of selling right and they have

so much invested both in terms of money

and um training and you know also

probably a little bit of like ego in

doing it their way that sometimes like

there definitely people who just are not

interested in thinking about oh maybe I

could sell differently right like maybe

I don't have to send you know a thousand

emails per you know inbox a day to hit

my number maybe I can do it much better

with say fewer emails or maybe I can you

know drop a few of these Point Solutions

and pick up something that builds a lot

of the functionality into one and

inherent in that is not relying so much

on having so like like micro level

control over everything that happens but

you know being tolerant of a

probabilistic system um within your

workflow which I think could be really

challenging for some people um and then

the last thing is just that there are a

lot of teams that like see an AI system

and they've been selling you know their

their sales teams uh especially at like

larger Enterprises Legacy companies

they've been selling one way for 30

years you know they don't use like any

sales Tech that's launched in the last

probably eight to 10 years like that

they have their relationships they meet

face to

face um

and you know actually what's really

interesting is that a lot of these teams

are the ones that are most eager to

adopt a tool like Omni or something in

our in our realm once they kind of get

comfortable with it I use it um and tral

it out because they don't even have to

learn all this super complex automation

right they don't have to have a revops

team they don't have to do all of this

it's it's literally like a plug-and

playay system that works um and almost

like without that baggage of like the

last 10 years of sales Tech um they're

they're a lot more tolerant of working

with these systems and and you know it's

just sort of a learning curve issue I

think people think like there is a lot

of oh AI has been around for so long

it's like well actual

usable no like I'm I I feel like a large

majority of people think that have just

suddenly it's almost suddenly come into

Consciousness since like chat GPT was

only what like a few years ago so to

expect I think it's so right to expect a

level of um almost like Grace not

lenience but Grace and understanding

that it's collaborative but then how do

you move that and like align that with

the whole like everything needs to be

Roi right now if it doesn't work like I

was doing a almost like a self study on

how quickly I will churn or leave a

product if it doesn't fit my exact

expectations and I was like I'm a

horrible end user I was mentioning to

someone on a call earlier today that

last year I tried 174 tools who like

Renee that's insane I went through like

my receipts my old signups and

everything because I'm an earlier doctor

of a lot of different stuff like really

enjoy it but I think my um I I

understand like on a on a not a just on

a level that it's like yeah you you do

need to have understanding for these

teams like it's a startup you you kind

of got like a window of expecting bugs

or lack of functionality but it made me

think about like the whole human in the

loop thing because you mentioned I know

that there's a right term for it and I'm

missing it right now where it's like it

will check what's that called do you

know what that's called um yeah like

like uh human feedback basically I mean

or or are you talking about like

reinforcement learning I don't really

know what I'm talking about yeah I mean

like like like yeah you'll hear like rhf

right so like okay yeah like

reinforcement learning uh with human

feedback yeah yeah yeah reinforcement

learning from Human feedback so it's

like you know the simple example is like

if you're using chachy BT there's the

thumbs up thumbs down and yeah I mean I

think maybe a better way of putting what

I was saying before is like I think what

we're definitely encountering is like

some people are are very open to using

these systems and are comfortable to

like deal with the underlying models

which are unpredictable inherently

they're probabilistic systems and I

think some teams just require absolute

certainty that everything that gets

output is what they expect it to be but

yeah these these the technolog is always

improving and I think you know similar

to how every kind of technological era

comes with you know a lot of people who

are uncomfortable using different

technology or thinking about different

workflows or just like different ways of

solving problems and then there are

people who are very comfortable in those

modes um we'll see the same things uh

here where I think over time more and

more people will just start to

understand probabilistic systems better

and say hey like this is going to work

um 95% of the time and 5% of the time

it'll throw some kind of weird response

but I'm kind of aware enough of that

that'll happen and I can catch it and

regenerate that response and just go

about my day that's such a good point

you started the call I said that you

were a little bit you struck me as like

very serious and like I I I know from

knowing you that like you're a pretty

realist person but that's a very

optimistic take for you to have um where

are you keeping up with like because

you're always online you're always awake

uh just for anyone actually really good

recently about getting my eight hours of

sleep oh well done yeah um but no I'm

definitely sort of a night owl sometimes

yeah where do I keep up I honestly like

Twitter is a great source of information

like as as much of a cesspool as it is

as everybody says like I hate it but I

also find it really valuable Reddit as

well and more and more I think I've just

been like trying to be more like

directed in my learning and so you know

when something peques my interest say on

Twitter then I go down the rabbit hole

of like looking at product documentation

if it's like a technological tool or

I'll go on YouTube and and try to find

longer form say video content so I I

think I've been like I try to spend as

little time on Twitter frankly as

possible uh but I I've enjoyed kind of

going there finding some interesting

things and then pursuing uh you know

more learning elsewhere that's very fair

do you have someone that you would

interview that you would love to

interview one of my classic questions oh

I don't know about on Twitter no they

don't have to be on Twitter okay okay

okay I mean in AI in general oh in

general um can it be a like a past

person like for person from history

because yeah so the person that that

comes to mind immediately is Carl Young

uh the psychoanalyst yeah he's so

interesting I'm reading an autobiography

of his where he just talks about like he

analyzes his own dreams and talks about

the origins of his theories on

Consciousness and identity and it's so

fascinating and I think when you go down

the rabbit hole of a lot of um

generative Ai and kind of the

discussions around um like you know kind

of um like when you go down the rabbit

hole of conversations around like

general intelligence um like his name

always comes up and his theory is always

come up as sort of this like framework

for understanding Consciousness within

like non-organic systems um yeah I I

think he'd be such a great uh

conversationalist strong opinions in AI

do you have strong opinions opinions I

do yeah I mean I think my strongest

opinion is um around interfaces other

than um chat input uh to interact with

these systems and when I say chat I both

Voice and text I I truly think that

while that's a great way of interacting

with language models and certainly it's

it's kind of like like you know text is

the fundamental unit of interacting with

language models of course but I think

from like a user perspective you'll see

more and more interfaces that are you

know I think more comfortable and more

natural to whatever the use case and the

context is and along with that I think

what what uh met necessitates is more um

special purpose tools so rather than

going to chat GPT for everything you

want to do I think it'll be look much

more like yeah like software prior to uh

chat gbt where you had you know your

tool that does this and your tool that

does that and and you know you you had

this like more diverse Tech stack that

you would go to um I really don't think

we'll see uh you know just like a

handful of software platforms dominate

everything um when it comes to like

where users spend their time to uh get

work done or or get play done or

whatever I I do think that we'll see

quite a bit of diversity still um and

that'll just be necessitated by like

it's impossible to build for every

single use case well that there's a lot

of value in building for specific person

and um that will just then necessitate

different design

decisions I've got a question that has

no relation to that if you had an extra

three hours a day what would you do oh

easy I'm really into uh so so actually I

wish I had way more than three hours you

can have five I'll give you five five

okay well i' spent three of those hours

I've been really interested in robotics

and I I want to learn Robotics and

specifically like robotics engineering I

think longterm I see myself doing work

that's much more technal technal than

what I do today so robotics engineering

and like I've been learning uh spending

some time self learning linear algebra

and so I'd spend more time to do that

and then the other one um is I love like

art and and arts and crafts and things

and um I've been wanting to learn how to

do woodworking for a long time and so I

think I would do that that's awesome you

can't say oh maybe you can um you say

the the wand up there yes that is a

Snape's hand with a wand from Harry

Potter and my dad wood whittel that my

dad whoa that's awesome my dad wood

whittles he I've got like most of my

cupboard back there is so cool wood

Wht say it probably that word it's

hilarious but yeah um where so you don't

live on Twitter I was going to say you

live on Twitter where are you able to be

found feel free to plug something well

you can find me on Twitter if you want

uh but I'm probably more often on link l

in uh just at my name um also check out

Omni docomo M ni.com

um yeah and come say hi we love to chat

with folks awesome thank you so much

that wraps up this week's episode of

unsupervised learning I'm your host

Renee and I've had a great time chatting

with you as always links to everything

we discussed will be in the show notes

make sure you reach out to our guests

questions or feedback reach out to pod

unsupervised learning. until then leave

a like follow or writing on Spotify

Apple podcast or YouTube and until next

week stay curious