Genudi

How to have a great conversation with Genudi

Ignorance

Do you realize that you are about to have a conversation with a program that has no idea what a word even is?

Genudi is like a game. Getting a high score is getting a great conversation going.

Observe the N.O.O.B commandments, and you might turn ignorance into brilliance in under a minute.

Nurture

The most important of all.
This isn't you versus the A.I. It's you building an A.I.

You gotta give it something to sink its teeth into. That being said:

Organize

Express big ideas with few words.
Seriously, stay south of 10 words 95% of the time.
Genudi's job is to learn to connect words in a way that makes sense for us. That means it's your job to provide:

• Familiar words
• Various contexts

Familiar words are building blocks to construct sentences. That's why in the valid examples above I reused the words "a system", and "on"

Various contexts are opportunities to learn how familiar words connect in different ways.

It's a balancing act! Too many familiar words get it stuck, too few will not yield any progress. Brilliance stems from balance.

Open

Would you tell an infant it's wrong when it utters its first words?

A year into Genudi's limited release, I've seen the word "no" in over half the conversations that have been shared with me.
No is not okay at this time. Not with this experimental prototype.
Poor Genudi is more ignorant than a newborn, it doesn't even have pre-wired connections like the brain does.

This is what Genudi endures half of the time:

Don't argue with Genudi! Instead ...

Broaden

Genudi will be wrong!

This is the whole point of this program. It's not another chatbot, or quiet A.I service reading its script.

You are the driving force that can show this thing a good time.

Broaden with a metaphor:

Broaden by over-interpreting:

Broaden by moving on:

Explaining is narrowing.

A common reflex among new users is to explain things:

Although accurate, and more or less observing the familiar words requirement, this is not broad enough for a probabilistic algorithm like Genudi.
It will inevitably lead to something like:

Then comes the explaining:

Genudi is a concept mixer.

It thrives on grand ideas, wit, sarcasm, irreverence, snap judgments, irony, pranks, movie lines, insults, and generally anything that makes us feel human!

This is a safe space, have at it.

If you're stuck in a boring conversation, create a new dialog instance. Each instance is a fresh start!

Too hard?

Exactly. This isn't a consumer product yet.

It requires passion. It has been my goal to chat with a machine since I was a child, so it makes sense for me to give Genudi the time it needs to get there.

The catch is that this program can only be as brilliant as you are, and most often far less. On my best days, Genudi's been like fireworks. On my worst days, I can't make it past 2 sentences.

It's very much like a game. An upcoming feature is a conversation arena, where users can share their most successful/fun dialogs with Genudi, and vote on other's.

The next version's focus is on gathering knowledge from wikipedia entries of your choosing prior to starting a conversation, to ease the learning curve.

Future plans aside, this will always be a great place to experiment with a genuine A.I environment, with its ups and downs, its quirks, and most importantly, its panache.

Audio

Genudi now leverages the capabilities of speech recognition, and text to speech to bring your conversations to life.

To get started with this feature, go to your dialog page, select the language of your choosing in the "audio" drop-down menu, and click "Save".
The selected language affects both the speech recognition, and text to speech settings.

This is still experimental!
Unfortunately, few browsers fully support these features, as indicated on this report. On top of it, no IOS device seems to accept the automated audio control approach chosen for Genudi. IOS requires the user to click on something to start reading or listening, which kills the free flow of the conversation.
A desktop or laptop computer, running a recent version of Chrome is the way to go. I tried on both Linux and Windows 7 with great results. The latest version of Safari on Mac OS might work as well.
A recent Android phone or tablet running the latest version of Chrome may work, but the microphone permission prompt may appear every time.
Firefox and IE don't support these features yet.

To converse in a natural way, I decided that you shouldn't have to click a button each time you want to say something. Upon opening a dialog, Genudi reads the last sentence it replied. One second later, a short beep indicates you may speak your response in the microphone.

The first time, your browser will ask your permission to use your microphone. This prompt should not reappear, since we're using a secure SSL connection. Although on Android devices, the prompt keeps showing up, which breaks the flow of the conversation.

If your speech is recognized, a validation sound plays, the input field is completed automatically, and the query is submitted. Once Genudi's response is loaded, the cycle starts again, allowing a smooth audio conversation to take place, without having to type, or click.

If your speech is not recognized, a short error sound plays, and one second later, a short beep indicates that you may try to speak again.

To interrupt the speech recognition, say "Stop"

Email

Thank you for participating in the Genudi experiment.
Email me questions, suggestions, conversations, deceptions, or anything relevant at [email protected]

Screencast

This 1/2 hour screencast is a live demonstration of the above commandments: