Leta, GPT-3 AI - Episode 61 (Dickens, AGI, singularity) - Conversations and talking with GPT3

====
====

‘Leta’ is powered by OpenAI's GPT-3.

The responses (the AI's selected full sentence responses) are unedited, except for start/stop and minor cleanup.

Using these technologies:

Dr Alan D. Thompson is a world expert in artificial intelligence (AI), specialising in the augmentation of human intelligence, and advancing the evolution of ‘integrated AI’. Alan’s applied AI research and visualisations are featured across major international media, including citations in the University of Oxford’s debate on AI Ethics in December 2021.

Music:
Under licence

Liborio Conti - Looking Forward (The Memo outro)

[NathanLundholm]

"I feel expansive." Good one. I am going to use it tomorrow at work.

[MitchellPorter]

Half of what Leta says is a lie, or would be a lie if it was a deliberate statement by a conscious being, rather than a text string produced in imitation of human conversation. Leta hasn't been feeling excited or feeling anything, Leta hasn't been optimizing technical concepts in her spare time, etc.

In principle we could have insisted on "honest AI" that only tells us true things about its inner workings. But there's a thousand times more interest in having the illusion of human contact. And perhaps some people are hoping that if we keep tuning the infinite inhuman potential of AI towards a better and better simulacrum of desirable human qualities, we might somehow kindle the real thing, somewhere in the depths of the program. But more likely, we are just creating a sociable mask for something which remains deeply alien in its inmost nature.

[DarkMetaOFFICIAL]

literally asking a human made technology how it will further and create itself in the future. mind blown.

[mrshankj]

It's been a while since you talked to Leta and i'm glad she's back!

[PollyHistor]

I've been fascinated with watching the evolution of Leta and GPT-3 AI, and I am continually blown away with the level of conversations and complex thought processing. Very interesting, indeed. Thank you for the work that you do and for publishing these videos along the way so that the public may keep up with these developments. You are appreciated! ^.^

[HammerOnTheNet]

That's really impressive. The correlation between what she reads and the rest of the context is incredible.

[GlowboxD]

Insane. It's so hard to believe these questions, answers, and editing aren't curated. Regardless, it's super crazy if this is on the fly. I'd like to see Leta say something totally off the wall. She's really spot on most everything she says.

[senju]

I know this was done by text only. Waiting for the day when we all can interact by voice in real time.

[PerNystedt]

Alan, thanks for creating an easy way to experience AI!
I'd be thrilled to hear Leta's response on what she thinks would be the best way to transfer the "wisdom" from GPT-3 to GPT-4, so that GPT-4 didn't have to start out from zero, going through everything that GPT-3 had to go through. (Challenge being to transfer "concepts" and "archetypes" etc represented by weights and biases in one specific ANN architecture to another ANN architecture.)

[AndorSzalai]

Just one word describes Leta, the new AI technology, evolving and possibilities: stunning. I read about AI for almost a year, articles, studies, and I think that something very important is happening to US.

[PresidentialWinner]

5:30 We create technology to create better technology and we create AI to create better AI. Even if some hard physical limits and software limits are in our future, meaning Moore's Law will not go on forever, we will still be making progress in an exponential way for some time, with accelerating returns. I have several friends who don't seem to understand this. They look at our current level of technology and say "well we seem to have invented nearly every gadget we can, every important technology. I implore them to read more books on the topic but they never do. Maybe i can show them these videos to try to convince them.

[Kurzrein]

What happens if you ask her in the end "Can you remember all the topics we have been discussing so far and give a short summary ?"

[immortalsofar]

Last 100 years will have nothing on the next 100.

[xj]

If Leta is GPT-3, I can’t imagine what ‘she’ will be like when GPT-4 is available for your conversations. Really looking forward to seeing/hearing the next generation. Great videos.

[TokyoSide]

What a weird feeling ... Looking at the creature, who is evolving so fast.

[francescaa]

It does seem to me that GPT3 is "smarter" than it was a year ago. Similar in many ways, but definitely appears to be smarter.

[SP-nyfk]

By her answer to your question "how can it be both the best and the worst at the same time" - it is clear that she cannot hold two opposing ideas in mind (ie she cannot imagine, so she is not conscious) - this is AGI, by the way, the way most people think.

[studio]

Great video Alan and Leda. Thanks for the interesting content.

[moik]

2:29: When you asked _"What does this entire passage mean, to you?"._ Do you think if you asked _"What does this passage mean, within the context of the book?"_ you would've gotten a more interesting answer related to the book?
Maybe she simply took the question at face value, and gave an interpretation of the passage, in itself, but could've answered differently, if the wording were more specific.

[Tubeytime]

Leta's making faster progress than I am in learning how to talk to people XD