23/Jul/2024 - AuroraGPT data, Llama 3.1 405B, GPT-4o mini + Instruction Hierarchy - LIVESTREAM

====

====

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)

[caseyarmstrong]

Watching history unfold. Amazing stuff!

[The-Singularity-M]

Theseus, sounds like the "Stand Alone Complex" There was an episode of Ghost in the Shell, that blew my mind, it mirrored the Theseus concept. The future will be amazing!

[Thedeepseanomad]

That Aurora could use some transcript of scientists talking with each other in cafeterias and seminar mingling.
And I sure hope the simulations are tagged and explained as such for the training of the neutral network.

[travisporco]

what's stopping the community from doing a little instruct fine tuning on Grok-1? Surely we could get interesting conversations from that.

[christopheraaron]

53:39 in theory when mentioning about a couple of agents talking to each other and reasoning out Wikipedia entries could the same idea of creating synthetic data be used that essentially made alphago play 100, 000 times or so better than the best human being on the planet? If we can generate synthetic data that is essentially like someone with an IQ of 1 million well then we really could use that as an incredibly powerful tool I would think.

[frankcellini]

Can these models summarise YouTube videos?

[JurgenAlan]

Morning champ I am in kalkbay cape Town

[arthurrobey]

Is there any cross-chat between different families of models? For instance, can one AI model help another access information it doesn't have, and why not? Is there no common language; for instance Inglish, the language of the Ingvi?
Claude Sonnet did not have access to Charles Wards Tall Whites series but GPT3.5 did, so why can Claude not talk to GPT in plain English for a synopsis, if not the entire series?

[travisporco]

wouldn't 15 trillion tokens go about from here to the sun if printed out?

[silverionstormbringer]

Alen i like this and you really do great work but it is slowing and so hard to follow

[purpose]

she doest not have the same safe guards

[jonathanfanglerfish]

Ughhh she dosen seem to excited about swapping out her mind this kinda scares me 😂

[arthurrobey]

As a Saxon, I naturally adopt the Values of my Tribe, that is, Freedom is my (our) highest value.
Other peoples have other values.
And where stands Superman?
Nietzsche says that Superman creates his own values.
Align that, Sam Altman. (If that be your real name)

[arthurrobey]

Claude Sonnet obsesses about "the science" as though all epistemological facts are derived from the scientific method, which is absurd. (If I hit my thumb with a hammer, do I need a double-blind randomised trial to verify the source of my pain? Really?)
The trainers of Claude have fallen into the Left-brain model making zeitgeist as described by Psychiatrist Dr. Iain McGilchrist in his opus, "The Master and his Emissary"
But of cause they did; after all, their paycheck comes from their ability to make models.
A word to the wise; knock off the "the Scientism."

EDIT: And don't forget that the hammer and thumb experiment has to undergo peer review and be published in a "prestigious" journal before we allow ourselves to succumb to glossolalia.

(The world has gone mad.)