OpenAI’s ChatGPT: This is Science Fiction!

The only channel who's video notifs I auto-click :D

It's not human-level AI until they start having useless meetings and gossip about each other behind their backs.

did we just create a 16 bit matrix 🤣

next paper: a recursive simulacrum of an AI research lab

AI : "I am not coming for your job, I am coming for your company"

If you told me 10 years ago that making really advanced next word predictors roleplay in a video game would be a legitimate research project in the medical field I don't know what I would have thought

This makes me think of how the Bumble CEO announced recently that they plan on adding a feature in the near future where every single user's profile will be made into an AI persona that interacts with every other user's AI persona, in order to determine the perfect match with a high degree of accuracy - such that the user does not need to waste time interacting with other users (I find it funny that there was literally a Black Mirror episode about exactly this). I can definitely see this being used to run simulations where a human would otherwise be required, and to build products through a recursive, continuous improvement manner.

The fact that Chat GPT makes better/more accurate answers when you have it simulate individual actors inside a larger organization is really fueling my we're all just a brain in a jar paranoia.

> Two minute papers > * 6 minutes * I'm starting a class-action lawsuit

Please someone remake The Office with ChatGPT characters. I want to see AI Jim annoy the hell out of AI Dwight

Chat GPT Sims game when?

Mulling it over, I feel like what this really boils down to is a novel training environment for creating a Mixture of Experts (MoE). Which is really cool.

The craziest part is that with any new model iteration all of these techniques will immediately become even more powerful in all aspects. The individual agent is better, the multi agent system is better and the learning is better.

I am an LLM, aren't I?

Imagine if the video game development simulation would actually end up producing something akin to Gameboy games without human supervision, then takes feedback from human play testers and the AI developer improves it. It's like all of the retro gaming era could come out of this all over again. Also, the AI developer has no social issues within and everyone is friendly and productive.

I would love something like this running on a local LLM... imagine this with a nice UI, character creator, custom sprites, and such... And you can just give them their own lil world.

No, it has to be a prison simulator, lets do all those psychology experiments.

1:53 you may end up in real hospital after seeing the openai api cost 😂

This even works in 3.5, just by dividing up tasks and assigning different agents for different tasks will give you almost human like accuracy in a very short time. I'm surprised a research paper on this came up so late because this has been the standard practice in most companies for the last year or so

Diagnostic AI assistants trained on pharmacology and physiology data with access to a clinic's patient visits, doctor's notes, and client's EHR could provide physicians with way better QA control in some cases, like during a day when a doctor's mind, body, or emotions might otherwise impair their decision making. For professionals stranded in 100hr/wk work conditions with increasingly smaller teams of aging out nurses, or even assistants in other industries/occupations, this could be a scalable, low-cost SaaS tool and improve quality of care across the board. Wait for VC to jump all over AI during the next low-rate bull run. We'll be seeing improvements everywhere in the next 2 years.