Sentient AI

[frorider]

Welp that was fast, looks like ‘the AI Boom is Over’.

https://www.vox.com/technology/2023/...-less-interest

[BCMtnHound]

Welp that was fast, looks like ‘the AI Boom is Over’.

https://www.vox.com/technology/2023/...-less-interest

“Generative AI can do some amazing things. There’s a reason why Silicon Valley is excited about it and so many people have tried it out. What remains to be seen is whether it can be more than a party trick, which, given its still-prevalent flaws, is probably all it should be for now.”

Pretty much. So much potential, but humans users, and their limitations, are the defining factor. We still haven’t dealt with the sharing the profits of tech advances for labour, forget about the white collar brain fields except for novelty or fringe. For now.

[schuss]

Well, training a domain specific LLM is expensive and relies on curating and training using good, expert info that's reasonably connected and consistent. The knowledge mapping situation in most companies is not great.

[SumJongGuy]

IDK.. It seems to me that AI is getting better at making it impossible to communicate with a human via online or phone support. Although that's human engineered dead end logic loop giving NO option for a problem they don't have a predetermined response for or none of the above no longer being an option..

[frorider]

https://www.politico.com/news/magazi...ze-ai-00111862

‘Nationalize AI’.

[Rideski]

Read most of the thread but not sure if this was posted.

[LegoSkier]

Well, training a domain specific LLM is expensive and relies on curating and training using good, expert info that's reasonably connected and consistent. The knowledge mapping situation in most companies is not great.

Running them is expensive too. I read recently that Githup CoPilot (a generative AI that is a code writing assistant) is costing Microsoft an average of $20 per user per month (and as much as $80 for some power users) but they are only charging $10 per user per month right now.

Like schuss said, these models are only as good as the data available to them and most companies aren't really good at mapping and consolidating their data right now (having a good meta-data landscape). Also, right now companies are hesitant to do too much with these open LLMs because it will expose private data to the outside world and running a private LLM is super expensive. But there are data platforms (Databricks, Microsoft Data Fabric, Snowflake - which I work with every day, ...) that are currently transitioning to allow private use of data using public models on shared hardware. In the near future I will be able to use one of these data platforms to deploy an LLM in a private, secure space that can access all my data, and using a few lines of Python code, deploy a chat bot that "knows" about our sales, orders, business processes, finances, etc. and be able to create analytics or distill knowledge in response to natural language questions from users, which is the holy grail to utilizing this technology

Also, you need to remember that an LLM is only really doing one thing really well which is predicting the next word given a context set by its user. Learning how to set that context correctly (a field called Prompt Engineering) is essentially a new language that current developers, and scientists, and other LLM users need to learn which will take time.

[LegoSkier]

So here is a weird one…Microsoft Autogen is a framework for multi agent chats for problem solving. So you can use it to define teams of autonomous agents based on any number of LLMs and taking on different roles in the team to chat with each other without human involvement.
https://microsoft.github.io/autogen/...etting-Started

Like this one where they have two LLM agents play each other at chess and make small talk https://github.com/microsoft/autogen...at_chess.ipynb

So now we have a way for the AIs to team up with each other and collaborate.

[schuss]

So here is a weird one…Microsoft Autogen is a framework for multi agent chats for problem solving. So you can use it to define teams of autonomous agents based on any number of LLMs and taking on different roles in the team to chat with each other without human involvement.
https://microsoft.github.io/autogen/...etting-Started

Like this one where they have two LLM agents play each other at chess and make small talk https://github.com/microsoft/autogen...at_chess.ipynb

So now we have a way for the AIs to team up with each other and collaborate.

I mean, we already did, it just wasn't easily accessible - https://openai.com/research/openai-five

Also Lego, a lot of what you mention is already possible, it's just the tuning, training and curation costs that are preventing real action on it. Don't worry though, that's coming. Domain-specific AI workers are out there already for insurance and health stuff, with more to follow. Really the challenge on this is it will completely destroy the junior talent pipe and learning process, so we have to figure out new ways to get people expertise to review the LLM stuff, as occasionally they just go batshit crazy.

[chaka]

Interesting development... Sam Altman told to leave OpenAI

[JimmyCarter]

Interesting development... Sam Altman told to leave OpenAI

He seemed like a bit of a shitweasel.

[up an down]

Interesting development... Sam Altman told to leave OpenAI

The board of directors fired him on the advice of their AI consultant

[JimmyCarter]

And team “Fuckloads of Money” looks like it’s going to win out over team “AI Safety” in the end

https://www.wsj.com/tech/openai-tryi...-back-4b728049

[SumJongGuy]

First they rounded up the buffalos and I did not speak out because I was not a buffalo..

[chaka]

https://twitter.com/griswold/status/1726599468022452595

[SumJongGuy]

^^^TRU STORY BRO^^^^^^^^^^^^

[TBS]

Altman is baacckk…
And all but one member of the old board is gone.

“Hey ChatGpt, write a plan to fire the Company’s CEO”

[JimmyCarter]

And just like that, the whole point of the non-profit structure is trashed.

[Joe Friday]

Open the pod bay door Hal…

[irul&ublo]

https://m.youtube.com/watch?v=jQT9QW1En5Q

[neufox47]

And just like that, the whole point of the non-profit structure is trashed.

Can’t say I know much about this, care to elaborate?

[MakersTeleMark]

Can’t say I know much about this, care to elaborate?

It's paywalled, but: https://www.nytimes.com/2023/11/22/p...an-openai.html

[JimmyCarter]

Can’t say I know much about this, care to elaborate?

https://openai.com/our-structure
Original point of the structure was that control would rest with a bunch of financially disinterested people who could make sure the company did the right thing for society vs profit if AGI was moving too quickly/getting dangerous/etc.

Most of the people who closely adhered to that viewpoint (and pushed Altman out for no explicit reason) were the “independent” board members, who have now been replaced.

Now Altman, a “master of persuasion” with no discernible qualifications besides being a VC douche-Svengali, has effectively proven that the “independent” non-profit board is powerless and ineffectual, and he has a path to do pretty much whatever he wants.

[CarlMega]

^no horse in this race but I think the number of employees protesting Altman's dismissal was interesting. It is hella hard to get that sort of employee backing. It'd assume that the rank and file are ultra-vision, mission focused - but maybe I'm wrong? Is your read that they want to monetize to their self-interests vs. the stated tenets of their mission?

I have a hard time reading the situation.

[MakersTeleMark]

I definitely see their compensation being 90% of their support.