The AI dooooooom thread

[Jacob]

All you people thinking AI is just a speculative waste of resources, I wish I shared your optimism.  Speculative bubbles are painful when they burst, but humanity has survived them plenty of times before.  I'm not sure humanity is ready to survive the AI that delivers on its promises.

Agreed, unfortunately.

[Josquius]

Could be. To an extent I think it's there already. Purposefully bad AI pictures used in an ironic way.

For AIs long term potential - long term AI is of course the end of the world as we know it. Absolutely society breaking.
If it is possible.

LLMs on the other hand are not that. They are a fad bubble which will pop. They won't go away completely. Look at amazon and others with the.com bubble. But the consensus increasingly seems to be   they're a bit of a dead end for AGI.

[Tamas]

There is a Microsoft Teams AI plugin called Facilitator (and Copilot can summarise for you from the transcript obvs). It's actually pretty good when the discussion on the call is easy to summarise and turned  into a list of actions.

I missed a call this morning I wanted to know where things are and how they are getting unfucked, and I went to read the Facilitator summary, which was exactly what I needed, confirmed with my direct report on the call if it was legit what was decided, and that was it.

Certainly has its uses. And people who think project managers just glorified note-takers and meeting organisers may very well conclude they no longer need PMs.

[Syt]

Note taking is definitely a big plus - but as CC pointed out above it probably depends on the topics discussed. Nuanced legal discussion may be more difficult to do accurately compared to our project progress updates (and even then one of us usually goes through it and makes edits before sending them to the other participants ... still a lot better than someone having to dedicate brain power constantly to taking notes).

[DGuller]

In my previous job, I had ChatGPT Enterprise edition, with access to some heavy duty reasoning models.  I used it to develop and document a custom algorithm, and it handled both extremely well.  Unfortunately in my current job all I have is Copilot, which frankly feels like three steps back.  Forget helping with developing a novel algorithm, sometimes just remembering what I was talking about in the previous prompt is a challenge.

[Syt]

We have ChatGPT Enterprise at work and I find its 5.1 model quite useful.