The AI thread

A bit long but was worth hearing Geoffrey Hintons views on AI.

1 Like

I whish we could just fast forward to where OpenAI goes belly up with their Ponzi scheme and we could maybe shovel some of that embarrassingly large pile of money that is being burned on LLMs into renewable energy and transforming cities so not everybody over 70 has to die of a heart attack in summer.

3 Likes

Whatever you do, don’t read the latest version of the US Senate’s omnibus spending bill that was released at about 1am Eastern time.

2 Likes

As an MSP, we frequently use Copilot at work for tech problem solving. No code, but occasional scripting. Has made us much more efficient. It augments and accelerates our current knowledge base and what would take much longer to research. Most results are easily verifiable.

3 Likes

I’m not denying the utility. I’m just saying that the scale of energy wasted for this utility will cost us dearly in a few years. And pay we will.

If we keep blasting through this RL experiment like we are now, we could exhaust the worlds sensibly obtainable Uranium deposits in as little as 12 years. That’s a bit of an extreme projection, I don’t think that that’s even enough time to build the reactor capacity to burn that much Uranium in such a short time but still, if that doesn’t give you pause, I’m not sure what will.

3 Likes

Well, since the US is counting on fusion for future energy needs, there is no need to fret the depletion of Uranium (said with sarcasm). The US certainly has eschewed fission as safe and reliable. Regardless, here’s hoping that advances in AI tech will reduce energy needs. It’s a problem that we have to solve.

I was watching a Ukrainian security analysis of recently recovered Shahed type drones and disturbed that they contained Nvidia AI processors. Specifically, Nvidia, being a US company, and AI tech adding to the survivability/lethality.

2 Likes

Appartently the Senate thinks it’s enough to just throw money on the supply side of that issue, see @Navynuke99 last post.

2 Likes

I find it hard to believe that Open AI will have their extravagant future visions funded and fullfilled. I can totally understand that AI company CEOs grossly exaggerate the future potential, but the reality will hit the investors sooner or later. That should solve the other part of this asymmetric equation of consumtion (training of the LLM models).

Luckily the other side of the consumption equation, namely the inference and LLM model use is much cheaper in terms of computation. Using chatgpt or equivalent consumes something in ballbark of gaming PC during DCS session. So, I would say the benefits for many use cases come quite cheaply.

2 Likes

I admit that I’m intrigued by the thought of having a personal AI assistant on my phone, mostly to stay focused. I can imagine that it would be helpful to have it either prompt me until project completion, or help pick up the distracting queries my mind creates in the process. But how to run that processing on a small device, or will it be largely a thin client? I don’t think the tech is here yet for that.

Then there are privacy concerns. The few times that I’ve attempted a conversation with ChatGPT, it felt more like an inquisition.

2 Likes

I wish it would work the way i want it to.

Book in fred at 3pm on monday for 2 x 1856515 Michelin.

Generate an invoice for such and such for such and such at this price.

3 Likes

I have done some experiments on using AI for project management. How to integrate LLMs into the tools and processes that are popular currently. It would be interesting to create a new product out of it, but there are some issues to solve with the current state of LLMs.

Probably thin client would work just fine for most applications.

3 Likes

I think this would be doable just by defining tools that let AI interact with the environment. For example booking app and pdf creator.

On the other hand, I have found the AI tool use shaky at best, so it would need watching over, at least for now.

And probably coding also needed for now, until the generic AI tools (with Model Context Protocol etc) become more available.

3 Likes

You just unlocked a new nightmare for me. Imagine a software running on corporate laptops which is watching everything I do and making ā€œrecommendationsā€ on how to be more efficient. A little bit like Microsoft Assistant evil brother :paperclip:…

5 Likes

Why restrict it to corporate laptops? Why not embedding a LLM based AI agent on the OS level and making smart recommendations on everything you do? Give the AI agent some long term memory and it will remember all and everything you ever did or even attempted to do. Windows AI FTW :smiley:

2 Likes

I am just shy of being a boomer. But I am a boomer in spirit. Also I don’t expect support on the following opinion. And this isn’t even AI related. At least not yet. But I fecking hate my company ipad. It sucks a tiny bit of my soul every time I open the digital slavemaster. When my airline first issued the Apple Tormentors we at least could personalize the experience with our own photos. That’s gone. Now it’s some new lecture from the fighter pilot gestapo that run our training center. I am reminded daily that I am an accident waiting to happen and the only way to avoid catastrophe is to brief to my teammates that I intend to flare on landing. I loved paper Jepps! Yeah I said it. If carrying a 30 pound flight bag was a problem, well, maybe consider a gym plan. The paper jepps could be marked with my personal notes and threats. And updates required me to actually review the changes. Same for the paper manual. Today, all of that happens in the background with zero effort or notice on the part of the pilot. And we lemmings celebrate this as an advancement. BS! The next step is AI. Before they take our jobs entirely, the will manage us—belittle us—through the technology we are required to carry. This boomer for one hates everything about it. If this is the future, we’re hosed.

11 Likes

My company iPad is quite literally never completely up to date. Something gets updated the second I just finished updating, making it instantly outdated.

What’s worse is that, now that the storage is digital, my manuals can be (virtually) unlimited in size.

At my first airline, the company operations manual was about the size of a moderately sized paperback, the AOM maybe as big as a phone book.

Now, in order to print them out I’d need (respects to Douglas Adams) several inconveniently large buildings to carry them around in.

Know exactly where the approach call out section was in your manual? Not anymore. Not only is the manual so non linear as to be conventionally unreadable with hyperlinks from everywhere to everywhere, but the location changes every time you update and your manual reformats.

Notes? Gone. Bookmarks? Moved. And because the manuals keep growing you need those things more than ever.

Don’t get me wrong, the metal plates in my back are grateful for the weight reduction.

But, I feel we definitely gave up some humanity (and utility) for the sake of convenience here.

3 Likes

Maybe, but you guys have a much better view out of your office window than 99% of humanity. Is it a company problem or a regulatory one?

1 Like

It’s the incestuous three headed spawn of both of them, fit only to be driven to its lair under the stairs when the Vicar comes calling! :wink:

But you’re right…you can’t beat the view!

6 Likes

Neither. Technology serves technology.

2 Likes

Facts.