Beep boop - this is a robot. A new show has been posted to TWiT…
What are your thoughts about today’s show? We’d love to hear from you!
I listened to this episode live (I’ll rewatch it VOD another day) and I followed in Andy’s footsteps: I did my taxes during the show. It was good background sound
This year I’m getting the biggest return of my entire life: $10K. I dedicate a portion of that to Andy for giving me the kick in the pants to do my taxes (although I really enjoy doing my taxes. It’s kind of fun)
I loved how Andy joked about Jason talking for 45 minutes giving him time to do his taxes. I hope he was actually doing it and it wasn’t just a joke.
I did my taxes weeks ago. I was more concerned about 3rd party interference with the IRS and wanted to make sure I got my refund.
There’s two sides to every story… right? This also means you’ve overpaid your taxes all year, loaning that money to the government, only to get to wait 6 weeks for them to give it back to you … 
3 Likes
Someme at work said that to me too. But this year my tax returns were huge because I put a loooot of money into my RRSP and I get to deduct taxes because I maintain a home office.
I usually put about 10K into my rrsp but this year I put 26K because I wanna buy a house in the relatively near future so I’m getting very serious about it.
This past year I got serious about having my money work for me and I even have a finance guy investing my money into the markets. I also made a lot more money this year because the gov’t owed us a lot of money in back pay for having an out of date collective agreement.
But you’re not wrong: getting a lot of money back does mean the government borrowed too much money from me during the year usually
I recently ran across an associate who said they avoided the Artificial Intelligences because of the huge electricity expense associated with using them. This was a surprise to me. Have other people heard this? Is avoiding AIs a big thing, or are few thinking about it? #WhatWouldGretaSay
Does any of Apple’s reluctance to rapidly deploy AIs have to do with their energy cost, or is it all about problems with the performance of the services?
Does anyone know how Apple offsets any energy cost with their own in-house computation and with ChatGPT’s energy cost? Do they have alternative energy powering all of their AI services (including ChatGPT)?
One of the reasons I switched from a Ryzen 7 desktop PC to a Mac mini M1 was that it had the same performance for photo editing, but used a fraction of the electricity.
Prices here increased around 60% at the beginning of the Russian invasion of the Ukraine, going up to over 45c/KwH.
The electricity requirements for AI is Jenseits von gut und böse, as they say in Germany.
Isn’t their goal to push as much of it to the end devices as possible?
That is their goal. But I suspect Mr. Bones was referring to the fact that Apple is also building data centers with M-chips when they need to defer to cloud Siri. Although I’m not really sure how any of that works. the M-chips are definitely not as power hungry as a farm full of nvidia GPUs
Eventually yes. I guess Apple’s AI stuff is fine for rewriting paragraphs and e-mail summaries. OTOH, I’ve found Apple AI’s “Hey, Siri” answers to questions is spectacularly useless. I’ve gotten in the habit of just tacking “ask chatgpt” to anything I ask of Apple Intelligence. Actually, I usually avoid the Siri interface completely and go directly to ChatGPT or to Grok.
Since Apple is deferring queries to ChatGPT, I thought that Apple might feel thenselves accountable for the electric power for those queries. The next time I’m in an Apple Store, I will ask the floor staff that question.
@big_D Wow. Average price in the US is 15¢ per KwH. In my area, it’s 14¢ per KwH. I wouldn’t have presumed power cost would be a reason anyone would move to a Mac.
It is one of the reasons why we build houses with good insulation, so we don’t need AC in summer or to heat as much in the winter.
Over the lifetime of the house (100+ years), it pays for itself.
We had temperatures over 42C (107F), but the house interior remained in the mid 20sC (mid 70sF) for most of the period, only the last 2 days got over 30C (86F) inside, with no AC. The house is about 110 years old, so relatively new for small town Germany, unless you move to one of the Siedlungen (housing estate). The town dates from around 800AD, although the first records are a bit later, it celebrated 1000 years towards the end of the 90s, and many of the houses are 17th to 19th century, in the centre of town, the town has spread outwards and newer houses have spring up on estates on the outskirts.
I know a few people who do have AC, but they usually run it for an hour or so before they go to bed, so the bedroom is cooler.
I don’t think I’ll play RollerCoaster Tycon on Apple Arcade; but I loooove OpenRCT2 which I play all the time in Windows & macOS
Maybe I’ll try it on iPad…
EDIT: I’ve been thinking about this all day but I totally want to play RollerCoaster Tycoon on my iPad!!! Although I don’t wanna pay for it so I’m glad it’s coming to Apple Arcade (I’m on family plan where my dad has Apple Arcade)
interest free as well
I avoid using AI mainly for that reason… and also because I don’t find the results very interesting. I’ve used it precisely once, to get a transcript of a 2019 podcast where I needed to trace the context of a throwaway remark made by one of the hosts.
I pay close attention to what other people are doing with it, but don’t see any reason I should contribute to its massive energy consumption in its current state.
2 Likes
Well, that leaves me with a couple of questions. Andy uses and recommends a Google Gemini Flash 2.0 $20/month service in MBW #964. Alex has discussed using ChatGPT voice services constantly when he’s driving and has discussed using Google NotebookLM extensively in previous episodes. AFAICT, Leo uses every AI under the sun. I haven’t heard Jason discuss AIs, but I’m guessing he uses them frequently.
First, don’t you feel left out? I would guess that the majority of users here are at least weekly and probably daily users of AIs. To me, they’re pretty darn useful – once you understand their limitations. There are lots of places where you are probably using AI services; many of those are difficult to avoid or even notice.
Second, do you have any resentment of hosts and the community at large using “wasteful” AI services?
This isn’t a personal question. I’m mostly interested in the juxtaposition of Apple’s aggressive initiatives on climate change with their embracing of energy-expensive AI services. They are tacitly approving of the global activity.
1 Like
I guess I don’t feel left out - I have looked carefully at what other people are doing with AI, and haven’t felt any desire to do the same. I just don’t seem to feel at all inspired by it at present, but that could change at any time of course. Right now I’m content to observe, and think about what organisations might need to do to avoid any potential PR disasters from things like unchecked hallucinations getting into corporate communications.
No resentment about other people’s usage, they are learning how to derive value from it. But having not identified any value to myself, I decided to sit on the sidelines and watch, rather than add to the energy consumption of the big models by experimenting without any real objective. I’m encouraged that the latest disruptors from China seem to be achieving similar results with much lower energy consumption, so that’s progress.
Just waiting to see if any of the results make me go “wow, I need that!”. Could still happen.
2 Likes
Also today there’s been a good example of the importance of being able to check AI output for factual accuracy, per this rather horrifying article:
1 Like
OWASP (Open Web Application Security Project, IIRC) have produced a Top 10 of things to guard against with LLMs, it’s very dense and aligned for corporate usage but still has relevant stuff to guard against:
I’m with Clayton. I have used some different models (OpenAI, Perplexity, Leo, Claude, Midjourney etc.) at the free level, but I have never found anything that I really want to use them for on a regular basis that makes them worth paying for. For the incredible amounts of electricity they consume, they results are wildly inaccurate and often useless.
I was doing an inventory of PCs recently and needed to know what generation processor was in each model (Windows 11 migration plan). Some of the PCs were properly documented, but many weren’t, so I was putting Tag numbers into the manufacturer’s website, but that was slow and painful, so I switched to AI, asking it for a particular model what generation processor it used, the first 4 or 5 answers seemed correct, then it threw out that a 2024 model Dell used a Core generation 1 processor from 2010! That meant I had to go back and double check the other results, because they had seemed plausible, but this last one was wildly wrong… So the time I had saved with the first 5 models was double as I had to go back and do a manual search to double check. I would have been faster doing the manual search for each in the first place.
And if it is getting something so badly wrong about something I can spot straight away is wrong, how am I supposed to trust it in a domain where I am trying to learn something new? It certainly isn’t worth throwing money at these services at the moment.
Resentment, no. But I find the absolute waste of resources pushing what isn’t even an alpha-quality product, let alone beta or a final release product, at the masses is crazy. It should still be in a lab somewhere, where experts can pull it apart and work out why they are constantly making errors (I refuse to use the term “hallucination”, it is an error pure and simple, romanticising it doesn’t change the fact that the answers are wrong and the LLMs are error prone).
1 Like
How do you reconcile that opinion against the experience and recommendations of Andy, Alex, and Leo? Andy’s recommending a $20/month Gemini service in this episode. Alex is clearly using both ChatGPT and NotebookLM – probably $50/month or more in services. Leo is spending more than $100/month. Are you asserting our hosts are not getting real value from those investments? What do they know that you don’t know? It’s OK to be skeptical in our society, but fast-based reasoning must accompany that skepticism. Pseudoskepticism is just another kind of gullibility.
In your prompts to AIs, you can ask them to provide links for you to verify what they find. This is a basic step; some of them do this automatically today. You could even ask the AIs to give you an easy way to verify what they say. With the Wolfram Research Notebook Assistant, it’s easy to have the AI run the Wolfram Language code it generates to make sure that it works before it presents it to use user.