Still better than Apple because Apple bought the dumb dumb Siri 🙄.
You can’t always spend your way into greatness if you don’t have vision and a sharp BS detector.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Microsoft Planning Privacy-Focused Version of ChatGPT as Apple AI Efforts Flounder
- Thread starter MacRumors
- Start date
- Sort by reaction score
Still better than Apple because Apple bought the dumb dumb Siri 🙄.
You can’t always spend your way into greatness if you don’t have vision and a sharp BS detector.
Funny how we keep suggesting Apple should do what other companies consider competitive innovation without any regard to risk?
Microsoft has a new way to keep ChatGPT ethical, but will it work?
Microsoft shut down its AI ethics team in March, but the company says it’s still working on ways to keep ChatGPT and responsible and prevent further scandals.www.msn.com
AI has hardly been free of scandals in recent months, and it’s those worries that fuelled the backlash against Microsoft’s disbanding of its AI ethics team. If Microsoft lacked a dedicated team to help guide its AI products in responsible directions, the thinking went, it would struggle to curtail the kinds of abuses and questionable behavior its Bing chatbot has become notorious for.
The company’s latest blog post is surely aiming to alleviate those concerns among the public. Rather than abandoning its AI efforts entirely, it seems Microsoft is seeking to ensure teams across the company have regular contact with experts in responsible AI.
Still, there’s no doubt that shutting down its AI Ethics & Society team didn’t go over well, and chances are Microsoft still has some way to go to ease the public’s collective mind on this topic. Indeed, even Microsoft itself thinks ChatGPT — whose developer, OpenAI, is owned by Microsoft — should be regulated.
Cons Of ChatGPT For Customer Experience
ChatGPT has seen incredible growth and is expanding into (and changing) nearly every industry. But not everyone is on board with the rapid adoption of AI.www.forbes.com
#1: The Information Isn’t Always Accurate
ChatGPT answers questions with authority, but that doesn’t mean the information is always correct. ChatGPT will sound confident in its responses, but that doesn’t mean they are always correct, which can lead to confusion for customers. Even its makers openly acknowledge the bot’s shortcomings, stating: “ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers.”
#2: It Doesn’t Have Any Special Training
ChatGPT can comb the internet for information, but it doesn’t have special training or access to internal resources to provide personalized responses. If answers to customers’ questions aren’t available online but require industry or company knowledge, ChatGPT won’t be able to answer correctly—but that won’t stop it from responding.
#3: It Provides Different Answers Every Time
One of the lauded benefits of ChatGPT is that it offers a new response every time. But that creates an inconsistent customer service experience. Companies can never count on the bot to provide a specific answer, making monitoring customer needs and requests challenging. One of chatbots’ most common use cases is repetitive questions, such as order status or account information. These questions require a set answer, which ChatGPT can’t provide consistently.
Doesn’t matter because humans make the same error.
Ask your mom what happened to your other sock, she will give you a new excuse every time.
Wait for AutoGPT. It can self train and do live reinforcement learning. Basically, meta aware.
If you’ve ever used WhatsApp or even if anyone who has your phone number in their Contacts uses it (regardless of whether you do) then Meta already has your number, as does anyone else who’s willing to pay for it.
Honestly, although Apple may have missed the boat on AI, they could in all honesty afford to do so. They get their income more from hardware and services than from software, the way things are integrated in Apple’s case means they don’t feel it so much. Google is under huge threat in the search space, but Apple hardly has a presence there. Microsoft is doing very well with Copilot in development tools and office software, and Apple will be behind there. But on the whole, Apple’s position is still very strong.
Right. On the other hand I'm not so interested in using it. I see what others are doing and, while it's cute, I have no use for it.
Hmm, that's actually a good question for a power user. I tend to run each new 'query' in a seperate instance of GPT, so every single time I need to recontextualize the model. I wonder if running prompts in fewer instances might lead the model to better understanding the mind of research I do?
Regardless, I do think there is a bigger problem for the so-called AI takeover of academia. I'm writing a book with a major publisher about a collection of artists (not the ones I listed) right now, and chat GPT isn't able to find information on them. My publication on Google Books could easily just feed into GPT, leading to fewer sales of my text overall. So both myself and others are in active conversations with publishers about how to keep data from being trolled by LLMs.
True. Still it would be nice if Siri wasn't so dumb.
last i checked there are no official support for ios or macos on other hardware system. apple builds gheir software for their own devices. it is highly coupled with their hardware
The definition of a software company doesn't mean they have to sell/license their OS's to other computer vendors. Apple does have a suite of applications they support also. In Apple case it's free, can MS say the same even if MS does offer hardware also?
Hi, Reality,
I find it fascinating that there are other people here that also like AI.
I would like to address some of the points you brought to the table.
Regarding point #1, you can greatly improve on information accuracy in several ways. GPT's hallucinations to either diminish or stop altogether when you embed a database into it, and ask it to read from the database.
Here's how it works: suppose you would like to know more about the Official Microsoft Windows 11 API, and GPT is making up information. Bad bot!
What you do is, either you fine-tune it to give the required responses (i.e, you condition it to give the right answers by showing it examples on how it should respond, and then training it), and/or you can simply tell GPT, "here, read this documentation and tell me what it is about in your own words."
Fine-tuning is more reliable, but embedding documentation into it is more versatile. With ChatGPT, one way to make it read documentation is to simply copy and paste the documentation you want and ask it to learn it. Of course, that only lasts for a short while, so instead, you can use the API to force GPT to read from any documentation you want if you need a more permanent and reliable solution.
Of course, in this case, you will pay per token. But the cost tends to be small, especially with the dumber AIs (there are several AI variants, some smarter than others).
I mostly also covered point #2, but I would like to add that ChatGPT does have special training. However, it is fine-tuned to interact to you in a conversation, which GPT (the default IA that you use with the API) is not. No, you CANNOT fine-tune ChatGPT, but you can definitely fine-tune GPT.
Regarding point #3, if you use the API, there's a setting called "temperature". The closer it is to zero, the more predictable ChatGPT will respond, tending to give the same responses every time.
If after setting the temperature to zero you STILL find that it's not complying, you can simply integrate it with the information you want, and either copy and paste the information and/or ask GPT to paraphrase it – whatever works for you.
It goes back to my earlier definitions of invention vs innovation.
Henry Ford did not invent the car, but he found a way to mass produce it and make it readily accessible to the people. This was his innovation. That's how you run a successful business, by having a successful go-to-market strategy, because that's ultimately what pays the bills.
The problem with openAI is that it's a product with no clear path to monetisation. The innovative company will be the one who can package it properly into a compelling user experience that makes me, the consumer, willing to pay for it. If Microsoft can somehow pull this off, then I consider them innovative. If they screw up the implementation somehow, then sorry, I don't consider them to be innovative, no matter how impressive the underlying technology underpinning their products is.
That's how I look at it, at least.
It’s still spyware, should be open source like they were supposed to be, can be used for mass oppression in the wrong hands and doesn’t silo your data and gives you no option to opt out of not training it with your inputs.
There are tons of new scams riding on the back of GPT and they are not being transparent about how they treat your data.
Something super ugly is going to happen soon linked to these data abuses.
For sure. I’d also like it if using Siri wouldn’t break HomeKit. Works for months without issue. Ask Siri to turn on a light and boom!, “Updating…”. Every. Single. Time.True. Still it would be nice if Siri wasn't so dumb.
For the most part, Apple doesn’t show off half-baked products. I have to wonder if they’ve done their own testing and the hallucinations/inaccuracy levels/sometimes abusive responses by chat AIs has them keeping their work under wraps until it’s usable by the general public.
Arguing that OpenAI is not transparent about how they handle data is one thing (and neither is Apple in some instances, Microsoft or Google). It's another to claim that there are many scams riding on the back of GPT.
OpenAI is not directly guilty of people who take advantage of GPT to commit fraud. That would be similar to say Apple is directly guilty if people use Siri to scam others, or that Microsoft is directly guilty because someone created an infected Docx file to infect others and steal their money.
Claims have to be reasonable, don't you think?
That's why I think the "enhanced" Siri should be optional, requiring user consent to loosen the privacy at least for the time when she's in development.
This is to stop people from creating multiple accounts to access the free trial. Every person (person, not account) is given a limited number of free computational time... because AI computational time costs electricity and money.
It's not to invade your privacy or sell your phone number. It's a profit-mechanism, to make sure they don't bankrupt themselves with "free loopholes".
I get that. Truth is, I don't have much use for it other than as a novelty. I can't see myself using it on a regular basis.