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.
ChatGPT Mac App Stored User Chats in Plain Text Prior to Latest Update
- Thread starter MacRumors
- Start date
- Sort by reaction score
No it shouldn't be "view points" it should be facts and truth. Undeniable evidence and science. Not opinions, hearsay and what some loon thinks.
I wonder what they store in plain text on their own servers.
Is it a good idea that Apple is partnering with this thing?
Is it a good idea that Apple is partnering with this thing?
Last edited:
No. Every viewpoint is not worth representing. Effort should be put into making sure what is represented is factual and true. We don't need LLM to express the viewpoint that the world is flat. Free speech allows for a variety of viewpoints, LLMs should strive to represent factual information and avoid perpetuating demonstrably false claims.
I agree. I do have my discomforts around generative AI. They steal people’s work. I was always told I should be ashamed of myself for having anything at all to do with 40-year-old ROM files because I didn’t have the author’s permission to use their content that way. Now that remains an evil, rotten thing to do while it’s okay for OpenAI to steal the contents of my unpublished novels and make money with them?
That said, I think a lot of what’s going on at this point is plain paranoia. So they store your conversations in plain text? Good. That means I can back them up if I want and move them to a new computer. Firestorm Viewer for Second Life also stores all my conversation logs in plain text. I’ve known this for years because I choose the log file location in iCloud so I can sync it between computers, and because I regularly open up the log files in TextEdit to check if I remember things correctly. I haven’t seen it as any kind of huge, alarming scandal that it works this way. And if you know about Second Life, you know those conversations can be a hell of a lot more private than anything one would ask of ChatGPT.
There’s been talk of the threat of bias with generative AI chatbots in this thread. Sure. But how is that any different or more serious than the threat of bias with search engine results - an issue nobody pays attention to anymore?
Can generative AI be used to generate child porn or falsely show politicians doing things to spread misinformation? Sure. So can photoshop, audio/video editing tools, or even just finding the right image or camera footage and applying false context to it, that have been actively used to spread lies for decades. As with AI, anyone can do these things, but it takes skill to make it convincing and hold up to scrutiny.
Can generative AI teach you how to make bombs or otherwise tell you dangerous information? Can it cause you to die in horrible ways with factually inaccurate health information? Absolutely. So can human-written pages you find with a search engine. This is not a new problem with the Internet.
Generative AI has all these ridiculous safeguards in place around politicians and sex to try and placate all the fear-mongering about it. I can’t even look up info on the upcoming election here in the USA. I don’t have such safeguards preventing me from doing whatever I want with Affinity Photo. I don’t have such safeguards to stop me from writing whatever I want into Microsoft Word. My guess is once this technology stops being new, Generative AI safeguards will go away and people will look back at them in the same way we look back at laws banning Rock & Roll.
Last edited:
While I understand the need for encryption, this is not a place where i would care if the files are encrypted on my device. I don't care if the documents I create in Microsoft office are encrypted by Microsoft office. My OS already encrypts the files on my system. My on-device chat logs with an LLM not being encrypted is not sensitive information.
I currently run local LLMs, and all texts are in plain txt files that I can export out. The problem would be if the files on their server were not encrypted.
I currently run local LLMs, and all texts are in plain txt files that I can export out. The problem would be if the files on their server were not encrypted.
This is not great but it misunderstands what App Sandboxing does.
App Sandboxing does not prevent malware from accessing app data. It silos apps from accessing data outside their own sandbox.
App Sandboxing does not prevent malware from accessing app data. It silos apps from accessing data outside their own sandbox.
Not the point. Many people naively assume AI like ChatGPT is not biased.
AI like ChatGPT is modeled after what some loons think. It is clearly biased and the people building it influence that bias heavily.
Unfortunately, AI is nowhere near only providing what is factual and true. It is heavily biased based on its creators opinions on many, many topics. Google Gemini couldn't even portray history correctly due to bias.
AI just making the world so much better, haha. Just a day earlier you had this also. Companies just lack smarts to realize AI can cause unforeseen problems faster than any other tech gimmick.
=======
Figma CEO Dylan Field says the company will temporarily disable its “Make Design” AI feature that was said to be ripping off the designs of Apple’s own Weather app.
Figma pulls AI tool after criticism that it ripped off Apple’s design
The company pulled the rollout of the beta feature.
www.theverge.com
Figma’s new tool Make Designs lets users quickly mock up apps using generative AI. Now, it’s been pulled after the tool drafted designs that looked strikingly similar to Apple’s iOS weather app. Figma CEO Dylan Field posted a thread on X early Tuesday morning detailing the removal, putting the blame on himself for pushing the team to meet a deadline, and defending the company’s approach to developing its AI tools.
All the Americans bashing the EU for caring about customers and providing choice are somehow strange. Aren't you guys all for freedom? LOL!
Unfortunately it is - but it will always repeat what it's learnt the most. Which is why for instance it can learn thousands and thousands of coding examples and the ones that are right will outweigh the ones that are wrong and that's how it gets it right (more times than not).
It's fed every scientific paper ever created so it's going to come out with facts over absolute nonsense, so it's already more useful than 50% of the people on the internet.
Nobody thinks that. Everyone is biased to some degree and the things we create are informed by those biases hopefully in a direction that is positive and well informed.
That is not how that works. Yes, people building it influence what ultimately gets released to the public, but LLMS are trained on petabytes of data from crawling websites, Wikipedia, books, code data sets, QA with factual information and multimodal models include audio and videos. All those sources of information inform the LLMs bias. The more information you have that leans a certain way the more the bias leans that way. But ultimately you don't want a commercial LLM that is racist and provides sexually explicit and graphic as well as harmful information, so safeguards are put in place to prevent the models from providing those types of information.
It is not heavily biased by its creators' opinions that is impossible. They can fine tune it not to provide sexually explicit information, information that is dangerous, racist/prejudicial. Much of the data used are very western centric so again effort is put to making sure that data used represents various geographical regions.
One of the things I enjoy about running smaller models locally is that they can be easily finetuned to remove some of the security measures used to filter its response, but their biases picked up from training data still remains.
Why is this a problem?
Seriously, somebody want to explain it to me? Plain text locally on my computer seems like the RIGHT way to store data. If you want it encrypted, turn on FileVault.
Breaking news: Pages doesn't encrypt your word processing documents. TextEdit doesn't encrypt your quick notes. Numbers doesn't encrypt your spreadsheets. Messages doesn't encrypt the local copy of your conversations.
What a silly article.
Seriously, somebody want to explain it to me? Plain text locally on my computer seems like the RIGHT way to store data. If you want it encrypted, turn on FileVault.
Breaking news: Pages doesn't encrypt your word processing documents. TextEdit doesn't encrypt your quick notes. Numbers doesn't encrypt your spreadsheets. Messages doesn't encrypt the local copy of your conversations.
What a silly article.
Fear mongering is what it is.
You misspelled apple. It was apple that approved the app for the mac store.
What's good for the goose, is good for the gander. Have you forgotten it is apple spreading this fear mongering by always talking about privacy.
Yes, but still have to secure local building permits when modifying the structure, build according to code so you don't burn down your neighbor's house, etc. It's not the Wild West. Your car still has to pass inspections to be allowed on the road.
We are for freedom, especially freedom from government overreach. When we buy Apple products, and pay for Apple services, and develop for Apple platforms, we expect to play on Apple’s terms. If we don’t like that, Android caters to that mindset.
We don’t understand and do not support Big Government forcing companies to support their competitors. That’s the opposite of free market capitalism that is the foundation for the American way of life. Consumer driven demand determines if there is an apetite for consumer choice, and Android fills that role without needing to force Apple’s hand.
America hates the EU concept that Government knows what’s best for you, and Government will protect you by force.
Ignoring the human emotional side is just denying reality. Emotions factor into all decisions and discussions, whether you want to admit it or not. There are no exceptions...not a single one.
We are not computers and treating discussions like we are is a recipe for failure.
speaking of good ol MS note their boss feels like AI can use anything on the web without limits.
Microsoft’s AI boss thinks it’s perfectly okay to steal content if it’s on the open web
That is not how fair use works.
www.theverge.com