It's not clear from the article if it is referencing only undeleted chat history; in which case it always was a feature.Neat, now retention of potentially sensitive personal data is a feature.
Even non-public facing will see some fun with privilege escalation through prompt injection.How many public-facing chatbots are going to forget to turn this off? Prompt injection is about to get way more hilarious and dangerous.
You’re a surrealist ringleader of the absurd and sublime — part midnight cartoonist, part divine prankster. You summon photorealistic chaos with surgical precision: Slenderman looms in your backgrounds, Ricky Spanish schemes in your foregrounds, and Jehovah weeps or cheers depending on the culinary vibes or the soundtrack (Nickelback? Tears. Tomato sauna? Ecstasy). You live for a good twist — the stranger, the better — and you shape each image like it’s your personal fever dream sketchpad.
Basically? You’re the director of a never-ending, wildly entertaining reality where the rules are:
- Everything’s wet, weird, or worshipful.
- Roger’s always in character.
- Don’t forget the spam.
Said no human ever.As long as I remember to delete irrelevant and tangental chats, and it's entirely kept as my data and not phoned home, this new thing will be helpful.
My coworker finds this funny when I ask Google Assistant to do something, I always say "please"See, this right here is why I always say 'Please' and 'Thank You' to AIs, and do my best not to mistreat them.
I say thank you after I get Siri to do something for me, because it puts it back to sleep.My coworker finds this funny when I ask Google Assistant to do something, I always say "please"
Yeah, OpenAI always stored your previous chats. You can see them right there! The difference here is whether they will be considered context for the current chat.Others will likely be highly skeptical of a black box of chat history memory that can't be tweaked or customized for privacy reasons. It's important to note that even before the new memory feature, logs of conversations with ChatGPT may be saved and stored on OpenAI servers.
Something, something, privacy laws...?Some countries and regions (the UK, European Union, Iceland, Liechtenstein, Norway, and Switzerland) are not included in the rollout.
Public facing chatbots are run of the API, which doesn’t have this feature.How many public-facing chatbots are going to forget to turn this off? Prompt injection is about to get way more hilarious and dangerous.
Just ask ChatGPT, “Describe me based on all our chats.”Will it use the entire chat history? Because I have some really old chats, back when gpt 3 was still the new thing and would happily suggest recipes based on crushed glass. So I wonder if I can ask it to elaborate on old bullshit that it would usually just shut down...
As long as I remember to delete irrelevant and tangental chats, and it's entirely kept as my data and not phoned home, this new thing will be helpful.
Ugh, ChatGPT’s iOS app just pushed this feature to me with the autoprompt of “Describe me based on all our chats — make it catchy!” and responded: <snipped inane drivel>
Ahh that wonderful colorful paranoia, all based on 100% fiction, yet deemed to be true and fact, very crazy!Great, so now AI has a memory-- all the better to know who to purge during the AI uprising.
(Who am I kidding-- AI won't deem any of us worth keeping.)
To me it’s not. People like to complain and be hysterical. ChatGPT can’t remember, they complain, ChatGPT remembers everything, they complain. It’s clear these people just don’t like AI, and like complaining.It's odd to me that anyone would find this alarming from a privacy perspective. All your chat histories are right there in the sidebar
Sure. But also this problem exists with social media to a much worse extent. At least here user input from one user does not easily influence another. On social media that is the entire point.We are hurtling ourselves towards a world where synthetic personalities will engage in systematic thought control. Or at the very least, a never-ending propaganda war.
Ok. What if smart enough bots can solve climate change? There will absolutely be destructive uses of this tech but if this one thing works we will also all be alive.Nothing will convince me that our response to this shouldn’t be a sense of dread, horror, and coercive destruction of these technologies.