OpenAI ChatGPT Is Testing A Memory Feature via @sejournal, @martinibuster

OpenAI announced that it is rolling out a test of ChatGPT memory, a new feature that remembers past conversations to learn a users preferences, remembers styles and tones and format preferences.

ChatGPT gradually becomes better with time as it remembers across all chats and doing away with having to repeat instructions.

The new feature is rolling out as a limited release to some ChatGPT free and Plus users for testing and feedback. There are plans to extend the feature to more users at a future time.

Memories are saved apart from the chat histories, which is how they can be summoned in later chat sessions.

Users Control Memory

The memory feature can be selectively turned on by telling it to remember something and it can also be told to forget specific details or instructions. Memory is controlled through the settings panel, where specific memories can be deleted or disabled entirely. There is also an option to delete all memories entirely.

The settings options are available at Settings > Personalization > Memory.

How To Use ChatGPT Memory

OpenAI shared examples of how to use the feature:

• “You’ve explained that you prefer meeting notes to have headlines, bullets and action items summarized at the bottom. ChatGPT remembers this and recaps meetings this way.
• You’ve told ChatGPT you own a neighborhood coffee shop. When brainstorming messaging for a social post celebrating a new location, ChatGPT knows where to start.
• You mention that you have a toddler and that she loves jellyfish. When you ask ChatGPT to help create her birthday card, it suggests a jellyfish wearing a party hat.
• As a kindergarten teacher with 25 students, you prefer 50-minute lessons with follow-up activities. ChatGPT remembers this when helping you create lesson plans.”

Memory May Be Used For Training

OpenAI advised that content provided to ChatGPT could be used for training, which includes the memories, with an option in Data Controls where this can be turned off. By default, no data is used for training. But this does not apply to ChatGPT Team and Enterprise customers.

Another way to control ChatGPT memory is through temporary chats, which will not invoke the memory feature. Temporary chats are activated in the GPT version control at the top of the chat page.

According to an OpenAI temporary chat explainer:

“With Temporary Chat, you can have a conversation with a blank slate. ChatGPT won’t be aware of previous conversations or access memories. It will still follow your custom instructions if they’re enabled.”

Read the OpenAI announcement:

Memory and new controls for ChatGPT

Featured Image by Shutterstock/rafapress

Why Did Google Gemini “Leak” Chat Data? via @sejournal, @martinibuster

It only took twenty four hours after Google’s Gemini was publicly released for someone to notice that chats were being publicly displayed in Google’s search results. Google quickly responded to what appeared to be a leak. The reason how this happened is quite surprising and not as sinister as it first appears.

@shemiadhikarath tweeted:

“A few hours after the launch of @Google Gemini, search engines like Bing have indexed public conversations from Gemini.”

They posted a screenshot of the site search of gemini.google.com/share/

But if you look at the screenshot, you’ll see that there’s a message that says, “We would like to show you a description here but the site won’t allow us.”

By early morning on Tuesday February 13th the Google Gemini chats began dropping off of Google search results, Google was only showing three search results. By the afternoon the number of leaked Gemini chats showing in the search results had dwindled to just one search result.

How Did Gemini Chat Pages Get Created?

Gemini offers a way to create a link to a publicly viewable version of a private chat.

Google does not automatically create webpages out of private chats. Users create the chat pages through a link at the bottom of each chat.

Screenshot Of How To Create a Shared Chat Page

Why Did Gemini Chat Pages Get Indexed?

The obvious reason for why the chat pages were crawled and indexed is because Google forgot to put a robots.txt in the root of the Gemini subdomain, (gemini.google.com).

A robots.txt file is a document for controlling crawler activity on websites. A publisher can block specific crawlers by using commands standardized in the Robots.txt Protocol.

I checked the robots.txt at 4:19 AM on February 13th and saw that one was in place:

I next checked the Internet Archive to see how long the robots.txt file has been in place and discovered that it was there since at least February 8th, the day that the Gemini Apps were announced.

That means that the obvious reason for why the chat pages were crawled is not the correct reason, it’s just the most obvious reason.

Although the Google Gemini subdomain had a robots.txt that blocked web crawlers from both Bing and Google, how did they end up crawling those pages and indexing them?

Two Ways Private Chat Pages Discovered And Indexed

• There may be a public link somewhere.
• Less likely but maybe possible is that they were discovered through browsing history linked from cookies.

It’s likelier that there’s a public links. But if there’s a public link then why did Google start dropping chat pages altogether? Did Google create an internal rule for the search crawler to exclude webpages from the /share/ folder from the search index, even if they’re publicly linked?

Insights Into How Bing and Google Search Index Content

Now here’s the really interesting part for all the search geeks interested in how Google and Bing index content.

The Microsoft Bing search index responded to the Gemini content differently from how Google search did. While Google was still showing three search results in the early morning of February 13th, Bing was only showing one result from the subdomain. There was a seemingly random quality to what was indexed and how much of it.

Why Did Gemini Chat Pages Leak?

Here are the known facts: Google had a robots.txt in place since the February 8th. Both Google and Bing indexed pages from the gemini.google.com subdomain. Google indexed the content regardless of the robots.txt and then began dumping them.

• Does Googlebot have a different instructions for indexing content on Google subdomains?
• Does Googlebot routinely crawl and index content that is blocked by robots.txt and then subsequently drop it?
• Was the leaked data linked from somewhere that is crawlable by bots, causing the blocked content to be crawled and indexed?

Content that is blocked by Robots.txt can still be discovered, crawled and end up in the search index and ranked in the SERPs or at least through a site:search. I think this may be the case.

But if that’s the case, why did the search results begin to drop off?

If the reason for the crawling and indexing was because those private chats were linked from somewhere, was the source of the links removed?

The big question is, where are those links? Could it be related to annotations by quality raters that unintentionally leaked onto the Internet?

Google Gemini Warning: Don’t Share Confidential Information via @sejournal, @martinibuster

Google Gemini privacy support pages warn that information shared with Gemini Apps may be read and annotated by human reviewers and also be included into AI training datasets.  This is what you need to know and what actions are available to prevent this from happening.

Google Gemini

Gemini is the name for the technology underlying the Google Gemini Android App available on Google Play, a functionality in the Apple iPhone Google App and a standalone chatbot called Gemini Advanced.

Gemini on Android, iPhone & Gemini Advanced

Gemini on mobile devices and the standalone chatbot are multimodal. Multimodal means that users can ask it questions with images, audio or text input. Gemini can answer questions about things in the real world, respond to questions, can perform actions, provide information about an object in a photo or provide instructions on how to use it.

All of that data in the form of images, audio and text are submitted to Google and some of it could be reviewed by humans or included in AI training datasets.

Gemini Uses Gemini Data To Create Training Datasets?

Gemini Apps uses past conversations and location data for generating responses, which is normal and reasonable. Gemini also collects and stores that same data to improve other Google products.

This is what the privacy explainer page says about it:

“Google collects your Gemini Apps conversations, related product usage information, info about your location, and your feedback. Google uses this data, consistent with our Privacy Policy, to provide, improve, and develop Google products and services and machine learning technologies, including Google’s enterprise products such as Google Cloud.”

Google’s privacy explainer says that the data is stored in a users Google Account for up to 18 months and they are able to limit the data storage to three months and 36 months.

There’s also a way to turn off saving data to a users Google Account:

“If you want to use Gemini Apps without saving your conversations to your Google Account, you can turn off your Gemini Apps Activity.

..Even when Gemini Apps Activity is off, your conversations will be saved with your account for up to 72 hours. This lets Google provide the service and process any feedback. This activity won’t appear in your Gemini Apps Activity.”

But there’s an exception to the above rule that lets Google hold on to the data for even longer.

Human Reviews Of User Gemini Data

Google’s Gemini privacy support page explains that user data that is reviewed and annotated by human reviewers is retained by Google for up to three years.

“How long is reviewed data retained
Gemini Apps conversations that have been reviewed by human reviewers (as well as feedback and related data like your language, device type, or location info) are not deleted when you delete your Gemini Apps activity because they are kept separately and are not connected to your Google Account. Instead, they are retained for up to 3 years.”

The above cited support page informs that human reviewed and annotated data is used to create datasets for Chatbots:

“These are then used to create a better dataset for generative machine-learning models to learn from so our models can produce improved responses in the future.”

Google Gemini Warning:  Don’t Share Confidential Data

Google’s Gemini privacy explainer page warns that users should not share confidential information.

It explains:

“To help with quality and improve our products (such as generative machine-learning models that power Gemini Apps), human reviewers read, annotate, and process your Gemini Apps conversations. We take steps to protect your privacy as part of this process. This includes disconnecting your conversations with Gemini Apps from your Google Account before reviewers see or annotate them.

Please don’t enter confidential information in your conversations or any data you wouldn’t want a reviewer to see or Google to use to improve our products, services, and machine-learning technologies.

…Don’t enter anything you wouldn’t want a human reviewer to see or Google to use. For example, don’t enter info you consider confidential or data you don’t want to be used to improve Google products, services, and machine-learning technologies.”

There is a way to keep all that from happening. Turning off Gemini Apps Activity stops user data from being shown to human reviewers, so there is a way to opt-out and not have the data stored and used to create datasets.

But, Google still stores data up to 72 hours in order to have a backup in case of a failure but also for sharing with other Google services and with third party services that a user may interact with while using Gemini.

Using Gemini Can Lead To 3rd Party Data Sharing

Using Gemini can start a chain reaction of other apps using and storing user conversations, location data and other information.

The Gemini privacy support page explains:

“If you turn off this setting or delete your Gemini Apps activity, other settings, like Web & App Activity or Location History, may continue to save location and other data as part of your use of other Google services.

In addition, when you integrate and use Gemini Apps with other Google services, they will save and use your data to provide and improve their services, consistent with their policies and the Google Privacy Policy. If you use Gemini Apps to interact with third-party services, they will process your data according to their own privacy policies.”

The same Gemini privacy page links to a page for requesting removal of content, as well as to a Gemini Apps FAQ and a Gemini Apps Privacy Hub to learn more.

Gemini Use Comes With Strings

Many of the ways that Gemini uses data is for legitimate purposes, including submitting the information for human reviews.  But Google’s own Gemini support pages make it very clear that users should not share any confidential information that a human reviewer might see or because it might get included into an AI training dataset.

Featured Image by Judith Linine / Shutterstock.com