Technologies

Gen AI Chatbots Are Starting to Remember You. Should You Let Them?

An AI model’s long memory can offer a better experience — or a worse one. Good thing you can turn it off.

Until recently, generative AI chatbots didn’t have the best memories: You tell it something and, when you come back later, you start again with a blank slate. Not anymore.

OpenAI started testing a stronger memory in ChatGPT last year and rolled out improvements this month. Grok, the flagship tool of Elon Musk’s xAI, also just got a better memory.

It took significant improvements in math and technology to get here but the real-world benefits seem pretty simple: You can get more consistent and personalized results without having to repeat yourself.

“If it’s able to incorporate every chat I’ve had before, it does not need me to provide all that information the next time,” said Shashank Srivastava, assistant professor of computer science at the University of North Carolina at Chapel Hill.

Those longer memories can help with solving some frustrations with chatbots but they also pose some new challenges. As with when you talk to a person, what you said yesterday might influence your interactions today.

Here’s a look at how the bots came to have better memories and what it means for you.

Improving an AI model’s memory

For starters, it isn’t quite a “memory.” Mostly, these tools work by incorporating past conversations alongside your latest query. “In effect, it’s as simple as if you just took all your past conversations and combined them into one large prompt,” said Aditya Grover, assistant professor of computer science at UCLA.

Those large prompts are now possible because the latest AI models have significantly larger “context windows” than their predecessors. The context window is, essentially, how much text a model can consider at once, measured in tokens. A token might be a word or part of a word (OpenAI offers one token as three-quarters of a word as a rule of thumb).

Early large language models had context windows of 4,000 or 8,000 tokens — a few thousand words. A few years ago, if you asked ChatGPT something, it could consider roughly as much text as is in this recent CNET cover story on smart thermostats. Google’s Gemini 2.0 Flash now has a context window of a million tokens. That’s a bit longer than Leo Tolstoy’s epic novel War and Peace. Those improvements are driven by some technical advances in how LLMs work, creating faster ways to generate connections between words, Srivastava said.

Other techniques can also boost a model’s memory and ability to answer a question. One is retrieval-augmented generation, in which the model can run a search or otherwise pull up documents as needed to answer a question, without always keeping all of that information in the context window. Instead of having a massive amount of information available at all times, it just needs to know how to find the right resource, like a researcher perusing a library’s card catalog.

Read more: AI Essentials: 27 Ways to Make Gen AI Work for You, According to Our Experts

Why context matters for a chatbot

The more an LLM knows about you from its past interactions with you, the better suited to your needs its answers will be. That’s the goal of having a chatbot that can remember your old conversations.

For example, if you ask an LLM with no memory of you what the weather is, it’ll probably follow up first by asking where you are. One that can remember past conversations, however, might know that you often ask it for advice about restaurants or other things in San Francisco, for example, and assume that’s your location. “It’s more user-friendly if the system knows more about you,” Grover said.

A chatbot with a longer memory can provide you with more specific answers. If you ask it to suggest a gift for a family member’s birthday and tell it some details about that family member, it won’t need as much context when you ask again next year. “That would mean smoother conversations because you don’t need to repeat yourself,” Srivatsava said.

A long memory, however, can have its downsides.

You can (and maybe should) tell AI to forget

Having a chatbot recommend a gift poses a conundrum that’s all too common in human memories: You told your aunt you liked airplanes when you were 12 years old, and decades later you still get airplane-themed gifts from her. An LLM that remembers things about you could bias itself too much toward something you told it before.

“There’s definitely that possibility that you can lose your control and that this personalization could haunt you,” Srivastava said. “Instead of getting an unbiased, fresh perspective, its judgment might always be colored by previous interactions.”

LLMs typically allow you to tell them to forget certain things or to exclude some conversations from their memory.

You may also deal with things you don’t want an AI model to remember. If you have private or sensitive information you’re communicating with an LLM (and you should think twice about doing so at all), you probably want to turn off the memory function for those interactions.

Read the guidance on the tool you’re using to be sure you know what it’s remembering, how to turn it on and off and how to delete items from its memory.

Grover said this is an area where gen AI developers should be transparent and offer clear commands in the user interface. “I think they need to be providing more controls that are visible to the user, when to turn it on, when to turn it off,” he said. “Give a sense of urgency for the user base so they don’t get locked into defaults that are hard to find.”

How to turn off gen AI memory features

Here’s how to manage memory features in some common gen AI tools.

ChatGPT

OpenAI has a couple types of memory in its models. One is called “reference saved memories” and it stores details that you specifically ask ChatGPT to save, like your name or dietary preferences. Another, “reference chat history,” remembers information from past conversations (but not everything).

To turn off either of these features, you can go to Settings and Personalization and toggle the items off.

You can ask ChatGPT what it remembers about you and ask it to forget something it has remembered. To completely delete this information, you can delete the saved memories in Settings and the chat where you saved that information.

Gemini

Google’s Gemini model can remember things you’ve discussed or summarize past conversations.

To modify or delete these memories, or to turn off the feature entirely, you can go into your Gemini Apps Activity menu.

Grok

Elon Musk’s xAI announced memory features in Grok this month and they’re turned on by default.

You can turn them off under Settings and Data Controls. The specific setting is different between Grok.com, where it’s “Personalize Grok with your conversation history,” and on the Android and iOS apps, where it’s “Personalize with memories.”

Technologies

U.S. diesel price tops $6 per gallon, a record high as Ukraine and Iran wars ripple through economy

U.S. diesel prices hit their highest level ever as fuel supply disruption stemming from the Ukraine and Iran wars lifts transportation costs.

U.S. diesel prices hit $6 per gallon on Friday for the first time ever, as fuel supply disruptions triggered by the Ukraine and Iran wars raises transportation costs across the entire economy.

Truckers and farmers are paying about 63% more to fill up their semis and tractors than they did at this time last year, according to data from AAA. The average price nationwide is now about $6.06 per gallon.

Prices are even higher in California, the biggest agriculture state in the U.S., at $7.98 per gallon.

Fuel costs are rising as crude oil prices have surged in response to a sharp escalation in fighting between the U.S. and Iran this month. U.S. crude oil futures topped $100 per barrel on Thursday for the first time since May. The contract has gained about 20% in September.

Diesel is the real lifeblood of the economy even though consumers tend to pay more attention to retail gasoline prices, said Bob McNally, president of Rapidan Energy, in an interview with CNBC’s “The Exchange” on Tuesday.

Higher diesel prices are passed down to consumers in what they pay for food, consumer goods and energy. Diesel fuels the trucks, trains and ships that bring goods to market. It powers the machinery that farmers use to plant and harvest food. And it heats homes and generates electricity in some cases.

This content is blocked because you are not allowing cookies.

To view this content, click on Cookie Preferences here or at the bottom of the page to allow all cookies.

“It’s the more insidious, more costly, and more impactful fuel,” McNally said. “As we climb higher, it is a real concern.”

Diesel prices at these levels will be a “silent killer” for the economy, said Patrick De Haan, head of petroleum analysis at GasBuddy, in an interview with CNBC’s “Power Lunch” Tuesday.

Gasoline prices, meanwhile, have never been this high this late in the year, De Haan said. Prices at the pump hit a Labor Day record of $4.15 per gallon earlier this week. Americans are spending about $700 million more per day on gas and diesel than they did a year ago, the analyst said.

“There’s sticker shock there for consumers,” De Haan said.

Fuel costs are rising as the Iran and Ukraine wars have disrupted global supplies. Kyiv has pounded Russian refineries, forcing Moscow to ban diesel exports. Iran and its militant Houthi allies in Yemen have also hit the refineries of U.S. Gulf allies. Fuel exports through the Strait of Hormuz are constrained due to the Iranian attacks on tankers.

The wars in Eastern Europe and the Middle East have shut down refineries with about 5 million barrels per day of capacity, said Valero Chief Operating Officer Gary Simmons on the U.S. refiner’s July 30 earnings call.

The world has lost nearly 8% of its diesel supply with little spare refining capacity available to make up the shortfall, said Andy Lipow, president of Lipow Oil Associates, in a Wednesday note.

Rising diesel prices pose an “enormous challenge” for the Trump administration, said Helima Croft, head of global commodity strategy at RBC Capital Markets, in a Sept. 4 interview with CNBC’s “Power Lunch.”

“U.S. refineries are running at 98% utilization rates — there is just no spare capacity,” Croft said.

Technologies

Buffett’s confidence in troubled decade-old acquisition finally pays off

Warren Buffett has said he paid too much for Precision Castparts in 2016. Now its complex metal castings are in high demand.

(This is the Warren Buffett Watch newsletter, news and analysis on all things Warren Buffett and Berkshire Hathaway. You can sign up here to receive it every Friday evening in your inbox.)

Buffett’s confidence in troubled decade-old acquisition finally pays off

Six years ago, when Berkshire Hathaway took an $11 billion write-down of its $37.2 billion 2016 acquisition of Precision Castparts, Warren Buffett wrote in his annual letter to shareholders he had paid “too much” for the company, which makes “complex metal components and products.”

While it was a “fine company – the best in its business,” he had been “simply too optimistic” about its profit potential, a “miscalculation … laid bare” by the enormous downturn for the aerospace industry, Precision Castparts’ largest customers, amid the Covid pandemic.

In a CNBC interview when the deal was first announced, Buffett admitted it was “a very high multiple for us to pay,” but told shareholders at the 2016 meeting he had great confidence in Mark Donegan, the company’s CEO, both then and now, and the company’s long-term profit outlook.

It’s taken longer than he planned, but Buffett’s purchase is now looking pretty good.

As Reuters puts it, there is currently a shortage of the “complex” products Precision Castparts makes that are essential for engine turbine blades.

They’re also used in natural gas turbines, which are in demand to produce energy for artificial intelligence data centers.

This week, GE Aerospace announced it would pay $11.75 billion to acquire Consolidated Precision Products, one of the few companies that competes against Precision Castparts.

Barron’s calls that “pricey” at 26 times projected 2027 earnings before interest, taxes, depreciation, and amortization.

Using the same multiple, Barron’s estimates Precision Castparts is worth around $100 billion. That’s well above the potential value of $60 billion to $75 billion it cited in an article last month that said the unit “probably has become one of the more valuable divisions” of Berkshire.

It’s also nearly three times the 2016 purchase price.

In the Barron’s piece, Andrew Bary said Berkshire, and its share price, aren’t “getting much credit” for the subsidiary’s rising value, in part because CEO Greg Abel, like Buffett, doesn’t do analyst conference calls or investor events that could draw attention to the unit’s performance.

His recommendation: “Without Warren Buffett at the helm, Berkshire may have to start telling its story if it wants to attract a new generation of investors. This year’s trading action suggests that something may need to change.”

Berkshire bounces a bit as Wall Street sells off

Berkshire Hathaway shares managed a modest gain this week even as Wall Street’s major averages declined, a small departure from the 2026 “trading action” Bary cites.

Both the Class A and Class B shares gained almost 0.9% while the S&P 500 fell by 0.8%.

Until Friday’s bounce, that benchmark index, along with the Dow Industrials and the Nasdaq Composite, had dropped four days in a row as oil and bond yields moved higher.

Even with this week’s outperformance, Berkshire’s B shares still trail the S&P 500 by more than 10 percentage points so far this year.

Nebraska candidate moves to replace ad that included Buffett’s image

The campaign team for the Republican running in Nebraska’s 2nd Congressional District accelerated the deployment of a new campaign ad after Susie Buffett complained about a previous commercial that briefly included an image of her father, Warren Buffett.

In the ad, a picture of Buffett and his name appear on screen for roughly two seconds as candidate Brinker Harding says, “Here in Omaha, we know a thing or two about the stock market, some more than others. But we do it without insider information.”

He then goes on to highlight his call for a ban on Congressional stock trading, saying some lawmakers “trade on secrets you’ll never know,” as they “get rich” while “we barely get by.”

In a report that led its 10 PM CT newscast Wednesday evening, ABC affiliate KETV in Omaha reported Susie Buffett had asked Harding on Sept. 2 to remove the ad.

She told the station, “I think it’s worth it to say that Warren did not give Brinker his permission to use his face or name in his ad.

“It implies that my dad endorses him. He did not have permission to use it.”

The KETV report quoted Harding as saying in a statement, “In Nebraska, we work hard and support each other, and we do it honestly. Warren Buffett exemplifies that, and that was the point of my ad.”

The report said Harding did not comment on whether the ad would be taken down but noted “it does look like new ads from his campaign are beginning to run on some stations.”

A Harding campaign spokesperson told me the campaign did not think its ad implied a Buffett endorsement, but to be respectful to the Buffett family, it responded to her concern by accelerating the rollout of its next planned ad by several days, although its effort was hampered by the Labor Day weekend.

The commercial now running does not show or mention Buffett.

BUFFETT & BERKSHIRE AROUND THE INTERNET

Some links may require a subscription:

– Best’s News and Research Service: 2026 Best’s Rankings: Berkshire Hathaway Takes DPW Top Spot Among Accident & Health Lines

– Financial Times: The day Warren Buffett saved Salomon Brothers

HIGHLIGHTS FROM CNBC’S BUFFETT ARCHIVE

The effects of 9/11 on Berkshire and the insurance industry (2002)

Warren Buffett shares his thoughts on the 9/11 attacks and explains how Berkshire’s insurance companies have started taking terrorism into account when writing policies.

AUDIENCE MEMBER: I know you lost a lot of money as a result of 9/11. But I would like to know how 9/11 changed your life and your investment strategy?

WARREN BUFFETT: It made everybody, I think, in the country aware, I mean, we’ve gone through world wars and all of that, and essentially felt quite protected within these borders.

And I have been quite worried about — Charlie can attest to — you know, the possibility, particularly of some kind of nuclear device in this country, by — probably more likely by terrorists than by some, at least, declared act of war by another state.

And 9/11 made everybody realize that as humans have not progressed, particularly, in terms of how they behave with each other over the years, they have progressed enormously in their ability to inflict damage on those they hate for one reason or another…

In terms of the business aspects of it, in your question, obviously the area at Berkshire that it effects most significantly, by miles, is insurance.

And prior to 9/11, even though we recognized that there could be huge monetary damages that flowed from the activities of what I would call deranged people, we hadn’t really written the contracts in such a way as to either get paid for taking that risk or to exclude the risk. In other words, we were throwing it in for nothing.

We had excluded risk for war. I mean, we knew that we’d seen what had happened in England in the 40s, and so we had taken account of something that some of us had seen with our own eyes, but we didn’t take account of something that we knew was possible, but we just hadn’t seen. And that’s, you know, that’s the human condition, to some degree.

Since September 11th, everybody in the insurance business recognizes that they had exposures that they weren’t charging for, and they either had to exclude those exposures or they had to charge for them.

We have written — first thing we had to do, of course, is we had lots of policies on the books that left us exposed to this, and most of those policies ran for a year, starting at different points. Those have run off to a great degree, but they’re not entirely run off.

The other thing we did was on new policies. We have sold a fair amount, quite a large amount, of terrorism insurance that excludes what we call NCB, nuclear, chemical, and biological, as well as fire following nuclear.

And, we can take a fair amount of exposure to that sort of terrorism, because it doesn’t — it won’t aggregate. It aggregated at the Twin Towers in a way that — World Trade Center — in a way that just about was as extreme as you could get for non-NCB-type activities.

I mean, that was a huge amount of damage done without nuclear, chemical, or biological.

But we can have tens of billions of dollars with NCB excluded throughout a greater New York area, or something, but we can’t have hundreds of billions of exposure that would be exposed, say, to, nuclear activities, because there an act or two, or three, coordinated, could cause damage that would destroy the insurance industry.

And if we had coverage on that, it would destroy us as well.

BERKSHIRE STOCK WATCH

Four weeks

Twelve months

BRK.A stock price: $766,000.00

BRK.B stock price: $510.37

BRK.B P/E (TTM): 12.83

Berkshire Cash as of June 30: $365.5 billion (Down 8.0% from March 31)

Excluding Rail Cash and Subtracting T-Bills Payable: $359.2 billion (Down 3.8% from March 31)

Berkshire repurchased $4.5 billion of its shares in Q2 2026.

BERKSHIRE’S TOP EQUITY HOLDINGS – Sep. 11, 2026

Berkshire’s top holdings of disclosed publicly traded stocks in the U.S. and Japan, by market value, based on the latest closing prices.

Holdings are as of June 30, 2026, as reported in Berkshire Hathaway’s 13F filing on August 14, 2026, except for:

– Mitsubishi, which is as of April 30, 2026

The full list of holdings and current market values is available from CNBC.com’s Berkshire Hathaway Portfolio Tracker.

QUESTIONS OR COMMENTS

Please send any questions or comments about the newsletter to me at alex.crippen@cnbc.com. (Sorry, but we don’t forward questions or comments to Buffett himself.)

If you aren’t already subscribed to this newsletter, you can sign up here.

Also, Buffett’s annual letters to shareholders are highly recommended reading. There are collected here on Berkshire’s website.

— Alex Crippen, Editor, Warren Buffett Watch

Technologies

Wall Street firm warns AI stock rally may be nearing its end: key reasons

Capital Economics says that while the S&P 500 may keep rising this year, the AI‑driven rally shows multiple bubble indicators and is expected to peak within months, with a projected decline to 6,500 by late 2027.

Various signs of a market bubble indicate that although the S&P 500’s rally can continue this year, its medium‑term outlook appears weak because the market has become overly frothy, according to Capital Economics.

James Reilly, senior market economist at Capital Economics, noted on Thursday that most indicators point to the AI equity rally being close to its end.

Since mid‑2023, Capital has been more optimistic than most about the stock market, viewing AI as a transformative technology.

The firm’s year‑end 2026 S&P 500 forecast has consistently exceeded consensus estimates.

Nevertheless, Capital maintains that the AI‑driven rally is a bubble destined to burst.

To identify a late‑stage bubble, Reilly examines eight metrics: valuations, earnings, index concentration, equity issuance, and foreign interest in U.S. stocks.

Several of these metrics are already at or near levels seen before past market peaks.

While earnings expectations appear aligned with a market top, measures such as volatility and leverage are somewhat less concerning.

Earnings are the most significant warning sign.

S&P 500 earnings growth expectations are hovering at levels only seen at the dot‑com bubble peak, and long‑term EPS forecasts have reached a record high.

Reilly argues that the tech sector’s heavy concentration of this growth means any weakness in tech earnings will heavily drag on the index.

Additional warning signals are also emerging.

Index concentration is approaching dot‑com era extremes, net equity issuance has turned positive, and foreign ownership of U.S. stocks is at a record level.

Reilly warns that another wave of IPOs and share sales could be especially significant, as past issuance booms have historically coincided with market peaks.

He adds that, based on history, the bubble’s end is likely just months away, not years.

Leverage measures are not yet alarming compared with other factors, though the analyst cautions they are moving in a concerning direction.

Volatility indicators resemble those of a mid‑stage bubble, but constituent‑level volatility is not as extreme as at the dot‑com bust’s end.

Reilly expects the S&P 500 to rise from roughly 7,650 now to about 8,250 by the end of 2026, but ultimately projects a decline to 6,500 by the end of 2027.

These projections imply an 8% gain this year and a 21% drop in 2027.

Most signs point to the AI equity rally being close to its conclusion, Capital Economics senior market economist James Reilly stated on Thursday in a note.

The platform has become part of Versant. By using this service, you accept our Terms and acknowledge the revised Privacy Policy, which also covers existing data. For details on your data rights, click the link.

We will also employ essential and service‑related cookies, including those for security and fraud prevention.

– Gather device data and leverage it for personalized ads, ad metrics, audience insights, and service improvements.

– Privacy Management

– Essential Functions

– Record and convey user privacy preferences 515 partners can use this special purpose

– Maintain security, thwart fraud, and resolve issues 649 partners can use this special purpose

– Provide and display ads and content 638 partners can use this special purpose

– Collect and retrieve device data 844 partners can use this purpose

– Tailor ads and content, measure performance, conduct audience research, and develop services 985 partners can use this purpose

– Targeted Ads

– Content Curation

– Employ limited data for ad selection 785 partners can use this purpose