AI companies reach out for help: OpenAI records computer actions, while Google uses uploaded content to train AI.
Axios begins investigating what user data AI companies are actually collecting.
The findings show that chat logs aren’t enough anymore. Now they also want to know what you click on your computer, what photos you upload, and what content you’ve viewed—then use that data for memory, personalization, and advertising.
OpenAI has added Computer History. Once Mac users manually enable it, ChatGPT and Codex can record your clicks, inputs, and switches within specific apps and websites, gradually building a long-term work memory about you. The feature is turned off by default; it doesn’t record your screen or audio. OpenAI says temporary action logs won’t be used to train its models.
Google is even more aggressive. A new Search Services History will save images, documents, and audio/video that users upload through services like Lens, Search Live, and Translate—and it can also be used to train generative AI. For accounts that meet the criteria, it’s enabled by default; if you want to opt out, you have to do it yourself. Even some data that is later unlinked from an account may be retained for up to 4 years after entering the training process.
Advertising has kept up too. Meta will use users’ interactions with Meta AI to adjust the content and ads on Facebook and Instagram; OpenAI and Google are also exploring ad models for chatbots. The better AI understands you, the more detailed the “portrait” the ad system has in its hands.
What used to be a concern was that chat logs would be used for training. Now, even how you work on your computer—and what you throw into Google—has started to become data in the hands of AI companies.
No matter whether open-source AI wins or closed-source AI wins, cybersecurity will be the ultimate winner! Palo Alto and CrowdStrike seize the windfall of the AI inference era!
The Wall Street financial giant Wells Fargo Bank recently released a research report stating that cybersecurity supergiant $CrowdStrike (CRWD.US)$ and $Palo Alto Networks (PANW.US)$ are seeing momentum in growth of orders for cybersecurity software products driven by the rapid expansion of the AI inference market. After AI moves into the stage of large-scale, massive inference and AI intelligent agent-style workflows (Agentic AI), cybersecurity demand is not simply likely to accelerate growth by “updating and iterating alongside cutting-edge AI technologies.” Instead, it may see a structural incremental expansion far beyond traditional IT spending.
No matter whether open-source AI wins or closed-source AI wins, cybersecurity will be the ultimate winner! Palo Alto and CrowdStrike seize the windfall of the AI inference era!
The Wall Street financial giant Wells Fargo Bank recently released a research report stating that cybersecurity supergiant $CrowdStrike (CRWD.US)$ and $Palo Alto Networks (PANW.US)$ are seeing momentum in growth of orders for cybersecurity software products driven by the rapid expansion of the AI inference market. After AI moves into the stage of large-scale, massive inference and AI intelligent agent-style workflows (Agentic AI), cybersecurity demand is not simply likely to accelerate growth by “updating and iterating alongside cutting-edge AI technologies.” Instead, it may see a structural incremental expansion far beyond traditional IT spending.
No matter whether open-source AI wins or closed-source AI wins, cybersecurity will be the ultimate winner! Palo Alto and CrowdStrike seize the windfall of the AI inference era!
The Wall Street financial giant Wells Fargo Bank recently released a research report stating that cybersecurity supergiant $CrowdStrike (CRWD.US)$ and $Palo Alto Networks (PANW.US)$ are seeing momentum in growth of orders for cybersecurity software products driven by the rapid expansion of the AI inference market. After AI moves into the stage of large-scale, massive inference and AI intelligent agent-style workflows (Agentic AI), cybersecurity demand is not simply likely to accelerate growth by “updating and iterating alongside cutting-edge AI technologies.” Instead, it may see a structural incremental expansion far beyond traditional IT spending.
No matter whether open-source AI wins or closed-source AI wins, cybersecurity will be the ultimate winner! Palo Alto and CrowdStrike seize the windfall of the AI inference era!
The Wall Street financial giant Wells Fargo Bank recently released a research report stating that cybersecurity supergiant $CrowdStrike (CRWD.US)$ and $Palo Alto Networks (PANW.US)$ are seeing momentum in growth of orders for cybersecurity software products driven by the rapid expansion of the AI inference market. After AI moves into the stage of large-scale, massive inference and AI intelligent agent-style workflows (Agentic AI), cybersecurity demand is not simply likely to accelerate growth by “updating and iterating alongside cutting-edge AI technologies.” Instead, it may see a structural incremental expansion far beyond traditional IT spending.
AI companies reach out for help: OpenAI records computer actions, while Google uses uploaded content to train AI.
Axios begins investigating what user data AI companies are actually collecting.
The findings show that chat logs aren’t enough anymore. Now they also want to know what you click on your computer, what photos you upload, and what content you’ve viewed—then use that data for memory, personalization, and advertising.
OpenAI has added Computer History. Once Mac users manually enable it, ChatGPT and Codex can record your clicks, inputs, and switches within specific apps and websites, gradually building a long-term work memory about you. The feature is turned off by default; it doesn’t record your screen or audio. OpenAI says temporary action logs won’t be used to train its models.
Google is even more aggressive. A new Search Services History will save images, documents, and audio/video that users upload through services like Lens, Search Live, and Translate—and it can also be used to train generative AI. For accounts that meet the criteria, it’s enabled by default; if you want to opt out, you have to do it yourself. Even some data that is later unlinked from an account may be retained for up to 4 years after entering the training process.
Advertising has kept up too. Meta will use users’ interactions with Meta AI to adjust the content and ads on Facebook and Instagram; OpenAI and Google are also exploring ad models for chatbots. The better AI understands you, the more detailed the “portrait” the ad system has in its hands.
What used to be a concern was that chat logs would be used for training. Now, even how you work on your computer—and what you throw into Google—has started to become data in the hands of AI companies.
On August 18, MaxForAI disclosed that an AI project called J-Space Cognition Suite, which has gone viral on the X platform today, is facing criticism from the community. The claim is that the DeepSeek V4 test results promoted by the project cannot be reproduced. The project previously stated that with V4 Flash combined with J-Space, it could match GLM-5.3, and that V4 Pro could surpass Fable 5 on multiple Agent Benchmarks while improving speed by 2.53x and token efficiency by 2.21x.
GitHub user GoForceX used the 87-question subset from Terminal Bench 2.1 to perform high-concurrency re-tests and confirmed that the J-Space related modules were loaded. The results showed that after adding J-Space, the benchmark scores actually dropped slightly, while token usage and cost both increased—opposite to the performance, speed, and token-efficiency improvement directions claimed by the project.
The community subsequently requested that the project release the complete evaluation configuration, per-question results, run logs, original runtimes, and data such as token consumption. So far, the project has mainly published aggregated results and has not provided complete raw experimental records sufficient to verify the precise data mentioned above. Notably, in response to the criticisms, the project’s authors previously said that the relevant data were “indeed exaggerated,” adding that the real improvements are roughly in the range of 1.6x to 3x. As of now, the author has not issued an official response to the community’s concerns, and some related Issues have been deleted.
Global Super-Rich Pile Into SpaceX (SPCX.US) Collectively—More Than a Dozen Family Offices Hold at Least $3.8 Billion!
Super-rich families of billionaires in the United States and across the globe are making large bets on SpaceX (SPCX.US). The latest regulatory filings show that several family offices from the Americas, Europe, and the Middle East have already built SpaceX holdings worth billions of dollars, indicating that ultra-wealthy investors—backed by strong capital—are increasingly occupying an important position in popular investment opportunities. According to an aggregation of media reports on second-quarter 13F regulatory filings, as of the end of June, the family office of Nick Pritzker, the heir to the Hyatt Hotels fortune, held about $1.8 billion worth of SpaceX shares. At that time, it had been only a few weeks since SpaceX completed its initial public offering, and the company’s business spans rocket launches, satellite communications, and artificial intelligence, among other areas. Other super-rich individuals have also built substantial positions.
An Apple research team ran a reinforcement learning experiment with 9 models and 11 languages to find out: if the training data is only in one language, can the model’s problem-solving abilities be applied to questions in other languages?
The answer is yes, and the effect is quite noticeable. Training with only one language still makes the model stronger across many other languages as well. For example, on a French test, training directly in French improved the average score by 25.6 percentage points. If you don’t train in French at all and only use Spanish questions for practice, you can still improve the score on the French test by 24.6 percentage points—just a 1-point difference. What the model learns isn’t only problem-solving for a single language; it also learns some problem-solving methods that can be transferred and continued to be used in other languages. So in the future, if you want to strengthen the model’s Chinese reasoning, you may not need to remake all reinforcement learning data into Chinese.
In a report published by Goldman Sachs on August 14, it said that during this year’s second-quarter earnings season, only 2% of the constituents of the $S&P 500 Index (.SPX.US)$ had quantified the specific impact of AI on earnings!
Goldman Sachs strategist Ben Schneidau said that companies are increasing their investments in artificial intelligence (AI) at an unprecedented pace, yet for most firms, this technology has not yet translated into tangible improvements in earnings. In a report published by Goldman Sachs on August 14, it said that during this year’s second-quarter earnings season, only 2% of the constituents of the $S&P 500 Index (.SPX.US)$ had quantified the specific impact of AI on earnings. Another 11% said they had observed measurable productivity gains in specific areas such as software programming and customer support. However, the companies that have implemented these efficiency improvements have not significantly outperformed overall market levels in terms of profit growth. Data shows that their median year-over-year profit growth was 17%, while companies that have not quantified the contribution of AI efficiencies were at 14%. Goldman Sachs noted that this gap is not statistically significant.
Cherishing hopes, moving forward steadily. May days be calm and worry-free, and may all persistence yield something, with good luck always by your side 🧧 🔥 Comment replies to claim红包 🧧👇
Major scam play exposed! Someone forged CZ to destroy tokens! Don’t be fooled by on-chain records
Major warning ⚠️ A carefully designed scheme has surfaced! Around 16:15 today, rumors spread that CZ’s publicly donated address appeared with three consecutive token destruction records: 4,444 Niu Lai, 4,444 MarsCoin, and 4,444 Binance Life tokens were sent to a “black hole” address.
Many people initially mistook it as CZ personally taking action to destroy tokens, rushing in to trade and bet on the market. However, after tracing on-chain data, the truth is far more shocking!
The token burn of “Niu Lai” was not done by CZ at all. The trader is actually the token issuer. The project team deployed a contract with reserved privileges, privately minted one billion tokens, and first transferred 800 million of them to the CZ address. Without obtaining any authorization from CZ, they then used the contract’s privileges to forcibly transfer out 4,444 tokens from that address to complete the destruction.
Just from the records shown by the blockchain explorer, the transaction initiator appears to be CZ—creating the illusion that a big-shot is backing and burning the token(s), in order to harvest market attention.
This kind of despicable marketing scheme is nothing new! Looking back at the market, the CAAB project team had previously moved 80% of its tokens to CZ’s donation address, promoting a false story about CZ holdings to inflate market value and lure retail investors into the game. After the SHORT token side batch-transferred tokens, the short-term surge caused by token burns also became a tool for the project to unload.
Here is a serious reminder to all participants! Do not determine that CZ is involved in the project, that they actively destroy tokens, or that they endorse the project based on a single on-chain transaction hash alone. As long as the contract reserves administrator privileges, the project team can perform whatever “show” they want. Meme tokens themselves lack real-world use cases, and the risk of extreme pump-and-dump price swings is very high, filled with all kinds of fabricated narratives. Blindly trading based on trending news can easily turn you into the chips on the market-maker’s cutting board—stay vigilant at all times!
⚠️ This content is for informational purposes only and does not constitute investment advice #CZ
When dealing with events, I’m still going strong, and as an old warhorse, I kneel and charge—every day I earn a bit of living expenses; as long as I can buy groceries and eat, that’s enough.