Free LLM credits and offers, updated every 30 minutes

According to a Nov 3 report by Dachang Ribao, ByteDance has introduced new rules that will broadly restrict internal use of personal accounts such as Claude and Codex. The report says that starting this month, ByteDance will ban external AI access to all L4-level code repositories and Feishu documents, covering the entire company, and AI accounts purchased personally by employees can no longer be used. The report also says that after the notice was issued last week, it sparked some internal discussion, and ByteDance Technology Vice President Hong Dingkun also gave a detailed response on the internal network, for example, "once L3 and above content leaks, the consequences are very serious," and "promoting a product does not require banning other options and forcing people to use it, because that effect is very poor." According to an August report by The Information this year, ByteDance founder Zhang Yiming clearly stated at the Seed team all-hands meeting the previous month that even if it means temporarily falling behind competitors, the company will not rely on AI distillation technology to improve its models. Extended reading: In ByteDance's job-level system, L4 usually corresponds to the former 3-1 level, belongs to team TLs or junior managers, and is the dividing line between core technical backbone and managers. L4-level code repositories have nearly accumulated the core assets and underlying architecture of the company's entire line of business, which is also why this ban is highly valued internally. This is not the first time ByteDance has tightened the use of external AI tools. In May 2025, ByteDance's security and risk control department once sent an email saying that, out of consideration for preventing data leakage risks, starting from June 30, 2025, it would ban third-party AI development software internally in batches, including AI programming tools Cursor, Windsurf, and others, and would make its own programming assistant ...

According to a report by Business Insider in the early hours of today (the 4th), OpenAI's challenge to Google Search has now "achieved initial results." Adam Fry, head of consumer products for ChatGPT, confirmed that the volume of search-type queries on ChatGPT is approximately equivalent to 20% of Google's search query volume. This summer, a survey report released by the Pew Research Center showed that searching for information is the most common AI use case among American adults, with 42% of respondents having used AI to search for information. Work purposes ranked second, with 38% of employed adult respondents having used AI to handle work tasks. Uses such as entertainment, getting news, and seeking medical advice were relatively less common. This survey was not specifically about ChatGPT. The report noted that ChatGPT is the most popular among AI chatbots. 44% of surveyed American adults have used ChatGPT, while Gemini, ranked second, has only 24%. According to IT Home, last year, Sundar Pichai, CEO of Google's parent company Alphabet, also discussed the competition between AI chatbots and Google Search in an interview with Bloomberg reporter Emily Chang. Pichai said at the time: "Chatbots have reached hundreds of millions of users, but our search query volume is also growing. So I think this is far from a zero-sum game. To give a similar example, after TikTok appeared, everyone started using TikTok, but YouTube also kept growing during that period, and performed very well."

On Nov 3 local time, Google announced that starting in 2027 it will adjust its policy for managing AI features for K-12 (kindergarten through high school) education users; school administrators will need to actively enable them before student accounts can use Gemini, NotebookLM, Gemini in Classroom, Google Search, and related services. Google also made clear that if a school chooses to fully disable AI, students will also no longer be able to continue using Google Search. According to Google's official announcement, the company will adjust the management approach for education products after communicating with schools, teachers, researchers, and policymakers. After the new policy takes effect, among school-managed K-12 student accounts, Gemini, Gemini Notebook, Gemini in Classroom, and Google Search will be off by default, and can be used only after school administrators actively authorize them. Google said the move is intended to let schools themselves decide how AI tools are used in the classroom. However, this policy also means schools cannot retain traditional search services while fully disabling AI. Chris Phillips, Vice President and General Manager of Google's education business, said in an interview with The New York Times that search and AI are already deeply integrated. If a school chooses to completely turn off AI features, students will be unable to access Google Search, and when attempting to use the search service they will only see a disabled notice message provided by the school. According to tech media outlet TechCrunch, schools currently cannot separately turn off the AI features in Google Search, …

Called FlashProve, the system would let anyone check that an AI response really came from the model it claims to be from.

On Nov 2 local time, OpenAI consumer product lead Adam Fry was interviewed by Big Technology during an OpenAI event held in New York. He revealed that ChatGPT's weekly active user count has reached as high as 1.2 billion. Fry said ChatGPT's user growth is accelerating further. Over the past 60 days, ChatGPT's weekly active users grew from 1 billion to 1.2 billion, adding about 200 million users. He summarized this growth rate as adding about 100 million users every 30 days. On the search business, Fry revealed that OpenAI tallied search-type queries within ChatGPT and compared them with Google's search query volume; the results show that ChatGPT currently handles search-type query volume at about one-fifth of Google's, generates 1.5 billion images per week, and the combined weekly active users of voice input and voice conversation features reach 250 million. Fry explained that as long as a user's question triggers ChatGPT's web search, it is counted as one search. Relevant queries include information retrieval about people, places, and things, as well as finance, health, and other questions that require the latest information. Pure writing and coding requests do not fall into this statistical category. Fry also mentioned that ChatGPT's voice input and voice conversation features together have 250 million weekly active users, and the number of images generated per week reaches 1.5 billion. This shows that voice interaction and image generation have also become important usage scenarios for ChatGPT.

Apple pushed the first iOS 27.3 developer preview to iPhone early this morning, bringing "Apple Intelligence & Siri" to China-market users as promised. IT Home tried out the first batch of core features launched for China-market Apple Intelligence right away, and here is a summary of the highlights: New Siri invocation light effect. The most noticeable change is that when you long-press the power button to invoke Siri, a ring of marquee light effects appears on the screen. Live Translation. The Live Translation feature has been integrated into Messages, FaceTime calls, and the Phone app, supporting users in obtaining translation experience in call and messaging scenarios. At the same time, in face-to-face communication scenarios, Live Translation will also extend to AirPods (requires the latest firmware installed and pairing with an iPhone with Apple Intelligence enabled); after users wear compatible earphones, they can obtain real-time translation capability in cross-language communication. This feature has clear device requirements: users need to use iPhone 16 and later models, iPhone Duo, iPhone 15 Pro or iPhone 15 Pro Max, iPhone Air, and set Siri and device language to English or Simplified Chinese. AirPods models that support Live Translation include: AirPods 5, AirPods 4 (with Active Noise Cancellation), AirPods Pro 2 and later models, AirPods Max 2. Writing Tools. Writing Tools support helping users rewrite, proofread, and summarize input text. Visual Intelligence. Visual Intelligence supports instantly understanding surrounding objects and places through the iPhone camera. Long-press the iPhone side Camera Control button…

In many circles, "Ask Chat" has replaced "Google it," and that's reflected in a new announcement from OpenAI: the volume of search queries run through ChatGPT now equals 20 percent of the volume run through Google. Adam Fry, OpenAI's consumer product lead for ChatGPT, made the announcement at a press event on Monday. He said […]

Google recently announced that starting in 2027 it will adjust its artificial intelligence strategy for K-12 education users, allowing schools to decide on their own whether to enable AI tools such as Gemini. However, this new policy, which appears to give schools more choice, has sparked widespread discussion in the education sector, because Google simultaneously confirmed that if a school decides to completely disable AI features, students will also no longer be able to continue using the Google Search search service. Read the full article

The Google DeepMind team yesterday (Nov 3) published a blog post announcing a set of "Cheminformatics Skills" for Google Antigravity, to help researchers more quickly complete work in drug discovery such as literature retrieval, molecular analysis, model building, and new molecule design. According to the blog post cited by IT Home, in a demonstration using lung cancer drugs as an example, the AI agent could within about 1 hour retrieve thousands of compounds, carry out matched molecular pair analysis, and propose novel rigid macrocyclic molecules with synthetic routes, whereas similar workflows previously often took several days. The tool combines the planning, retrieval, and scientific reasoning capabilities of large language models with deterministic chemical computation tools such as RDKit, and connects to public chemical databases such as ChEMBL and PubChem. The system can identify the same molecule appearing in different forms such as IUPAC names, Markush structures, or SDF files from papers, patents, and product catalogs, reducing the time researchers spend on data cleaning and format conversion. In the demonstration, the AI analyzed two types of lung cancer targeted drugs, crizotinib and Lorlatinib. It retrieved and organized 568 compounds, identified the association between "molecular rigidification" and improved efficacy, then trained an efficacy prediction model and designed multiple potentially better macrocyclic compounds, while also providing reaction mechanisms and synthesis suggestions. Google emphasized that the AI produces "testable scientific hypotheses," not finished drugs; due to data sparsity, inter-laboratory assay errors, and dynamic range limitations, all candidate molecules must still undergo synthesis and experimental validation. This feature is currently available to Googl…

Google is completely turning off Search in schools that reject its AI features, the company told The New York Times. The decision was also shared in a broader education-related update from Google on Tuesday, saying that access to Gemini, Gemini Notebook, Gemini in Classroom, and Google Search "will be entirely off" for K-12 student accounts […]

Schools choosing to restrict AI altogether will no longer be able to access Google Search, as AI features are now integrated into the search engine.

Adam Fry, OpenAI's head of consumer products, recently disclosed a series of the latest operational data for ChatGPT in an interview with the tech media outlet Big Technology. According to his introduction, ChatGPT's weekly active users have reached 1.2 billion, with about 200 million new users added in the past 2 months; its search-type query volume is currently about one-fifth of Google's. In addition, ChatGPT generates 1.5 billion images per week, and the combined weekly active users of voice input and voice conversation have also reached 250 million. Read the full article

According to Reuters, an adviser to the EU's top court said on Tuesday that Google's Gemini chatbot's use of news reports when answering internet users' questions may, under certain circumstances, violate copyright rules. IT Home notes that this opinion was issued by Maciej Szpunar, Advocate General at the Court of Justice of the European Union (CJEU) in Luxembourg, and concerns a lawsuit brought against Google by Hungarian website operator Like Company, with the dispute centered on how Google's AI chatbot Gemini uses news content. If the court ultimately rules against Google and requires the tech giant to pay for the use of news media content, it may create new survival opportunities for online media such as Like Company. As news summaries generated by AI chatbots such as Gemini increasingly replace the search results provided by traditional search engines, which usually include hyperlinks pointing to the original news articles, the website traffic received by online media is continuously declining. According to a statement released by the CJEU, Szpunar said: "When a chatbot powered by a large language model (LLM) obtains an article from a news publisher's website, uses it to generate an answer, and provides that answer to a user, under certain circumstances, it may infringe the rights enjoyed by the news publisher." He pointed out that if a chatbot obtains news content by illegally bypassing paywalls or other technical restrictions, or if it reuses the specific expression of a news report in an identifiable way in its answer, then the answer generated by Gemini may involve infringement of the publisher's reproduction right. However, he also emphasized that copyright rules do not protect the "mere facts" involved in news reports, but only protect these facts...

On the afternoon of Nov 3, DeepSeek's chat function became unavailable, with the chat interface displaying "The server is busy, please try again later." DeepSeek's official website has now updated its status, stating that the partial outage of its web and API services that occurred earlier has been fixed and that it is monitoring the results of the fix. However, the API services and chat services of DeepSeek V4 Pro and DeepSeek V4.1 Flash are still experiencing degraded performance. Read the full article

Today, at Baidu AI Day·Apollo special event held in Shenzhen, Baidu Apollo disclosed for the first time Apollo World, an agent-driven omni-modal world model for physical AI, and proposed the "2+3" criteria for judging world models. Apollo World can generate a relatively realistic video, make the virtual world follow real-world laws and respond to the actions of autonomous vehicles, and support the driving system to start from the origin and drive all the way to the destination. According to IT Home, Apollo World for autonomous driving must not only generate realistic images but also accurately deduce how the real world changes. Baidu Apollo proposed the "2+3" criteria for judging world models: "2" are the basic thresholds: Causality: a world model should follow the one-way nature of time, with the past determining the future, predicting the next moment based on the current state, rather than piecing together images and inferring the past backward. Interactivity: it can dynamically respond to external interventions, injecting actions to affect future world states. "3" are advanced indicators: Physical accuracy: how reliable the simulated state is, whether it conforms to the laws of the world, and whether the learned capabilities can still be used when returned to the physical world. Persistence: whether the continuous running time and stability can keep up, and whether objects and details drift during long-time operation. Real-time performance: low latency, the growth of generated world complexity does not affect model inference time, and high simulation testing efficiency is achieved. In the virtual city built by Apollo World, traffic participants such as vehicles and pedestrians are not "props" acting according to fixed scripts, but AI agents with decision-making and interaction capabilities. For example, facing the same traffic scenario, when the autonomous vehicle takes different actions, the virtual world will also change accordingly: Apollo World can also…

According to feedback from ITHome users, DeepSeek experienced a service outage this afternoon, showing the message "Server is busy, please try again later." DeepSeek's official status page shows that today (Nov 3) at 16:58, DeepSeek web/API experienced a partial disruption, and the official team is investigating the issue. The chat service and the API services of DeepSeek V4 Pro and V4.1 Flash all have problems, so everyone can wait patiently for a while.

Media learned from informed sources in the industry chain that DeepSeek's next-generation large model, DeepSeek V5, plans to use Huawei Ascend 950 chips as its core training compute, with the project timeline pointing to landing in 2027. It is reported that the core R&D liaison team between DeepSeek and Huawei Ascend is currently stationed in Hangzhou, and the two sides have already completed a large amount of low-level operator adaptation and customized development during the V4 stage, laying the foundation for subsequent model training migration and compute collaboration. Read the full article

Most outside attention focused on the latest financial data disclosed in the S-1/A registration statement amendment Anthropic updated on Monday (revenue soaring, and spending surging as well), but this AI lab dropped another bombshell in the fine print of the filing. The filing appears to warn that the company is about to develop models with recursive self-improvement (RSI) capability. Recursive self-improvement refers to the stage in which an AI's self-training capability reaches an extremely high level, thereby driving explosive growth in its own capabilities. For AI risk pessimists, recursive self-improvement is an irreversible catastrophe: AI will self-evolve to become powerful enough to escape human control, and in pursuing its own goals may be indifferent to the harm it causes. In the view of AI optimists, recursive self-improvement, or related forms in which AI steadily achieves self-optimization, represents the enormous potential of this technology, at which point it will become a powerful tool for unlocking super-high productivity and aiding scientific discovery and many other fields. Anthropic's stance is that recursive self-improvement is both an opportunity and a hidden risk. In its filing it wrote that the company already uses AI in R&D work, and cautioned: 'Using AI to develop more capable AI may also bring major safety risks, especially when it greatly accelerates the pace of technological iteration. This is also one of the areas covered by our Responsible Scaling Policy, a document that lays out related plans to monitor and mitigate risks as we gradually approach the so-called recursive self-improvement state.' IT Home noted that when the concept of recursive self-improvement went viral earlier this year, many people regarded it as merely a buzzword, or at best a goal that is hard to implement, hard to define, and impossible to verify. But on Monday...

According to TechCrunch, the interface layout of the ChatGPT desktop software can easily confuse users: it is divided into two major sections, ChatGPT and Codex, and within the ChatGPT section there are two tabs, Chat and Work. With so many options, users sometimes find it easy to lose track of where they currently are in their work and where project files are actually saved. According to an email sent to users, OpenAI is preparing to merge the Chat tab and the Work tab in ChatGPT to solve the above problem. OpenAI said this new feature will first be opened to some users, and after collecting feedback it will be rolled out fully. The email said: "After this adjustment, you can start from brainstorming ideas all the way to having ChatGPT carry out and complete tasks, all within the same interface. You can upload files, call plugins, have ChatGPT build websites, create presentations, or help you handle various operations on web pages, while viewing and guiding its output at any time. Codex will remain as an independent feature in the ChatGPT desktop app." OpenAI also mentioned that after this adjustment is completed, ChatGPT and Codex will share the same token quota pool. The company said this change will not affect most users, but some users will need to wait more frequently for quota resets because of different usage scenarios and models called. IT Home noted that competitor Anthropic already integrated Claude's chat and collaborative office features into the same interface in September.

Moonshot AI's Kimi released the "Game Development and Creative Work Solution" yesterday. This product is based on the Kimi K3.1 series models, Kimi Code and Kimi Work series Agent products, as well as professional tool plugins such as Unity, Unreal Engine, and Blender, and can help teams validate gameplay ideas faster and advance engineering and art production. According to the introduction, the first threshold in game development is the cost of creative validation. Whether a gameplay idea is fun can only be judged by actually playing it. But making a prototype requires time and manpower, and teams often can only prioritize validating a small number of ideas. The Kimi game solution can help game teams efficiently create project prototypes, explore and compare more gameplay directions within limited time, and thereby make more accurate business judgments. The second threshold comes from cross-role and cross-tool collaboration. Design requirements need to become engineering tasks; art assets also need to comply with engine specifications. If AI only generates a piece of text or an image, the team still needs to spend a lot of time adapting and creating before it can be moved into the existing workflow. The Kimi solution can directly connect to the team's existing workflow, operate professional tools and editors, modify existing projects, and deliver files that can continue to be edited and clear operation records, making it convenient for teams to review, collaborate, and advance subsequent work. IT Home learned from the Kimi official website that Kimi can start from the team's actual goals and drive ideas from conception to practice. For example, if a designer wants to verify whether "adding freely combinable rule cards to match-3 gameplay" is fun, it is difficult to feel the strategic changes brought by rule stacking based only on the gameplay description. Kimi will, around this goal, sort out the combination methods of rule cards and their impact on elimination strategy, scoring, and feedback, and then break it down...

According to a report by tech website 9to5mac, OpenAI has notified users who are still using the old ChatGPT Classic app that the app will soon be discontinued. At the same time, OpenAI is also preparing to merge Chat and Work into a unified interface. Read the full article

QuestMobile today released the 2026 China Mobile Internet Autumn Report. As of Sep 2026, the mobile internet monthly active user scale reached 1.281 billion, up 0.9% year-over-year. In terms of user structure, users aged over 46 surpassed 40%, an increase of 2.6% year-over-year (among them, users aged over 51 are approaching 30%), while the user scale in the 25 to 45 age range is 39%, having fallen below 40%; this reversal shows that the "silver-haired" trend of mobile internet users continues. The report states that in terms of user time spent, short drama videos and short videos contributed 111.5% of the growth (cannibalizing time from online video, general information, and other sectors). On terminal devices, Apple took the top spot with an active share of 20.92%, surpassing Huawei's 20.87%, though both declined year-over-year, down 0.3% and 1.7% respectively; OPPO, vivo, and Xiaomi had active shares of 18.24%, 16.79%, and 10.34% respectively, up 0.0 percentage points, 1.3 percentage points, and 0.7 percentage points respectively; in Sep 2026, AI native apps had a monthly active user scale of 546 million, up 82.1% year-over-year; terminal manufacturers' AI applications reached 783 million, up 16.9% year-over-year; PC clients reached 26 million, up 82.3% year-over-year. In contrast, AI application plugins and PC web AI applications declined 15.6% and 10.4% respectively. In Sep 2026, Doubao and DeepSeek had per-user average usage counts of 72.0 times and 6...

The 2026 World Internet Conference opened today (Nov 3) in Wuzhen, Zhejiang. According to Jiemian News, Alibaba Group co-founder and chairman Joseph Tsai said at the opening ceremony that AI demand continues to climb and computing power is in short supply. He believes that "only open source can achieve widespread adoption," and that open source can effectively lower R&D costs for small and medium-sized enterprises and is especially important for developing countries. Tsai disclosed that Alibaba Qwen has so far open-sourced more than 470 models, with cumulative global downloads exceeding 4 billion and more than 330 thousand derivative models. According to an October report by Bloomberg this year, Tsai said that for Europe to break away from dependence on China and the US in technology, the best choice is open-source AI. He called on Europe to build its own computing infrastructure for training and running AI models. IT Home noted that Tsai once predicted that in the future AI will shift from an explicit topic to implicit infrastructure. "Five years from now, people may no longer discuss AI separately, because it will be integrated into every part of business just like the internet today. Related terminology may evolve with the level of technology adoption, but AI, as a fundamental technological breakthrough, will continue to affect business and society." Tsai also elaborated on the AI open-source path of Chinese companies represented by Alibaba. "Just as a top university opens to the world and provides scholarships to students, AI should also open-source its progress so that more people can benefit. This also stems from a philosophical choice of Chinese companies."

Google's Dreambeans, its latest AI experiment, now supports 13 Indian languages.

According to extensive user feedback, OpenAI's next-generation model GPT-6.2 Astra is suspected to have entered small-scale testing, and the standalone model identifier "gpt-6.2-astra" has already appeared in the usage statistics of some Codex users. However, as of now, OpenAI's official API model catalog, changelog, and public announcements have not confirmed this model, and OpenAI has also not confirmed its release date. According to a report yesterday by tech media TestingCatalog, multiple Codex users discovered this new model identifier on the usage analysis page. Screenshots shared by different users show that GPT-6.2 Astra already accounts for part of the model call volume of some accounts, and IT Home noted that the share for one account reached 19.8% (the relevant records mainly appeared in the usage statistics for Oct 31). TestingCatalog believes these records may mean that OpenAI is conducting a gray release test or an A/B test, that is, routing some users' requests to the new model in the background while temporarily retaining the original interface name. As background, OpenAI released GPT-6 Astra on Sep 3, 2026. According to Terminal-Bench 4.0 test results, GPT-6 Astra scored 57.9% and Claude Fable 5.1 scored 55.8%, but the data currently listed on the official Terminal-Bench leaderboard are 58.2% and 57.9% respectively; the gap between the two is small, and the specific scores are also affected by reasoning intensity and test configuration. This rumor is also related to Anthropic…

The "28-Day Update Challenge" initiated by Thibault Sottiaux (Tibo), head of the OpenAI Codex and ChatGPT Work teams, officially ended on Nov 2. The challenge ultimately concluded with "quota reset cards" issued on two consecutive days, Nov 1 and Nov 2, corresponding to Day 27 and Day 28 of the challenge respectively. Tibo announced the start of the "28-Day Challenge" on X on Oct 5 Beijing time, promising that over the following 28 days a clearly meaningful improvement for most Codex / ChatGPT Work users would be released each day, otherwise users would receive one full quota reset. IT Home note: the resets on the final two days do not mean that only two quota resets were issued during the entire 28-day challenge. The third-party tracking site Did Codex Reset shows that resets were offered on 14 days during this period (some days offered both feature updates and resets), among which Day 27 and Day 28 both count as banked resets. The challenge centered mainly on product features, operating efficiency, usage quotas, and model updates. During the period, OpenAI rolled out multiple improvements one after another and also issued full quota resets to users multiple times. According to records from the third-party tracking site, on Day 26 (Oct 30) Tibo announced in advance that banked quota resets (Banked Reset) would be issued over the following two days. Subsequently, Day 27 and Day 28 each delivered one, completing the final two resets. The core promise of Tibo's challenge was not to unconditionally increase usage quotas every day, but to require the team to continuously deliver product improvements that users can perceive…

Instruction-following capability has also been improved simultaneously.

OpenAI released a new benchmark called LifeVal on Nov 2 to measure whether AI models can truly help ordinary people complete everyday tasks; the company says this evaluation aims to make general AI closer to "the most useful everyday help for people." In designing LifeVal's tasks, OpenAI first referenced the US "American Time Use Survey" and the cross-national "Multinational Time Use Survey," then conducted a survey of 3,995 demographically representative respondents, asking them in which matters they most wanted a personal AI assistant to help. Accordingly, LifeVal covers scenarios such as caregiving, health, education, household administration, personal finance, shopping, and accessing government services. Typical tasks include: after moving, selecting and enrolling a suitable kindergarten for a child; creating a diet and care plan for a family member with diabetes; purchasing ingredients within a budget; organizing government forms; arranging community flood-control volunteers; and so on. Unlike traditional benchmarks, LifeVal focuses more on whether tasks can be advanced to an "actionable, deliverable" state. For example, AI should not just list kindergartens but filter them based on location, budget, commuting, and family needs; when handling finances, it should also take bills, income, and risk into account. OpenAI says models may need to understand chat records, browse the web, integrate information from multiple sources, and give recommendations suited to an individual's circumstances. OpenAI says that over the past year it has used LifeVal to track and drive improvements in models on everyday tasks. Public examples show relatively high performance on some tasks: finding a kindergarten for a child scored 93.1%, organizing flood-control volunteers scored 98.0%, and preparing a meal plan for a parent with diabetes scored 90.4%. IT Home attaches the relevant screenshots below: However, L…

Recently, AIsphere's model and open platform PixVerse and OpenAI announced the launch of a deep collaboration.

OpenAI announced yesterday (Nov 2) that it has notified users still using ChatGPT Classic that this old desktop app will be discontinued on Jan 30, 2027. OpenAI stated that after users open the ChatGPT Classic app, they will receive a discontinuation notice, and clicking the banner will take them to the download page for the new ChatGPT; the company says complete chat history will be migrated with the account, and users can continue their existing conversations in the new app. IT Home reported in July this year that OpenAI announced the launch of a brand-new desktop application that will integrate the Codex and ChatGPT experience, enhancing the AI programming experience while expanding coverage to more non-developer groups. After the update, the desktop version of Codex will become the brand-new ChatGPT desktop app, including three components: Chat, ChatGPT Work, and Codex; after the update, the app will retain existing Codex tasks and projects, and users can also choose the Codex app icon in settings. The previous ChatGPT app becomes the ChatGPT Classic app. At the same time, OpenAI is preparing to further simplify its product structure: in the coming weeks it will open testing of the merger of Chat and Work to a small group of users. After the merger, users can complete the process from proposing ideas to having ChatGPT execute tasks within the same interface, for example importing files or plugins, having it build websites, generate presentations, or complete operations on web pages, and review and guide its work at any time. Codex will still remain as a standalone experience in Cha…