Hy is Tencent's family of AI models, and its X account is @TencentHunyuan. Tencent released an earlier Hy4 preview, which it uses as the baseline for most comparisons. Other leading open weight models from China include DeepSeek, Kimi and GLM, which Tencent compares against in its chart.

Tencent's Hy team (@TencentHunyuan) announced Hy4 early on Nov 4 ET, saying it is firmly among the leading open models and adds vision for the first time.

According to Tencent's announcement page, Hy4 substantially scales up pre training and post training compared with Hy4 preview. Tencent says it adds image understanding and, for the first time, computer use across command line and graphical interfaces, with targeted training on software such as Blender and CAD. Tencent also reports gains in instruction following, long context understanding, long horizon tasks and deep research.

Tencent cites several internal results. It says Hy4 improves by 19.1% on GameDev over Hy4 preview, cut a search hallucination rate from 15.76% to 5.45%, and ranked first among open weight models on 579 finance tasks judged with 213 finance professionals, scoring 3.04 against 2.97 for DeepSeek V4.1 Flash 1017. These figures come from Tencent and its partners and have not been independently verified.

The chart in the post compares Hy4 with GLM 5.3, DeepSeek V4.1 Flash, Kimi K3.1 Pro, Claude Opus 5.5 and GPT 6.1 Sol. Hy4 beats GLM 5.3 and DeepSeek V4.1 Flash on every chart where they appear, but Claude Opus 5.5 scores higher than Hy4 on all twelve benchmarks shown. Tencent itself says a notable gap remains with top proprietary models.

The weights are not out yet. Tencent says it plans to open source them, working with the vLLM and SGLang communities, and will publish a system card then. For now Hy4 is available on Tencent Cloud, OpenRouter and Tencent products such as CodeBuddy, WorkBuddy and Yuanbao, priced at $0.834 per million input tokens and $2.501 per million output tokens.

The key thing to watch is when Tencent actually releases the Hy4 weights and system card, since the open model claim depends on it. Independent benchmark groups will likely test Hy4 against DeepSeek, Kimi and GLM. Tencent also says a cheaper and faster Hy4 flash is coming soon. Builders will test whether the computer use features hold up outside Tencent's own demos.

Codex is OpenAI's coding agent and app. Pierre (@pierrecomputer), the pierre computer company, builds open source developer tools. OpenAI Newsroom (@OpenAINewsroom) is OpenAI's official news account.

OpenAI Newsroom (@OpenAINewsroom) posted on Nov 4 that OpenAI will acquire Pierre (@pierrecomputer). OpenAI said the deal is meant to accelerate its vision for making Codex a place where teams can build software together with AI, from the first idea to the version they ship.

According to OpenAI, Pierre's open source developer tools already power parts of Codex today, and the two will build on that foundation for what OpenAI called the agentic era. Pierre posted its own announcement video about a minute earlier, captioned as an announcement from the pierre computer company.

The post does not give a price, a closing date, or say what happens to Pierre's open source projects. Those details are not public in either post.

The framing matters. Most coding agents today work for one developer at a time. OpenAI's wording points at shared work: teams, the full path from idea to shipped version, and collaboration with AI agents along the way. Buying a company whose tools already sit inside Codex is a quicker way to build that than starting fresh.

The OpenAI post passed 340,000 views and Pierre's video passed 650,000 within hours.

Watch for OpenAI to say what happens to Pierre's open source tools and whether they stay open. Expect new Codex features aimed at teams, such as shared projects and review flows, in the coming months. Rivals such as Anthropic, Cursor and GitHub may answer with their own team features or acquisitions. Deal terms may come out if regulators or reporters dig in.

US export rules limit sales of the most advanced AI chips to China, but Chinese companies have rented computing power in data centers outside China. Oracle is one of the largest US cloud providers. Earlier in the Nov 3 deck, Watcher.Guru reported that Nvidia told partners to cut China business.

Zerohedge (@zerohedge) posted shortly after midnight ET on Nov 4 that Oracle canceled a $7B chip leasing deal with Tencent, citing the Financial Times, and added that things are starting to move.

According to summaries of the FT report posted by Wall St Engine (@wallstengine) and others, the canceled agreement was a five year deal that would have given Tencent access to as many as 100,000 advanced AI chips across Oracle data centers in Southeast Asia. The same summaries say Oracle dropped it after pressure from the Trump administration. CNBC TV18 also posted a headline, citing a report, saying Oracle scrapped the deal after US pressure.

These details come from posts relaying the FT, not from Oracle or Tencent statements, and the full FT report was not reviewed here. Neither company had commented in the posts found during this scan.

The report matters because renting compute abroad has been one of the ways Chinese technology companies could use advanced AI chips without buying them directly. A $7B, 100,000 chip arrangement would be one of the larger known deals of that kind.

The zerohedge post passed 440,000 views overnight, while US attention was largely on election results. It follows other China related chip news this week, including a Nov 3 report that Nvidia told partners to cut China business.

Watch for statements from Oracle, Tencent or the US Commerce Department confirming or explaining the cancellation. Investors will look for any impact on Oracle's cloud backlog. A key open question is whether US officials will press other cloud providers to limit compute rentals to Chinese companies in third countries, and how Beijing responds. Chinese cloud users and the chip suppliers behind these deals will also be watching closely.

Ricursive Intelligence (@RicursiveAI) trained Ripple from scratch. A dense model uses all its parameters for every token, unlike mixture of experts models. Agentic coding tests measure how well a model completes multi step programming tasks with tools.

Ricursive Intelligence (@RicursiveAI) posted on Nov 4 that it is releasing Ripple 65B Preview, a 65 billion parameter dense language model that it pretrained from scratch.

According to Ricursive, Ripple sets a new state of the art for US open weight models in agentic coding tasks. The company says it beats much larger models, naming Muse Spark Open, Beam, Inkling and Nemotron 3 Ultra, each listed in the post with hundreds of billions of parameters.

These results come from the company and are not independently confirmed in the post. The post does not list the benchmarks used or the scores, which sit behind the linked page. The word Preview suggests a fuller release is planned.

The claim is narrow on purpose: best among US open weight models, not best overall. That matters because the strongest open weight models have mostly come from Chinese labs, including Moonshot's Kimi K3.1 Pro, whose weights were released today. A smaller US model winning on coding agents would give American companies a strong option they can run themselves.

The post passed 340,000 views.

Watch for independent agentic coding results, the license, and a full release after the preview. Developers will compare Ripple with Kimi K3.1 Pro and other Chinese open models. If the claims hold, inference providers and US companies that avoid Chinese models may adopt it quickly, and Ricursive may raise money or announce larger models.

A full model card with training data, compute and evaluation details would also help outsiders judge the preview fairly.

Walter Bloomberg (@DeItaone) is a markets headline account that relays reports from banks and news outlets. The post relays a Goldman Sachs note after the US elections of Nov 3. Data center growth has become a local political issue because of power, water and tax break costs.

Walter Bloomberg (@DeItaone) posted on Nov 4 that Goldman Sachs warns Democratic victories at state and local levels could trigger temporary moratoria on data center construction.

According to the post, Goldman says 45% of planned US data center capacity could be affected, and Texas alone accounts for roughly 25% of planned capacity. The post says Goldman sees Texas, Georgia and Michigan as states that could introduce restrictions on tax breaks, electricity, water and environmental permits.

The post also says Goldman expects any moratoria to be short lived, with political pressure easing over time.

These figures are secondhand. The post relays the Goldman view but does not link the note itself, so the exact wording and assumptions could not be checked.

The warning lands on the same day as other election fallout in this deck, including a report that Michigan's newly elected Democratic senator wants to pause AI. Together they show data centers becoming a voting issue. For AI labs and cloud companies, any pause in permits or power hookups would slow the buildout they are counting on to train and serve larger models.

The post passed 460,000 views.

Watch for the full Goldman note and for statements from new state and local officials in Texas, Georgia and Michigan on permits, power and tax breaks. Data center developers and power companies may comment on project timing. If any local moratorium is actually proposed, AI companies may shift planned sites to other states or abroad, and the timeline Goldman expects for pressure to ease will be tested.

CellType (@celltypeinc) is a biology AI company. The benchmark asks models to predict how cells respond to drugs, gene knockdowns and immune signals, using results from real experiments. It was built on NVIDIA AI tools.

CellType (@celltypeinc) posted on Nov 4 asking whether LLMs can simulate biology, and introduced BioSimBench, built on NVIDIA AI.

According to CellType, the team tested frontier models on real lab experiments, asking them to predict how cells respond to drugs, gene knockdowns and immune signals. CellType says even the best models get only part of the response right, and that there is lots of room to improve.

The post opens a thread with more details, but the main post does not name which models were tested or give their scores. Those results should be read in the linked thread before quoting numbers.

The benchmark matters because predicting cell behavior is one of the main hopes for AI in drug discovery. If models could reliably simulate what a drug does to a cell, labs could skip many costly experiments. A benchmark grounded in measured results, rather than textbook questions, gives a harder and more honest test of that hope. CellType's own finding is that today's frontier models are not there yet.

The post passed 660,000 views.

Watch for model makers to report BioSimBench scores in new releases, and for other labs to test their biology models on it. CellType may publish the full method and data so results can be reproduced. Scores rising quickly on this kind of test would signal real progress for AI in drug discovery, while slow gains would temper claims about AI replacing lab work soon.

Conductor is an app for running several coding agents at once. Charlie Holtz (@charlieholtz) is one of its founders. MSL is Meta Superintelligence Labs, where Alexandr Wang and Nat Friedman are leaders.

Charlie Holtz (@charlieholtz) posted on Nov 4 that Conductor is joining Meta. He wrote that the team has big plans for the next interface, and that people who liked Conductor will love what is coming.

He said the team will work with Nat Friedman (@natfriedman) and Alexandr Wang (@alexandr_wang) at MSL, and wrote that he is convinced MSL has the highest talent density of all the labs. He thanked early backers @aaron_epstein, @ilyasu and @nabeel, and the users who tried Conductor and sent feedback.

The post does not say whether this is an acquisition or a hire of the team, what it cost, or what happens to the existing Conductor app. Those details are not stated.

Conductor became known among developers as a way to run many coding agents side by side. Its move to Meta follows a pattern of Meta bringing in founders and small teams to strengthen MSL. It lands on the same day OpenAI announced it is buying Pierre, another developer tools company, which shows the big labs competing for the teams that build interfaces around AI agents.

The post passed 380,000 views.

Watch for word on whether the current Conductor app keeps running and for how long. Meta may show the next interface Holtz mentioned in a future product launch. More small developer tool teams may join large labs, as OpenAI's Pierre deal today suggests. Details on deal terms may come from Meta or from reporting.

A hiring spree by MSL around agent interfaces would confirm this is part of a wider push by Meta.

Claude Code is Anthropic's coding agent. Subagents are helper agents that run tasks in the background. ClaudeDevs (@ClaudeDevs) is Anthropic's account for developer news. Haiku is Anthropic's smallest and cheapest Claude model line.

ClaudeDevs (@ClaudeDevs) posted on Nov 4 that background subagents in Claude Code now use less of your usage limit.

According to the post, a running subagent updates its status line every 30 seconds, and each update used to re read the subagent's whole conversation. Now Haiku 5.5 writes the update from short notes instead, so each update costs about 99% less, Anthropic says.

The change is small in code but large in effect for heavy users. A status line that refreshes twice a minute and re reads a long conversation each time adds up quickly, especially when several subagents run in parallel. Moving that job to a small model working from notes removes most of that hidden cost.

The post does not say how much total usage a typical session saves, since that depends on how many subagents run and for how long.

It comes the same day as another Claude Code change in this deck, the looser rule on building games in the style of a named game, and shows Anthropic tuning Claude Code for long, parallel work.

The post drew about 840,000 views and nearly 14,000 likes, the most likes of any new item today.

Watch for users to report how much longer their sessions last before hitting limits. Anthropic may move other background chores in Claude Code to Haiku in the same way. Competing coding agents such as Codex and Cursor may answer with their own savings on parallel agents, since usage limits are a frequent complaint from heavy users.

Anthropic may also publish more detail on how the short notes are kept accurate.

Claude Code is Anthropic's coding agent. Last week Addy Osmani (@addyosmani) said two sentences had been added to Claude Code's prompt on Opus 5.5 to steer it away from recreating someone else's specific game, because using Claude to infringe IP is against Anthropic's usage policy.

Addy Osmani (@addyosmani) posted an update on building games with Claude Code shortly after midnight ET on Nov 4.

He says that last week some users ran into Claude Code pushing back on game projects it should have just built. The team had added a short rule to the Opus 5.5 prompt to steer Claude away from recreating specific existing games, but it was also kicking in when people referenced a game only to describe the mechanics they wanted.

According to the post, the rule was updated in Claude Code version 2.1.317. Users can now name a game to describe what they want, and Claude should build an original game in that genre. If someone asks for a copy of the original game, Claude will offer to build something similar instead.

Osmani says the goal is for Claude to be as helpful as possible when people build something original, while also protecting their work if someone else tries to copy it. He asks users to report cases where Claude turns down work that is theirs.

In his earlier post on Oct 30, Osmani said the original change was two sentences in the prompt, added because using Claude to infringe IP is against Anthropic's usage policy, and that original games, genre games and games built from a user's own designs were meant to keep working. The update post passed 100,000 views.

Watch for whether game developers report fewer refusals after updating to version 2.1.317, and whether new edge cases appear around famous franchises. The episode may push other coding agent makers to explain how they handle requests to recreate existing products. A wider open question is how AI tools should judge when inspiration becomes copying. Anthropic may adjust the rule again based on user reports.

Gemini 3.8 Flash TTS is Google's text to speech model. Google AI Studio is Google's developer tool for trying Gemini models. TTS means text to speech.

Google (@Google) posted on Nov 4 that Voice Remixing for Gemini 3.8 Flash TTS is now live in Google AI Studio and the Gemini API.

According to Google, developers can start with a preset from a library of more than 2,000 voices or a custom designed voice, then use prompts to fine tune timbre, pitch, pace and accent. Google's examples include adding a subtle Southern US accent and making a voice sound like a dragon that breathes fire.

The post does not give pricing or limits in its text, and it does not describe safeguards against copying real people's voices, which is a common concern with voice tools.

The feature changes how voices are made. Instead of recording or cloning, a developer describes changes in words, the same way they prompt a text model. That lowers the work needed to give an app, game or assistant its own voice.

It also fits a wider push in voice AI this week, from voice agents to phone calling assistants. The post passed 130,000 views.

Watch for developers to share remixed voices in apps and games, and for Google to publish pricing and rules on voice use. Competitors such as ElevenLabs and OpenAI may add similar prompt based voice editing. Policy discussions on synthetic voices may also cite tools like this one.

Google may also expand the voice library, add more languages, or bring remixing to its own consumer apps such as the Gemini app and its assistant products.

Christel Buchanan (@ladyxtel) introduced Living Brain. LoCoMo is a public benchmark that tests how well systems remember facts across long conversations. Memory layers store what an AI learns about a user or project outside the model itself.

Christel Buchanan (@ladyxtel) posted on Nov 4 that her team decoupled memory from the model. She introduced Living Brain, described as one persistent brain for ChatGPT, Claude, Cursor, Codex and your own agents, where you write in one tool and retrieve in any.

She claims Living Brain scores 92.2% on the LoCoMo memory benchmark, has a median retrieval time of 464 milliseconds, and uses up to 22.6 times fewer tokens per question. She closed with the line that LLMs have parameters while Living Brain has plasticity.

All of these numbers come from the maker and are not independently confirmed in the post. The post also does not say how the benchmark was run, what the token savings are compared against, or how the product is priced.

The idea answers a real pain. Each AI assistant keeps its own memory, so users switch tools and lose context. A shared layer that any tool can read would make it easier to move between assistants.

The post drew about 1.39 million views and almost 2,000 reposts, the widest reach of any new item in today's scan, though likes were lower at about 2,400.

Watch for outside developers to test Living Brain on LoCoMo and real work, and for details on privacy, storage and pricing. Integrations with each named tool will show how deep the connection goes. Built in memory features from OpenAI, Anthropic and others compete directly with this idea, so their next memory updates will matter for whether a separate layer catches on.

The New York Times, citing the AP, reported that a progressive Democrat won Michigan's Senate race on election night and will become the first Muslim senator when seated in January. Earlier posts in this deck tied the election results to fears for the AI data center buildout.

Chubby (@kimmonismus) posted early on Nov 4 ET that the Michigan Senate result is incredibly bad news for AI.

According to the post, the winner wants to pause AI, while at the same time almost every AI data center is being blocked by local communities. Chubby adds that people fail to realize how transformative AI is for humanity. The post quotes a video clip of just under two minutes posted by another account on Nov 2.

The New York Times posted overnight, citing the Associated Press, that the progressive Democrat won Michigan's Senate race and will become the first Muslim senator when he is seated in January. That post passed 128,000 views.

The description of the senator elect's AI position comes from Chubby's post and the quoted clip, not from a policy document or statement reviewed here. The exact scope of any pause he supports, such as on frontier model training or on data centers, is not spelled out in the post.

The post passed 100,000 views and drew more than 200 replies, many debating whether AI policy played any role in the race.

Watch for the senator elect's own statements on AI once he takes office in January, including whether he backs specific legislation. AI companies and their lobbying groups will likely court new members of Congress on the issue. The combination of skeptical lawmakers and local data center fights could shape where and how fast new compute gets built. Polls on AI as an election issue may also get more attention.

Pranav Sharan (@pranavctl) built Aether (@aetherruntime). Cloud coding agents run tasks on remote machines so work continues when a laptop is closed. Claude Code and Codex are the coding agents from Anthropic and OpenAI.

Pranav Sharan (@pranavctl) posted on Nov 4 introducing Aether, which he called the fastest cloud coding agents in the world that work with your existing subscription.

According to the post, you give Aether a task and about 20 seconds later the build is running on its own cloud machine. You can pick it up from your phone, watch every line, and step in as if it were local, and work continues after you close your laptop.

Sharan says Aether runs Claude Code or Codex on the Claude or ChatGPT plan you already pay for, so there is no second AI bill, and you pay only for the machine. When code review is enabled, a reviewer agent runs the code with the changes, which he says catches bugs that static analysis cannot find.

He says he previously built infrastructure at Apple, Uber and HeyGen, and that Aether is free to start.

The speed claim is the maker's own and is not independently confirmed. The post also does not detail machine pricing.

The post passed 150,000 views.

Watch for developers to compare Aether with cloud agent options from OpenAI, Anthropic, Cursor and others on speed and price. Anthropic and OpenAI rules on using personal subscriptions in outside tools will matter for how long the model works. Machine pricing and team plans are likely next steps if adoption grows.

Independent tests of the 20 second start time and of the reviewer agent on real projects would show whether the claims hold up for larger codebases.

Grok Bot is an assistant agent that can browse and complete tasks for users. Matt Palmer (@mattyp) posted a short tutorial video. Pages are documents and layouts Grok Bot can build and share.

Matt Palmer (@mattyp) posted a video walkthrough titled Pages in Grok Bot on Nov 4. The post lists chapters in the video: what Pages are at 0:15, an example page at 0:40, sharing a Page at 1:34, tips on using your own data at 1:55, using X posts at 2:46, and a recap at 3:44.

The post itself is a chapter list and does not describe new features in its text, so what Pages can do is shown in the video rather than stated in writing. Based on the chapter titles, the video covers building a page, sharing it with others, and filling it with your own data and with posts from X.

The walkthrough comes a day after a run of Grok Bot launches covered in the Nov 3 deck, including shopping and phone calls. It is a tutorial rather than a new launch, which is why it sits in the middle of today's deck.

Features like Pages show a wider shift in AI assistants: instead of only answering in a chat window, the assistant produces a lasting result that can be opened and shared like a web page.

The video passed 420,000 views and about 1,500 bookmarks.

Watch for more tutorials and examples from Grok Bot users sharing their own Pages, and for any updates that add new layouts or data sources. Other assistants with similar features, such as artifacts and canvases, may push their own sharing tools. Usage numbers or a public gallery would show how many people actually build and share Pages.

More videos in the same series may cover other Grok Bot features.

Theo Browne (@theo) runs the T3 developer tools and a popular tech video channel, and he worked at Anthropic. On Nov 3, Cat Wu, an early Claude Code team member, said they had also left Anthropic, as covered in the Nov 3 deck.

Theo Browne (@theo) posted on the morning of Nov 4 ET that it was his last day at Anthropic.

He writes that he is thankful for the time he spent there and the connections he made. According to the post, he will take some time off while he prepares for his next thing, and he says he will share more soon. He does not say why he left or what the next thing is.

The post came hours after he wrote that he was about to vaguepost harder than ever, and about half an hour before he quote posted a viral claim that Anthropic's IPO was cancelled. In that quote post he said he is quick to call out misinformation from the account that made the claim, but that he had heard similar news from a few sources. That claim remains unverified, and it is covered in the Nov 3 deck.

The departure follows Cat Wu's post on Nov 3 about leaving Anthropic after helping build Claude Code. Anthropic had not commented on either move in the posts found during this scan.

Theo's post passed 150,000 views within a few hours.

Watch for Theo's announcement of what he does next, which he says is coming soon. Followers will look for whether more Anthropic staff announce exits in the coming days. Any comment from Anthropic on team changes, or a clear update on its IPO plans, would help separate real news from rumor. His large audience means his next move will get wide attention among developers.

Lauren (@poteto) works on Grok Bot and is giving a talk at AI Engineer (@aiDotEngineer), a conference for people building with AI. The post is a teaser for that talk.

Lauren (@poteto) posted slides on Nov 4 from an upcoming talk at AI Engineer. She wrote that the talk covers how her team ships crazy amounts of code, which she put at 30x, without reading it, while performance is as good as or better than at launch.

She wrote that there is still a lot to improve and the ceiling is far from reached, but that given the shipping speed and the team's high bar for polish, its agent friendly architecture bets are paying off at taming the slop. She added that about 230 people landed at least one pull request to Grok Bot in October.

The first slide shows merged pull requests a month that touch the Grok Bot team's code, rising from 809 in June to 1,784 in July, 4,203 in August, 5,648 in September and 10,926 in October, with 23,370 in total. The second compares May with now: cold launch fell 31% to 1,831 ms, relaunch fell 35% to 1,200 ms, and typing delay in a long chat fell 61% at the median.

These are the team's own figures. The post is a preview, so the full method will come in the talk.

Watch for the full talk from AI Engineer, which should explain the architecture choices and how the team checks code it does not read by hand. Other teams may share similar numbers on agent written code. Critics will likely ask how bugs and security issues are caught, so any data on defect rates would matter.

The slides may also spark debate on how much human code review teams still need.

On election night, market accounts tracked strong Democratic results, and zerohedge said the results could slow the data center buildout, as covered in the Nov 3 deck. Musk has promoted the idea of data centers in space, and SemiAnalysis said on Nov 3 that SpaceX will become the world's biggest AI cloud.

Elon Musk (@elonmusk) posted early on Nov 4 ET that data centers in space are looking pretty appealing right now.

The post came during a night of US midterm election results that market accounts described as a strong showing for Democrats. Earlier that night, zerohedge said Nasdaq futures fell as the Texas race swung, arguing that a Democratic win would effectively ice the data center buildout, and Polymarket posted that AI stock futures plunged across the board as a Democratic blue wave took shape. Those claims are interpretations by market accounts, not stated policies of the winners.

Musk's post does not mention the election directly, but it was widely read as a response to it. Polymarket relayed it as Musk declaring that data centers in space look appealing following the results.

The remark connects to a theme Musk and his companies have pushed for months: putting AI compute in orbit, where power and land are not limited by local permits or voters. On Nov 3, SemiAnalysis said SpaceX will become the world's biggest AI cloud. Space based data centers still face open engineering questions, including cooling, launch costs and maintenance.

The post passed 3 million views within a few hours, making it one of the most viewed AI related posts of the night. A separate Musk post joking about space lasers drew even more views but had no AI angle.

Watch for whether SpaceX or xAI share concrete plans or timelines for orbital compute. Official results and the winners' actual positions on data centers, power and permits will show whether the buildout really slows on Earth. Investors will watch AI and chip stocks as markets digest the election. The cost and engineering case for space data centers remains unproven at large scale.

Dario Amodei is the CEO and a cofounder of Anthropic. He has warned in past interviews that AI could eliminate many entry level white collar jobs within a few years. Indian Tech and Infra (@IndianTechGuide) is a news aggregator account.

Indian Tech and Infra (@IndianTechGuide) posted on the morning of Nov 4 ET a quote it attributes to Anthropic founder Dario Amodei.

According to the post, the quote says that 50% of all entry level lawyers, consultants and finance professionals will be completely wiped out within the next five years. The post includes a photo but no link, video, date or source for when or where the remark was made.

This is a relayed quote. The post does not say whether it comes from a new interview or repeats something said earlier, and the wording was not checked against a transcript here. Amodei has made similar warnings in past interviews about AI eliminating many entry level white collar jobs, so this may be a repeat of an older remark rather than a new statement.

The post passed 139,000 views and drew close to 200 replies, many arguing over whether the timeline is realistic. It spread on a morning when AI and jobs were already in the news because of the US election results.

Anthropic had not commented on the post in the results found during this scan.

Watch for whether a new interview or source for the quote surfaces, which would show if this is fresh commentary or a recycled remark. Data on hiring of junior lawyers, consultants and finance analysts will keep being cited on both sides of the debate. Lawmakers newly elected this week may raise AI and jobs as an issue. Other AI lab leaders may respond with their own forecasts.