📰 Welcome back!

Welcome back to Big Machines after a brief hiatus.

It’s been a busy couple weeks at BM Towers, and we couldn’t spool up the AI intern to get last week’s edition out on time.

Luckily for you, we’ve got a real-life human rounding up all the major stories from the past two weeks (Oct 13-27), so you don’t miss any of the real headlines shaping the world of tech and artificial intelligence.

And I guess that’s kind of a running theme – AI’s evolution wreaking havoc in the corporate world. From OpenAI’s long-awaited ChatGPT-powered browser launch, to Meta dispatching with 600 roles across its AI teams, or Deloitte’s epic AI-powered fuck-up, the tech has played an important role in the way corporations integrate and use tech to shape both workforces and workflows.

The only way is up, but let’s dive on in. 🤿 

🚀 What we’re covering today…

• 🗺 OpenAI’s ChatGPT Atlas arrives

• 🪓 Meta axes 600 AI jobs

• 🦾 Amazon one ups them, cutting 30,000 jobs

• 💩 Deloitte publishes report full of AI slop

• 🧠 Anthropic introduces “hybrid reasoning”

• 📰 BBC study says AI gets news wrong

• 🍎 Apple loses more top talent to Meta

• 🖥️ Databricks develops self-improving AI trick

• 💸 Anthropic bets on ARR trebling

• 🤖 Boston Dynamics’ AI robot gets opposable thumbs

🔴 Quick Note: We like to cover loads of AI news in our newsletter, so for a better reading experience, we suggest opening this in your browser for the full experience! 

Head to the ‘READ ONLINE’ tab at the top of this email.

👁️ 👁️ What you might have missed

• It’s been some time coming, but has Google’s great usurper finally arrived? We have no clue tbh, but know OpenAI has launched ChatGPT Atlas, the AI-powered web browser designed to rival Chrome. Built around ChatGPT, Atlas replaces the traditional address bar and introduces an “agent mode” that autonomously searches and assists users, although available only to paying ChatGPT subscribers atm. The browser debuts on macOS, reflecting OpenAI’s efforts to monetise its AI technology and expand partnerships with platforms like Etsy, Shopify, Expedia, and Booking.com.
  
  Atlas’s release follows OpenAI’s proclamation ChatGPT has reached 800 million weekly users. Analysts remain cautious, noting that mainstream users may prefer familiar browsers like Chrome or Edge, which already integrate similar AI features. They also argue that widespread adoption will depend on user trust, interface ease, and how seamlessly AI tools can be embedded into existing browsing habits. But without clear advantages over established browsers, how will Atlas fare in capturing significant market share?

• Well, well, well – Meta is cutting up to 600 jobs in its AI division. The move is targeting roles in product, infrastructure, and research groups under – but not including – Zuckerberg’s much vaunted “superintelligence” division. The layoffs aim to streamline decision-making, reduce bureaucracy, and give remaining team members broader responsibilities, according to Meta insiders. A newly formed unit, TBD Labs, which is focused on next-generation AI and LLMs, is to be spared from cuts and looks likely to continue hiring. Meta is encouraging affected staff to apply internally for other roles, and many are expected to transition rather than leave outright.
  
  The move to axe corporate roles in the group’s hottest division could hint at more than Zuckerberg’s bet on hiring star roles over legacy muscle – rather it could mark the moment AI enters its first real correction with a shift from exuberant experimentation to measured maturity. Like the dot-com and cloud shakeouts before it, this retrenchment suggests the technology is settling into sustainable scale. TBD Labs could also indicate where the real bets now lie: fewer sprawling teams and more concentrated ambition. Whether this is AI’s growing-up moment or the first sign of overreach, it shows the gold rush is giving way to pragmatism.
  

• ‘Big Four’ professional services consultancy Deloitte is in deep doo-doo, after it emerged a report it delivered to the Australian government was riddled with AI-induced errors, including misleading data points, insights and references – and when we mean ‘misleading’, we mean totally fucking made up. Hallucinations included citations from non-existent academic papers and a fabricated quote from a federal court judgment. Tucked-tail, Deloitte publicly apologized for the snafu and admitted oversight procedures were not followed. The firm issued a revised version and agreed to partially refund ~AUD 440,000 (US$290,000) to Oz gov coffers. It gets better, however; citations used in this v2 report, again comprised using AI, turned out to be “irrelevant” and erroneous.

  
  Critics have framed this as a “wake-up call” for the corporate world: don’t be lazy shits and try pull the wool – AI is an aide not a replacement, nor is it an infallible oracle, principal consultant or junior colleague. Deloitte-gate has sparked broader scrutiny of how firms deploy generative AI in high-stakes work such as finance and government consulting, and highlights the need for any corporate to have a mix of robust internal controls and human oversight when integrating systems into workflows. For me personally, it begs an even greater existential question: “What are managing consultants actually good for?”
  
  

  

  Source: @JeremyPoxon/X.com

  
  

• An indicator Zuck is doubling down on hiring marquee talent over numbers to scale, is the recent poaching of a top AI exec Ke Yang from Apple. Yang headed uo Apple’s ‘Answers, Knowledge and Information’ (AKI) project, but has swapped Cupertino for Menlo Park to join Meta’s AI ‘superintelligence’ division. It marks an extraordinarily late move in the AI Transfer Window^TM (wait, it’s not closed??), and signals another grim chapter in the Apple AI division’s history, which perviously lost a dozen or so senior bods from foundation models’ teams earlier in the year. Yang was charged with powering a ChatGPT-style web search functionality within Siri after the previous AKI head did a runner, and has now followed suit. Will Apple ever be able to fulfil its AI ambitions?
  

• Amazon will begin cutting up to 30,000 corporate jobs – its largest layoff ever. The reduction represents nearly 10% of Amazon’s 350,000 corporate employees, though it’s just a small fraction of the company’s 1.55 million total workforce globally. Multiple divisions, including HR, operations, devices, and AWS, are affected. Managers have been to communicate the news, with email notifications for the dammed expected immediately. The cuts follow pandemic-era overhiring and previous layoffs of 27,000 jobs since 2022, as CEO Andy Jassy aims to streamline Amazon’s corporate structure and leverage more automation.
  

• A comprehensive study by the European Broadcasting Union (EBU) and BBC reveals that leading AI assistants, including ChatGPT, Copilot, Gemini, and Perplexity, often misrepresent news content. Analyzing 3,000 responses in 14 languages, researchers found that 45% contained significant errors, with 81% showing some form of issue. Notably, 72% of Gemini's responses had serious sourcing problems, compared to under 25% for others. Examples include outdated information, such as ChatGPT incorrectly stating Pope Francis was still alive months after his death. The study underscores the need for AI companies to improve accuracy and accountability in news reporting.

  
  

  

  “I don’t have a clue what’s real anymore”. Source: BBC/EBU

  
  

• Just like me after Indian food, Anthropic believes what comes next will be explosive. Sources say the Claude developer plans to nearly triple its annualized revenue run rate by 2026. It expects to finish 2025 at about US$9 billion, and has set a base target of $20 billion, making an ambitious upside of $26 billion next year. Right now, Anthropic’s run rate is approaching $7 billion, up from $5 billion in August. Around 80% of its revenue comes from over 300,000 enterprise customers using its AI tools via APIs, while key products like Claude Code have already hit nearly $1 billion in annualized revenue.
  

• OpenAI has partnered with Samsung Electronics and SK hynix as part of its Stargate initiative to enhance global AI infrastructure. Samsung and SK will accelerate production of DRAM chips, targeting 900,000 wafer starts per month to support OpenAI’s AI models. Additionally, OpenAI plans to develop next-generation AI data centers in South Korea, including collaborations with SK Telecom and Samsung affiliates. The partnership aims to bolster Korea's position as a global AI leader.

🧩 Other Bits

• If you didn’t think AI-powered automatons could be made to sound any creepier, Boston Dynamics has given its Atlas humanoid robot ‘advanced grippers’. Like a drunk uncle at a wedding, these robots are able to get even handsier thanks to their opposable thumbs and human-like dexterity. Powered by adaptive AI and designed for industrial tasks, this innovation marks a major step toward humanoid robots integrating seamlessly into real-world environments like warehouses and factories.
  
  

  

  Source: Boston Dynamics

  
  

• A groundbreaking AI trial involving over 30,000 workers from the UK’s National Health Service (NHS) has demonstrated significant time and cost savings. The pilot of Microsoft 365 Copilot across 90 NHS organisations revealed that AI-powered administrative support could save staff an average of 43 minutes per day, equating to five weeks annually. A full roll-out could potentially save up to 400,000 hours of staff time per month, translating to millions of hours annually – a huge bone of contention in practical and political spheres alike.
  

• Anthropic has made its Claude memory feature available to all paying subscribers (Pro and Max). The model can now remember context from past conversations, while offering an incognito mode for temporary chats and a dashboard to manage stored data. This update improves personalization in AI interactions while emphasizing privacy and responsible AI use, giving users more control over what the chatbot retains and how their data is handled.
  

• Google and Yale University have created an advanced AI model called Cell2Sentence-Scale 27B, built on Google’s Gemma architecture, to analyze single-cell RNA data and interpret biological information using language-model techniques. The system not only improves understanding of cellular behaviour but also generated a new cancer therapy hypothesis validated in living cells, marking a major milestone for AI’s role in scientific discovery.
  
  

📈 Trending tools, models & apps this week

📋 LLM Leaderboard

📲 Trending tools & apps

• ChatGPT Atlas – OpenAI’s web browser killer has finally landed (TBC). ChatGPT promises to come “anywhere across the web – helping you in the window where you are, understanding what you’re trying to do, and completing tasks for you… all or leaving the page.”
  

• Grokipediai – Say whaaaaat? An xAI rival to Wikipedia? This AI-generated encyclopedia was pitched by Musk as a “more truthful” alternative to Wiki. It’s launched with nearly 900k articles already, and aims to reshape online knowledge using AI. We wonder what Sam Altman’s page looks like?
  

• Dynal.AI – One to put us out of a job. Turn ideas, links, PDFs, and images into ready-to-post LinkedIn posts in minutes. Dynal learns your tone and uses proven patterns to draft editable copy and visuals, making it easier to post consistently.
  
  

💸 Financials

• AI talent platform Mercor raised $350 million at a $10 billion valuation in a Series C led by Felicis Ventures, and backed by Benchmark, General Catalyst, and Robinhood Ventures. Mercor connects AI research labs with domain experts for human-led model training.
  

• Periodic Labs, a new venture by former OpenAI and DeepMind researchers, raised a huge $300 million in seed funding, backed by a veritable who’s who in the tech industry including Andreessen Horowitz, DST, Nvidia, Accel, Elad Gil, Jeff Dean, Eric Schmidt, and Jeff Bezos.
  

• Document intelligence platform Reducto secured $75 million in Series B funding led by Andreessen Horowitz, with participation from Benchmark, First Round Capital, BoxGroup, and Y Combinator.
  

• Multimodal AI infrastructure startup Fal.ai raised $250 million at a $4 billion valuation in a round led by Kleiner Perkins and Sequoia Capital, highlighting continued investor appetite for AI infrastructure plays.
  

• AI-powered smart glasses startup Sesame raised $250 million in Series B funding, led by former Oculus executives. The company is building glasses with always-on personal assistant capability.
  

• Semiconductor design startup ChipAgents secured $21 million in Series A funding led by Bessemer Venture Partners, with strategic investments from Micron Ventures, MediaTek Ventures, and Ericsson.
  

• AI infrastructure startup 1001 AI raised $9 million in seed funding led by CIV, General Catalyst, and Lux Capital. The company is building AI infrastructure for critical industries in the Middle East and North Africa.
  

• AI-powered talent recruitment platform Findem raised $51 million, comprising a $36 million Series C led by Silver Lake Waterman plus a $15 million growth loan from JPMorgan.
  

• Enterprise AI agent platform Serval secured $47 million in funding to expand its autonomous AI capabilities for business workflows.
  

🕵️‍♂️ FREE ENTRY TO OUR INVITE-ONLY AI CHAT ON TELEGRAM…

If you share this newsletter with a friend and they actively sign up for the Big Machines newsletter, we will send you access to our invite-only Big Machines Telegram group, which is full of builders, investors, founders, and creators.

Access is now only granted to those who refer our newsletter to active subscribers, which means if you sign up on your work email, we will know you sneaky bastards.

This would kill our open rate, so please don't do that, we beg.

👋 Here’s a nice story to get you through the week… see you next time.

Took me a day to write this and ChatGPT Atlas is already living up to its hype.

Wonder if BBC would call this fake news? Till next week. 👋

Sam, Grant, Matt, Mike and the Big Machines team.